lenguaje by geraldine villanueva torres

Visual Prolog Version 5.0

Language Tutorial

Copyright The documentation for this software is copyrighted, and all rights are reserved. It may not be reproduced, transmitted, stored in a retrieval system, or translated, either by electronic, mechanical or any other means, without the prior written consent of Prolog Development Center A/S. The software products described in these manuals are also copyrighted, and are licensed to the End User only for use in accordance with the End User License Agreement, which is printed on the diskette packaging. The prospective user should read this agreement carefully prior to use of the software. Visual Prolog is a registered trademark of Prolog Development Center A/S. Portions of the Software and Documentation under license from: •

Other brand and product names are trademarks or registered trademarks of their respective holders.

Table of Contents Using This Manual...........................................................................................ix PART 1 xii Using the Environment.......................................................................................xii CHAPTER 17....................................................................................................xiii Using the Visual Prolog Environment...............................................................xiii What Needs to be Installed for This Book ?...................................................xiii Starting the Visual Prolog Environment.........................................................xiv Opening an Editor Window............................................................................xiv Running and Testing a Program.....................................................................xiv Loading Examples From Disk.........................................................................xv Handling Errors..............................................................................................xvi

PART 2 xvii Tutorial Chapters 19 -- 20: Learning Prolog2122Visual Prolog........................xvii CHAPTER 23..................................................................................................xviii Prolog Fundamentals.......................................................................................xviii PROgramming in LOGic.............................................................................xviii Sentences: Facts and Rules.........................................................................xix Queries.......................................................................................................xxi Variables: General Sentences...................................................................xxiv Overview...................................................................................................xxv From Natural Language to Prolog Programs................................................xxvi Clauses (Facts and Rules)........................................................................xxvi Predicates (Relations)................................................................................xxx Variables (General Clauses)......................................................................xxx Goals (Queries)......................................................................................xxxiv Comments.............................................................................................xxxvii What Is a Match?.....................................................................................xxxviii Summary..........................................................................................................xl CHAPTER 30....................................................................................................xlii Visual Prolog Programs.....................................................................................xlii Visual Prolog's Basic Program Sections........................................................xlii The Clauses Section..................................................................................xliii The Predicates Section..............................................................................xliii The Domains Section................................................................................xlvi The Goal Section............................................................................................l A Closer Look at Declarations and Rules..........................................................l Other Program Sections..................................................................................lix The Facts Section........................................................................................lix The Constants Section.................................................................................lix The Global Sections....................................................................................lxi The Compiler Directives.............................................................................lxi Summary........................................................................................................lxii CHAPTER 57...................................................................................................lxiv Unification and Backtracking...........................................................................lxiv Matching Things Up: Unification..................................................................lxv Backtracking...............................................................................................lxviii Visual Prolog's Relentless Search for Solutions.........................................lxx Controlling the Search for Solutions.......................................................lxxxi Using the fail Predicate..........................................................................lxxxii Preventing Backtracking: The Cut........................................................lxxxiv Prolog from a Procedural Perspective .........................................................xciii How Rules and Facts Are Like Procedures..............................................xciii iii

Summary......................................................................................................xcix CHAPTER 77...................................................................................................101 Simple and Compound Objects.........................................................................101 Simple Data Objects......................................................................................101 Variables as Data Objects..........................................................................101 Constants as Data Objects.........................................................................101 Compound Data Objects and Functors..........................................................104 Unification of Compound Objects.............................................................105 Treating Several Items as One...................................................................106 Declaring Domains of Compound Objects................................................111 Compound Mixed-Domain Declarations...................................................116 Summary.......................................................................................................118 CHAPTER 88...................................................................................................120 Repetition and Recursion..................................................................................120 Repetitive Processes......................................................................................120 Backtracking Revisited..............................................................................120 Implementing Backtracking with Loops....................................................124 Recursive Procedures................................................................................126 Tail Recursion Optimisation......................................................................128 Using Arguments as Loop Variables.........................................................134 Recursive Data Structures.............................................................................138 Trees as a Data Type.................................................................................139 Binary Search Trees..................................................................................145 Summary.......................................................................................................152 CHAPTER 122.................................................................................................153 Lists and Recursion...........................................................................................153 What Is a List?..............................................................................................153 Declaring Lists..........................................................................................154 List Processing..............................................................................................155 Using Lists....................................................................................................156 Writing Lists.............................................................................................157 Counting List Elements.............................................................................158 Tail Recursion Revisited...........................................................................160 List Membership.......................................................................................164 Appending One List to Another: Declarative and Procedural Programming ..................................................................................................................165 Finding All the Solutions at Once.................................................................168 Compound Lists............................................................................................170 Summary.......................................................................................................175

CHAPTER 142.................................................................................................176 Visual Prologs fact sections..............................................................................176 Declaring the facts-sections...........................................................................177 Using the facts sections.............................................................................178 Accessing the facts sections......................................................................178 Updating the facts section.........................................................................179 Facts determiner-keywords.......................................................................184 Saving a database of facts at runtime........................................................186 Examples.......................................................................................................186 Summary.......................................................................................................190 CHAPTER 148.................................................................................................191 Arithmetic and Comparison..............................................................................191 Arithmetic Expressions.................................................................................191 Operations.................................................................................................192 Order of Evaluation...................................................................................192 Functions and Predicates...............................................................................193 Generating Random Numbers...................................................................194 Integer and Real Arithmetic......................................................................196 Comparisons..................................................................................................200 Equality and the equal (=) Predicate..........................................................201 Comparing Characters, Strings, and Symbols...........................................204 CHAPTER 166.................................................................................................206 Classes and objects............................................................................................206 Encapsulation............................................................................................206 Objects and classes....................................................................................206 Inheritance.....................................................................................................207 Identity..........................................................................................................207 Visual Prolog Classes....................................................................................207 Class declarations..........................................................................................208 Class implementation....................................................................................208 Class instances..............................................................................................209 Destroying Objects........................................................................................210 Class Domains...............................................................................................210 Sub-classing and inheritance.........................................................................211 Virtual Predicates..........................................................................................212 Static facts and predicates.............................................................................214 Class Scopes..................................................................................................214 Constructors and Destructors........................................................................216 Reference to the Object Itself (This).............................................................217 Abstract Classes............................................................................................218 Protected facts and predicates.......................................................................219

Derived class access control..........................................................................219 Formal definition for classes.............................................................................221 CHAPTER 168.................................................................................................222 Advanced Topics...............................................................................................222 The Flow Analysis........................................................................................222 Specifying Flowpatterns for Predicates.....................................................224 Controlling the Flow Analysis...................................................................225 Reference Variables..................................................................................226 Declaring Domains as Reference..............................................................226 Reference Domains and the Trail Array....................................................227 Using Reference Domains.........................................................................228 Flow Patterns Revisited.............................................................................229 Using Binary Trees with Reference Domains...........................................230 Sorting with Reference Domains...............................................................231 Functions and Return Values.........................................................................233 Determinism Monitoring in Visual Prolog....................................................235 Visual Prologs determinism checking system...........................................236 Predicates as Arguments...............................................................................238 Predicate Domains.....................................................................................238 Examples...................................................................................................240 The Binary Domain.......................................................................................244 Implementation of binary terms................................................................244 Text syntax of Binary Terms.....................................................................245 Creating Binary Terms..............................................................................245 Accessing Binary Terms...........................................................................246 Unifying Binary Terms.............................................................................247 Example....................................................................................................247 Converting Terms to Binary Terms...........................................................248 Errors and Exception Handling.....................................................................250 Exception Handling and Error Trapping...................................................251 Error reporting...........................................................................................253 Handling Errors from the Term Reader.....................................................255 Break Control (Textmode Only)...................................................................256 Manual Break and Signal Checking in UNIX...........................................258 Critical Error Handling under DOS Textmode..........................................263 Dynamic Cutting...........................................................................................265 Free Type Conversions..................................................................................266 Programming Style........................................................................................267 Rules for Efficient Programming..............................................................267 Using the fail Predicate.............................................................................270 Determinism vs. Non-determinism: Setting the Cut..................................271

PART 3 272 Tutorial Chapters 201 -- 202: Using Visual Prolog.203204Visual Prolog.........272 CHAPTER 205.................................................................................................272 Writing, Reading, and Files...............................................................................272 Writing and Reading.....................................................................................272 Writing......................................................................................................272 Reading.....................................................................................................280 Binary Block Transfer...............................................................................284 Visual Prolog's File System...........................................................................285 Opening and Closing Files........................................................................286 Redirecting Standard I/O...........................................................................289 Working with Files....................................................................................290 File Attributes...........................................................................................295 File and Path Names......................................................................................299 Directory Searching......................................................................................300 Manipulating File Attributes.........................................................................305 Handling terms in text files...........................................................................306 Manipulating Facts Like Terms.................................................................307 Summary.......................................................................................................309 CHAPTER 231.................................................................................................311 String-Handling in Visual Prolog......................................................................311 String Processing...........................................................................................311 Basic String-Handling Predicates..............................................................311 Type Conversion...........................................................................................318 Summary.......................................................................................................322 CHAPTER 240.................................................................................................324 The External Database System..........................................................................324 External Databases in Visual Prolog.............................................................324 An Overview: What's in an External Database?........................................325 Chains.......................................................................................................327 External Database Domains......................................................................329 Manipulating Whole External Databases..................................................330 Manipulating Chains.................................................................................336 Manipulating Terms..................................................................................338 A Complete Program Example..................................................................339 B+ Trees........................................................................................................343 Pages, Order, and Keylength.....................................................................343 Duplicate Keys..........................................................................................344 Multiple Scans...........................................................................................344 The B+ Tree Standard Predicates..............................................................344 Example: Accessing a Database via B+ Trees...........................................347 vii

External Database Programming...................................................................349 Scanning through a Database....................................................................350 Displaying the Contents of a Database......................................................351 Implementing a Database That Won't Break Down...................................353 Updating the Database...............................................................................354 Using Internal B+ Tree Pointers................................................................358 Changing the Structure of a Database.......................................................360 Filesharing and the External Database.......................................................362 Filesharing Domains.................................................................................362 Opening the Database in Sharemode.........................................................363 Programming with Filesharing..................................................................366 Implementing highlevel locking................................................................368 A Complete Filesharing Example..............................................................369 Implementation Aspects of Visual Prolog Filesharing .............................374 Miscellaneous................................................................................................375 Summary.......................................................................................................375 CHAPTER 253.................................................................................................376 System-Level Programming..............................................................................376 Access to the operating system......................................................................376 Timing Services........................................................................................380 Bit-Level Operations.....................................................................................384 Access to the Hardware: Low-Level Support................................................387 Summary.......................................................................................................390 chapter 260........................................................................................................391 Example Prolog Programs.................................................................................391 Building a Small Expert System....................................................................392 Prototyping: A Simple Routing Problem.......................................................396 Adventures in a Dangerous Cave..................................................................398 Hardware Simulation.....................................................................................401 Towers of Hanoi............................................................................................402 Dividing Words into Syllables......................................................................404 The N Queens Problem.................................................................................408 PART 4 411 Programmer’s Guide.........................................................................................411 CHAPTER 286.................................................................................................412 Elements of the Language.................................................................................412 Names...........................................................................................................412 Program Sections..........................................................................................413 The Domains Section................................................................................414 The Predicates Section..............................................................................419

The Facts/Database Section.......................................................................421 The Clauses Section..................................................................................422 The Constants Section...............................................................................427 Conditional Compilation...........................................................................429 Including Files in Your Program...................................................................430 Compiler Directives......................................................................................430 Visual Prolog Memory Management.............................................................437 Releasing Spare Memory Resources.........................................................437 Modular Programming..................................................................................438 Global Declarations...................................................................................438 CHAPTER 296.................................................................................................441 Interfacing with Other Languages.....................................................................441 Using DLL’s.................................................................................................441 Calling Other Languages from Visual Prolog...............................................442 Declaring External Predicates...................................................................442 Calling Conventions and Parameter Passing.............................................442 Domain Implementation................................................................................446 Simple Domains............................................................................................447 Complex Domains.....................................................................................447 Memory Considerations................................................................................449 Memory Alignment...................................................................................449 Memory Allocation...................................................................................451 Examples.......................................................................................................455 List Handling.............................................................................................455 Calling Prolog from Foreign Languages...................................................458 Calling an Assembler Routine from Visual Prolog....................................460 Index.................................................................................................................464

Using This Manual If you have never programmed in Prolog before, you should read all of this manual. Chapters 1-2 cover Prolog fundamentals, and you should read them before attempting any serious application development. The later chapters become very important as soon as you want to do serious programming. If you program in a procedural programming language such as C, Pascal, or Basic, pay close attention to the procedural discussions. At the end of Chapter 3, you will find a procedural overview of the material covered in the first three tutorial chapters. We also provide procedural discussions of recursion in Chapter 4. If you have programmed in other Prologs and have a good understanding of Prolog fundamentals, you won't need much review. However, Visual Prolog has

several extensions and is different from interpreted Prologs. We recommend that you read the release notes and Chapters 5 as an introduction. Chapter 6 explains the structure of a Visual Prolog program and Chapter 7 introduces the declarations. We also recommend that you read Chapter 8 on Visual Prolog's facts section, and Chapter 9, on the external database. Chapters 10 through 11 provide valuable information that you will need if you plan to do serious programming. If you think you could use a review of Visual Prolog programming, we recommend that you read from Chapter 12 on. This user's guide is divided into four parts: a short introduction to the Visual Prolog environment; then the first ten tutorial chapters--which teach you how to program in Visual Prolog; then six chapters--which gives an overview of the predefined features of Visual Prolog - the standard predicates, the last part gives a complete systematic overview of the language, modular programming and interfacing to other languages. Here's a summary of each chapter in this book: Part 1: Introduction to Visual Prolog Chapter 1: Getting Started describes how to run Visual Prolog on your system, provides a quick guide to the menus and the editor, and takes you through the steps involved in creating, running, and saving your first Visual Prolog program. Part 2: Tutorial Chapters 2--10: Learning Visual Prolog Chapter 2: Fundamentals of Prolog provides a general introduction to Prolog from a natural language perspective and discusses how to convert natural language statements and questions into Prolog facts, rules, and queries. Chapter 3: Visual Prolog Programs covers Visual Prolog syntax, discusses the sections of a Visual Prolog program, and introduces programming in Visual Prolog. Chapter 4: Unification and Backtracking describes how Visual Prolog solves problems and assigns values to variables. Chapter 5: Simple and Compound Objects discusses declaring and building structures in Visual Prolog. Chapter 6: Repetition and Recursion explains how to write repetitive procedures using backtracking and recursion; also introduces recursive structures and trees.

Chapter 7: Lists and Recursion introduces lists and their use through recursion, as well as covers general list manipulation. Chapter 8: The facts section discusses Visual Prolog's facts section for adding facts to your program at run time and for storing global information. Chapter 9: Classes and Objects gives a short introduction to object oriented programming and introduces the object mechanism in Visual Prolog. Chapter 10: Arithmetic and Comparison introduces the full range of arithmetic and comparison functions built into Visual Prolog and gives examples that demonstrate how to use them. Chapter 11: Advanced Techniques controlling the flow analysis, using reference variables, pointers to predicates, the binary domain, term conversions, using the dynamic cut, tools and techniques for error and signal handling, and programming style for efficient programs. Part 3: Tutorial Chapters 13--14: Using Visual Prolog Chapter 12: Writing, Reading, and Files introduces I/O in Visual Prolog; covers reading and writing, and file- and directory-handling. Chapter 13: String-Handling in Visual Prolog covers string manipulation, including string comparison and string conversion, plus constructing and parsing strings. Chapter 14: The External Database System covers Visual Prolog's external database system: chained data, B+ trees, storing data (in EMS, conventional memory, and hard disk), and sorting data. Includes examples of constructing real database applications. Chapter 15: System-Level Programming introduces the low-level control supported within Visual Prolog: system calls, BIOS, low-level memory addressing, and bit manipulation. Chapter 16: Example Prolog Programs provides a diverse group of Prolog program examples demonstrating some of the elegant ways Prolog solves complex problems. Part 4: Reference Chapters 15--16: An overview Chapter 17 Elements of the Language gives a systematic overview of all the features in the Visual Prolog language. The chapter also introduces modular programming. Chapter 18 Interfacing with Other Languages gives a describtion on how to interface with C and other languages

PA RT Using the Environment

CHAPTER

Using the Visual Prolog Environment 18This chapter describes the basic operation of the Visual Prolog system focusing on running the examples described in this book. We assume, that you have experience using the Graphical User Interface system, the windowing system. This might be either Windows 3.x, Windows 95, Windows NT, Win-OS/2 or OS/2 PM. You should thus know about using menus, closing and resizing windows, loading a file in the File Open dialog etc. If you do not have this knowledge, you should not start off trying to create an application that runs under this environment. You must first learn to use the environment. If you are a beginner to Prolog, you don’t want to mix learning the Prolog language with the complexity of creating Windows applications with event handling and all the Windows options and possibilities. The code for the examples in this book are platform independent: They can run in DOS text mode, under UNIX, or with the adding of a little extra code for the User Interface handling, in a Windowing environment like MS Windows or OS/2 PM. We do suggest, that you at an early stage try compiling some of the examples in the VPI subdirectory, and work your way through the Guided Tour in the Getting Started book. This gives you an impression what can be done with Visual Prolog just so you know what you should be able to do, when you have learned to master Visual Prolog. However, if you are going to seriously use the Visual Prolog system, you need to learn the basic concepts of Prolog properly. You will not be able to build a skyscraper without having a solid foundation. In Visual Prolog, the foundation is understanding the Prolog language and the VPI layer.

What Needs to be Installed for This Book ? To Run and test the examples in this book you need during installation to: Install the Visual Development Environment. You can choose the Win16 version or the Win32 version if you own the Professional version of In the Dialog for choosing Documentation; install "Answers to exercises" and "Examples". xiii

Starting the Visual Prolog Environment The installation program will install a program group with an Icon which are normally used to start the Visual Prolog Environment. However, there are many ways to start an application in the GUI World, if you prefer another method, you can just start the VIP.EXE down in the BIN\WIN\16 or the BIN\WIN\32 directories under the main VIP directory. If Visual Prolog had an open project (a .VPR file) last time it was closed on your computer, it will automatically reopen this project next time it starts. For the examples in this manual, you do not need to have an open project.

Opening an Editor Window To create a new edit window, you can use the menu command File | New. This will bring up a new editor window with the title "NONAME". The editor is a fairly standard text editor. It is documented in the VDE manual, but you should be able to use cursor keys and the mouse as you are used to in other editors. It supports cut, copy and Paste and Undo / Redo, which you can all be activated from the Edit menu. Also the Edit menu shows the accelerator keys associated for these actions.

Running and Testing a Program To check, that your system is set up properly, you should try to type in the following text in the window: GOAL write("Hello world"),nl.

This is what is called a GOAL in the Prolog terminology, and this is enough to be a program that can be executed. To execute the GOAL, you should activate the menu item Project | Test Goal, or just press the accelerator key Ctrl+G. If your system is installed properly, your screen will look like the following:

The result of the execution will come up in a separate window, which you must close before you can test another GOAL. (The Visual Prolog system treats the GOAL as a program which it compiles, links and generates a Windows executable from. It uses the EASYWIN strategy, which is described in the VDE manual.).

Loading Examples From Disk The examples in this manual are provided on the disk. You can find them in the subdirectory: DOC\EXAMPLES. Also there are some answers to exercises in the subdirectory DOC\ANSWERS. They're named after the chapter they appear in: chCCeNN.pro, where CC will be 02, 03, 04, etc. according to chapter, and NN is the example number within that chapter (01, 02, 03, etc.).

You should now try to open one of these example, and test one. You should just use the File | Open command to open the file, and the press Ctrl+G to test the GOAL.

Handling Errors If you, like all programmers do, happen to make some errors in your program, the Visual Prolog system will display an error window, which contains a list of errors. You can double click on one of these errors to come to the position of the error in the source text.

PA RT

Tutorial Chapters 19 -- 20: Learning Prolog2122Visual Prolog

xvii

CHAPTER

Prolog Fundamentals 24This is the first in a sequence of chapters giving a step-by-step tutorial introduction to the Visual Prolog language. We begin this chapter with an introduction to programming in logic. After that we discuss some of Prolog's basic concepts, including clauses, predicates, variables, goals, and matching.

PROgramming in LOGic In Prolog, you arrive at solutions by logically inferring one thing from something already known. Typically, a Prolog program isn't a sequence of actions--it's a collection of facts together with rules for drawing conclusions from those facts. Prolog is therefore what is known as a declarative language. Prolog is based on Horn clauses, which are a subset of a formal system called predicate logic. Don't let this name scare you. Predicate logic is simply a way of making it clear how reasoning is done. It's simpler than arithmetic once you get used to it. Prolog uses a simplified variation of predicate logic syntax because it provides an easy-to-understand syntax very similar to natural language, and because computers are not as fast, large, or as inexpensive as we would like. If Prolog were to accept English statements, the compiler would need to know every possible way something could be worded in English. In many cases, it would take many times longer to translate the program into something the computer understands than it would to run the program. The computer hardware needed to run such a system would be monstrous. Prolog includes an inference engine, which is a process for reasoning logically about information. The inference engine includes a pattern matcher, which retrieves stored (known) information by matching answers to questions. Prolog tries to infer that a hypothesis is true (in other words, answer a question) by questioning the set of information already known to be true. Prolog's known world is the finite set of facts (and rules) that are given in the program. One important feature of Prolog is that, in addition to logically finding answers to the questions you pose, it can deal with alternatives and find all possible solutions

rather than only one. Instead of just proceeding from the beginning of the program to the end, Prolog can actually back up and look for more than one way of solving each part of the problem. Predicate logic was developed to easily convey logic-based ideas into a written form. Prolog takes advantage of this syntax to develop a programming language based on logic. In predicate logic, you first eliminate all unnecessary words from your sentences. You then transform the sentence, placing the relationship first and grouping the objects after the relationship. The objects then become arguments that the relationship acts upon. For example, the following sentences are transformed into predicate logic syntax: Natural Language:

Predicate Logic:

A car is fun. A rose is red.

fun(car). red(rose).

Bill likes a car if the car is fun.

likes(bill, Car) if fun(Car).

Sentences: Facts and Rules A Prolog programmer defines objects and relations, then defines rules about when these relations are true. For example, the sentence Bill likes dogs.

shows a relation between the objects Bill and dogs; the relation is likes. Here is a rule that defines when the sentence Bill likes dogs. is true: Bill likes dogs i f

the dogs are nice.

Facts: What Is Known In Prolog, a relation between objects is called a predicate. In natural language, a relation is symbolized by a sentence. In the predicate logic that Prolog uses, a relation is summarized in a simple phrase--a fact--that consists of the relation name followed by the object or objects (enclosed in parentheses). As with a sentence, the fact ends with a period (.). Here are some more facts expressing "likes" relations in natural language: Bill likes Cindy. Cindy likes Bill. Bill likes dogs.

Here are the same facts, written in Prolog syntax:

xix

likes(bill, cindy). likes(cindy, bill). likes(bill, dogs).

Facts can also express properties of objects as well as relations; in natural language "Kermit is green" and "Caitlin is a girl." Here are some Prolog facts that express these same properties: green(kermit). girl(caitlin).

Rules: What You Can Infer from Given Facts Rules enable you to infer facts from other facts. Another way to say this is that a rule, as conclusions is a conclusion that is known to be true if one or more other conclusions or facts are found to be true. Here are some rules concerning a "likes" relation: Cindy likes everything that Bill likes. Caitlin likes everything that is green.

Given these rules, you can infer from the previous facts some of the things that Cindy and Caitlin like: Cindy likes Cindy. Caitlin likes Kermit.

To encode these same rules into Prolog, you only need to change the syntax a little, like this: likes(cindy, Something):- likes(bill, Something). likes(caitlin, Something):- green(Something).

The :- symbol is simply pronounced "if", and serves to separate the two parts of a rule: the head and the body. You can also think of a rule as a procedure. In other words, these rules likes(cindy, Something):- likes(bill, Something) likes(caitlin, Something):- green(Something).

also mean "To prove that Cindy likes something, prove that Bill likes that same thing" and "To prove that Caitlin likes something, prove that it is green." In the sameside effects procedural way, a rule can ask Prolog to perform actions other than proving things--such as writing something or creating a file.

Queries Once we give Prolog a set of facts, we can proceed to ask questions concerning these facts; this is known as querying the Prolog system. We can ask Prolog the same type of questions that we would ask you about these relations. Based upon the known facts and rules given earlier, you can answer questions about these relations, just as Prolog can. In natural language, we ask you: Does Bill like Cindy?

In Prolog syntax, we ask Prolog: likes(bill, cindy).

Given this query, Prolog would answer yes

because Prolog has a fact that says so. As a little more complicated and general question, we could ask you in natural language: What does Bill like?

In Prolog syntax, we ask Prolog: likes(bill, What).

Notice that Prolog syntax does not change when you ask a question: this query looks very similar to a fact. However, it is important to notice that the second object--What--begins with a capital letter, while the first object--bill--does not. This is because bill is a fixed, constant object--aconstants known value--but What is a variable. Variables always begin with an upper-case letter or an underscore. Prolog always looks for an answer to a query by starting at the top of the facts. It looks at each fact until it reaches the bottom, where there are no more. Given the query about what Bill likes, Prolog will return What=cindy What=dogs 2 Solutions

This is because Prolog knows likes(bill, cindy).

and

xxi

likes(bill, dogs).

We hope that you draw the same conclusion. If we were to ask you (and Prolog): What does Cindy like? likes(cindy, What).

Prolog would answer What = bill What = cindy What = dogs 3 solutions

This is because Prolog knows that Cindy likes Bill, and that Cindy likes what Bill likes, and that Bill likes Cindy and dogs. We could ask Prolog other questions that we might ask a person; however, a question such as "What girl does Bill like?" will yield no solution because Prolog, in this case, knows no facts about girls, and it can't draw any conclusions based on material not known (supplied to it). In this example, we have not given Prolog any relation or property to determine if any of the objects are girls. Putting Facts, Rules, and Queries Together Suppose you have the following facts and rules: A fast car is fun. A big car is nice. A little car is practical. Bill likes a car if the car is fun.

When you read these facts, you can deduce that Bill likes a fast car. In much the same way, Prolog will come to the same conclusion. If no fact were given about fast cars, then you would not be able to logically deduce what kind of a car Bill likes. You could take a guess at what kind of a car might be fun, but Prolog only knows what you tell it; Prolog does not guess. Here's an example demonstrating how Prolog uses rules to answer queries. Look at the facts and rules in this portion of Program 1:

likes(ellen, tennis). likes(john, football). likes(tom, baseball). likes(eric, swimming). likes(mark, tennis). likes(bill, Activity):- likes(tom, Activity).

The last line in Program 1 is a rule: likes(bill, Activity):- likes(tom, Activity).

This rule corresponds to the natural language statement Bill likes an activity if Tom likes that activity.

In this rule, the head is likes(bill, Activity), and the body is likes(tom, Notice that there is no fact in this example about Bill liking baseball. For Prolog to discover if Bill likes baseball, you can give the query Activity).

likes(bill, baseball).

When attempting to find a solution to this query, Prolog will use the rule: likes(bill, Activity):- likes(tom, Activity).

Load Program ch02e01.pro into the Prolog System and run it. PREDICATES nondeterm likes(symbol,symbol) CLAUSES likes(ellen,tennis). likes(john,football). likes(tom,baseball). likes(eric,swimming). likes(mark,tennis). likes(bill,Activity):likes(tom, Activity). GOAL likes(bill, baseball).

The system replies in the Dialog window yes

It has combined the rule likes(bill, Activity):- likes(tom, Activity).

with the fact xxiii

likes(tom, baseball).

to decide that likes(bill, baseball).

Try also this query: likes(bill, tennis).

The system replies no

Visual Prolog replies no to the latest query ("Does Bill like tennis?") because There is no fact that says Bill likes tennis. Bill's relationship with tennis can't be inferred using the given rule and the available facts. Of course, it may be that Bill absolutely adores tennis in real life, but Visual Prolog's response is based only upon the facts and the rules you have given it in the program.

Variables: General Sentences In Prolog, variables enable you to write general facts and rules and ask general questions. In natural language, you use variables in sentences all the time. A typical general statement in English could be Bill likes the same thing as Kim.

As we mentioned earlier in this chapter, to represent a variable in Prolog, the first character of the name must be an upper-case letter or an underscore. For example, in the following line, Thing is a variable. likes(bill, Thing):- likes(kim, Thing).

In the preceding discussion of rules, you saw this line: likes(cindy, Something):- likes(bill, Something).

The object Something begins with a capital letter because it is a variable; it must be able to match anything that Bill likes. It could equally well have been called X or Zorro. The objects bill and cindy begin with lower-case letters because they are not variables--instead, they are symbols, having a constant value. Visual Prolog can

also handle arbitrary text strings, much like we've been handling symbols above, if the text is surrounded by double quotes. Hence, the token bill could have been written as "Bill", if you wanted it to begin with an upper-case letter.

Overview 1. A Prolog program is made up of two types of phrases (also known as clauses): facts and rules. Facts are relations or properties that you, the programmer, know to be true. Rules are dependent relations; they allow Prolog to infer one piece of information from another. A rule becomes true if a given set of conditions is proven to be true. Each rule depends upon proving its conditions to be true. 2. In Prolog, all rules have two parts: a head and a body separated by the special :- token. The head is the fact that would be true if some number of conditions were true. This is also known as the conclusion or the dependent relation. The body is the set of conditions that must be true so that Prolog can prove that the head of the rule is true. 3. As you may have already guessed, facts and rules are really the same, except that a fact has no explicit body. The fact simply behaves as if it had a body that was always true. 4. Once you give Prolog a set of facts and/or rules, you can proceed to ask questions concerning these; this is known as querying the Prolog system. Prolog always looks for a solution by starting at the top of the facts and/or rules, and keeps looking until it reaches the bottom. 5. Prolog's inference engine takes the conditions of a rule (the body of the rule) and looks through its list of known facts and rules, trying to satisfy the conditions. Once all the conditions have been met, the dependent relation (the head of the rule) is found to be true. If all the conditions can't be matched with known facts, the rule doesn't conclude anything. Exercises Write natural language sentences that represent what these Prolog facts might convey to a human reader. (Remember that, to the computer, these facts are simple pieces of information that can be used for matching answers to questions.) 1.

likes(jeff, painting).

xxv

male(john).

building("Empire State Building", new_york).

person(roslin, jeanie, "1429 East Sutter St.", "Scotts Valley", "CA", 95066).

Write Visual Prolog facts that represent the following natural language statements: Helen likes pizza. San Francisco is in California. Amy's telephone number is 476-0299. Len's father is Alphonso Grenaldi.

From Natural Language to Prolog Programs In the first section of this chapter we talked about facts and rules, relations, general sentences, and queries. Those words are all part of a discussion of logic and natural language. Now we're going to discuss the same ideas, but we're going to use more Prolog-ish words, like clauses, predicates, variables, and goals.

Clauses (Facts and Rules) Basically, there are only two types of phrases that make up the Prolog language; a phrase can be either a fact or a rule. These phrases are known in Prolog as clauses. The heart of a Prolog program is made up of clauses. More About Facts A fact represents one single instance of either a property of an object or a relation between objects. A fact is self-standing; Prolog doesn't need to look any further for confirmation of the fact, and the fact can be used as a basis for inferences. More About Rules In Prolog, as in ordinary life, it is often possible to find out that something is true by inferring it from other facts. The Prolog construct that describes what you can infer from other information is a rule. A rule is a property or relation known to be true when some set of other relations is known. Syntactically, these relations are separated by commas, as we illustrate in example 1 below.

Examples of Rules 1. This first example shows a rule that can be used to conclude whether a menu item is suitable for Diane. Diane is a vegetarian and eats only what her doctor tells her to eat.

Given a menu and the preceding rule, you can conclude if Diane can order a particular item on the menu. To do this, you must check to see if the item on the menu matches the constraints given. a. Is

Food_on_menu

a vegetable?

b. Is Food_on_menu on the doctor's list? c. Conclusion: If both answers are yes, Diane can order Food_on_menu. In Prolog, a relationship like this must be represented by a rule because the conclusion is based on facts. Here's one way of writing the rule: diane_can_eat(Food_on_menu):vegetable(Food_on_menu), on_doctor_list(Food_on_menu).

Notice here the comma after vegetable(Food_on_menu). The comma introduces a conjunction of several goals, and is simply read as "and"; both vegetable(Food_on_menu) and on_doctor_list(Food_on_menu) must be true, for diane_can_eat(Food_on_menu) to be true. Suppose you want to make a Prolog fact that is true if Person1 is the parent of Person2. This is easy enough; simply state the Prolog fact parent(paul, samantha).

This shows that Paul is the parent of Samantha. But, suppose your Prolog database already has facts stating father relationships. For example, "Paul is the father of Samantha": father(paul, samantha).

And you also have facts stating mother relationships; "Julie is the mother of Samantha": mother(julie, samantha).

If you already had a collection of facts stating these father/mother relationships, it would be a waste of time to write parent facts into the database for each parent relationship.

xxvii

Since you know that Person1 is the parent of Person2 if Person1 is the father of Person2 or if Person1 is the mother of Person2, then why not write a rule to convey these constraints? After stating these conditions in natural language, it should be fairly simple to code this into a Prolog rule by writing a rule that states the relationships. parent(Person1, Person2):- father(Person1, Person2). parent(Person1, Person2):- mother(Person1, Person2).

These Prolog rules simply state that Person1 is the parent of Person2 if Person1 is the father of Person2. Person1 is the parent of Person2 if Person1 is the mother of Person2.

3. Here's another example: A person can buy a car if the person likes the car and the car is for sale.

This natural language relationship can be conveyed in Prolog with the following rule: can_buy(Name, Model):person(Name), car(Model), likes(Name, Model), for_sale(Model).

This rule shows the following relationship: Name can_buy Model if Name is a person and Model is a car and Name likes Model and Model is for sale.

This Prolog rule will succeed if all four conditions in the body of the rule succeed. 4. Here is a program designed to find solutions to this car-buying problem: /* Program ch025e02.pro */ PREDICATES nondeterm can_buy(symbol, symbol) nondeterm person(symbol) nondeterm car(symbol) likes(symbol, symbol) for_sale(symbol)

CLAUSES can_buy(X,Y):person(X), car(Y), likes(X,Y), for_sale(Y). person(kelly). person(judy). person(ellen). person(mark). car(lemon). car(hot_rod). likes(kelly, hot_rod). likes(judy, pizza). likes(ellen, tennis). likes(mark, tennis). for_sale(pizza). for_sale(lemon). for_sale(hot_rod).

What can Judy and Kelly buy? Who can buy the hot rod? You can try the following goals: can_buy(Who, What). can_buy(judy, What). can_buy(kelly, What). can_buy(Who, hot_rod).

Experiment! Add other facts and maybe even a rule or two to this Prolog program. Test the new program with queries that you make up. Does Prolog respond in a way you would expect it to? Exercises Write natural-language sentences corresponding to the following Visual Prolog rules: eats(Who, What):- food(What), likes(Who, What). pass_class(Who):- did_homework(Who), good_attendance(Who). does_not_eat(toby, Stuff):- food(Stuff), greasy(Stuff). owns(Who, What):- bought(Who, What).

xxix

Write Visual Prolog rules that convey the meaning of these natural-language sentences: a.

A person is hungry if that person's stomach is empty.

Everybody likes a job if it's fun and it pays well.

Sally likes french fries if they're cooked.

Everybody owns a car who buys one, pays for it, and keeps it.

Predicates (Relations) The symbolic name of a relation is called the predicate name. The objects that it relates are called its arguments; in the fact likes(bill, cindy)., the relation likes is the predicate and the objects bill and cindy are the arguments. Here are some examples of Prolog predicates with zero or more arguments: pred(integer, symbol) person(last, first, gender) run insert_mode birthday(firstName, lastName, date)

As we've shown here, a predicate might not have any arguments at all, but the use of such a predicate is limited. You can use a query such as person(rosemont,Name,male). to find out Mr. Rosemont's first name. But what can you do with the zero-argument query run? You can find out whether the clause run is in the program, or--if run is the head of a rule, you can evaluate that rule. This can be useful in a few cases--for instance, you might want to make a program behave differently depending on whether the clause insert_mode. is present.

Variables (General Clauses) In a simple query, you can use variables to ask Prolog to find who likes tennis. For example: likes(X, tennis).

This query uses the letter X as a variable to indicate an unknown person. Variable names in Visual Prolog must begin with a capital letter (or an underscore), after which any number of letters (upper-case or lower-case), digits, or underline characters (_) can be used. For example, the following are valid variable names:

My_first_correct_variable_name Sales_10_11_86

while the next three are invalid: 1stattempt second_attempt "disaster"

(Careful choice of variable names makes programs more readable. For example, likes(Person, tennis).

is better than likes(X, tennis).

because Person makes more sense than X.) Now try the goal GOAL likes(Person, tennis).

Visual Prolog replies Person=ellen Person=mark 2 Solutions

because the goal can be solved in just two ways; namely, by taking the variable Person and successively matching it with the values ellen and mark. In variable names, except for the first character (which must be an upper-case letter or an underscore), Visual Prolog allows lower-case or upper-case letters in any position. One way to make variable names more readable is by using mixed upper-case and lower-case letters, as in IncomeAndExpenditureAccount

How Variables Get Their Values You may have noticed that Prolog has no assignment statement; this is a significant distinction between Prolog and other programming languages. Variables in Prolog get their values by being matched to constants in facts or rules. Until it gets a value, a variable is said to be free; when it gets a value, it becomes bound. But it only stays bound for the time needed to obtain one solution to the query; then Prolog unbinds it, backs up, and looks for alternative solutions.

xxxi

This is a very important point: You can't store information by giving a value to a variable. Variables are used as part of the pattern-matching, process, not as a kind of information storage. Take a look at the following example, which uses program 3 to demonstrate how and when variables get their values. /* Program ch026e03.pro */ PREDICATES nondeterm likes(symbol,symbol) CLAUSES likes(ellen,reading). likes(john,computers). likes(john,badminton). likes(leonard,badminton). likes(eric,swimming). likes(eric,reading).

Consider this query: Is there a person who likes both reading and swimming? likes(Person, reading), likes(Person, swimming).

Prolog will solve the two parts of this query by searching the program's clauses from top to bottom. In the first part of the query likes(Person, reading)

the variable Person is free; its value is unknown before Prolog attempts to find a solution. On the other hand, the second argument, reading, is known. Prolog searches for a fact that matches the first part of the query. The first fact in the program likes(ellen, reading)

is a match (reading in the fact matches reading in the query), so Prolog binds the free variable Person to the value ellen, the relevant value in the fact. At the same time, Prolog places a pointer in the list of facts indicating how far down the search procedure has reached. Next, in order for the query to be fully satisfied (find a person who likes both reading and swimming), the second part must also be fulfilled. Since Person is now bound to ellen, Prolog must search for the fact likes(ellen, swimming)

Prolog searches for this fact from the beginning of the program, but no match occurs (because there is no such fact in the program). The second part of the query is not true when Person is ellen. Prolog now "unbinds" Person and attempts another solution of the first part of the query with Person once again a free variable. The search for another fact that fulfills the first part of the query starts from the pointer in the list of facts. (This returning to the place last marked is known as backtracking, which we'll cover in chapter 27.) Prolog looks for the next person who likes reading and finds the fact likes(eric, reading). Person is now bound to eric, and Prolog tries once again to satisfy the second part of the query, this time by looking in the program for the fact likes(eric, swimming)

This time it finds a match (the last clause in the program), and the query is fully satisfied. Prolog returns Person=eric 1 Solution

Anonymous Variables Anonymous variables enable you to unclutter your programs. If you only need certain information from a query, you can use anonymous variables to ignore the values you don't need. In Prolog, the anonymous variable is represented by a lone underscore ("_"). The following parents example demonstrates how the anonymous variable is used. Load Program 4. /* Program ch028e04.pro */ PREDICATES male(symbol) female(symbol) nondeterm parent(symbol, symbol) CLAUSES male(bill). male(joe). female(sue). female(tammy).

xxxiii

parent(bill,joe). parent(sue,joe). parent(joe,tammy).

The anonymous variable can be used in place of any other variable. The difference is that the anonymous variable will never get set to a value. For example, in the following query, you need to know which people are parents, but you don't need to know who their children are. Prolog realizes that each time you use the underscore symbol in the query, you don't need information about what value is represented in that variable's place. GOAL parent(Parent, _).

Given this query, Prolog replies Parent=bill Parent=sue Parent=joe 3 Solutions

In this case, because of the anonymous variable, Prolog finds and reports three parents, but it does not report the values associated with the second argument in the parent clause. Anonymous variables can also be used in facts. The following Prolog facts owns(_, shoes). eats(_).

could be used to express the natural language statements Everyone owns shoes. Everyone eats.

The anonymous variable matches anything. A named variable would work equally well in most cases, but its name would serve no useful purpose.

Goals (Queries) Up to now, we've been mixing the word query when talking about the questions you ask Prolog, with the more common name goal, which we'll use from now on. Referring to queries as goals should make sense: when you query Prolog, you are actually giving it a goal to accomplish ("Find an answer to this question, if one exists: ...").

Goals can be simple, such as these two: likes(ellen, swimming). likes(bill, What).

or they can be more complex. In the "Variables" section of this chapter, you saw a goal made up of two parts: likes(Person, reading), likes(Person, swimming).

A goal made up of two or more parts is known as a compound goal, and each part of the compound goal is called a subgoal. Often you need to know the intersection of two goals. For instance, in the previous parents example, you might also need to know which persons are male parents. You can get Prolog to search for the solutions to such a query by setting a compound goal. Load the Program 4 and enter the following compound goal: Goal parent(Person, _), male(Person).

Prolog will first try to solve the subgoal parent(Person, _)

by searching the clauses for a match, then binding the variable Person to a value returned by parent (Person is a parent). The value that parent returns will then provide the second subgoal with the value on which to search (Is Person--now bound--a male?). male(Person)

If you entered the goal correctly, Prolog will answer Person=bill Person=joe 2 Solutions

Compound Goals: Conjunctions and Disjunctions As you have seen, you can use a compound goal to find a solution where both subgoal A and subgoal B are true (a conjunction), by separating the subgoals with a comma, but this is not all. You can also find a solution where subgoal A or subgoal B is true (a disjunction), by separating the subgoals with a semicolon. Here's an example program illustrating this idea:

xxxv

/* Program ch029e05.pro */ PREDICATES car(symbol,long,integer,symbol,long) truck(symbol,long,integer,symbol,long) nondeterm vehicle(symbol,long,integer,symbol,long) CLAUSES car(chrysler,130000,3,red,12000). car(ford,90000,4,gray,25000). car(datsun,8000,1,red,30000). truck(ford,80000,6,blue,8000). truck(datsun,50000,5,orange,20000). truck(toyota,25000,2,black,25000). vehicle(Make,Odometer,Age,Color,Price):car(Make,Odometer,Age,Color,Price); truck(Make,Odometer,Age,Color,Price).

Load and run this program, then try the goal GOAL car(Make, Odometer, Years_on_road, Body, 25000).

This goal attempts to find a car described in the clauses that costs exactly $25,000. Prolog replies Make=ford, Odometer=90000, Years_on_road=4, Body=gray 1 Solution

But this goal is slightly unnatural, since you'd probably rather ask a question like: Is there a car listed that costs less than $25,000?

You can get Visual Prolog to search for a solution by setting this compound goal: car(Make, Odometer, Years_on_road, Body, Cost), Cost < 25000.

/*subgoal A and*/ /*subgoal B */

This is known as a conjunction. To fulfill this compound goal, Prolog will try to solve the subgoals in order. First, it will try to solve car(Make, Odometer, Years_on_road, Body, Cost).

and then Cost < 25000.

with the variable Cost referring to the same value in both subgoals. Try it out now. Note: The subgoal Cost < 25000 involves the relation less than, which is built into the Visual Prolog system. The less than relation is no different from any other relation involving two numeric objects, but it is more natural to place the symbol for it between the two objects. Now we will try to see if the following, expressed in natural language, is true:goals, disjunctivedisjunctive goals queries, disjunctive Is there a car listed that costs less than $25,000?, or is there a truck listed that costs less than $20,000?

Prolog will search for a solution if you set this compound goal: car(Make,Odometer,Years_on_road,Body,Cost), Cost<25000 ; /* subgoal A or */ truck(Make,Odometer,Years_on_road,Body,Cost), Cost < 20000. /* subgoal B */

This kind of compound goal is known as a disjunction. This one sets up the two subgoals as alternatives, much as though they were two clauses for the same rule. Prolog will then find any solution that satisfies either of the subgoals. To fulfill this compound goal, Prolog will try to solve the first subgoal ("find a car ..."), which is composed of these subgoals: car(Make, Odometer, Years_on_road, Body, Cost.)

and Cost < 25000.

If a car is found, the goal will succeed; if not, Prolog will try to fulfill the second compound goal ("find a truck ..."), made up of the subgoals truck(Make, Odometer, Years_on_road, Body, Cost),

and Cost < 20000.

Comments It's good programming style to include comments in your program to explain things that might not be obvious to someone else (or to you in six months). This

xxxvii

makes the program easy for you and others to understand. If you choose appropriate names for variables, predicates, and domains, you'll need fewer comments, since the program will be more self-explanatory. Multiple-line comments must begin with the characters /* (slash, asterisk) and end with the characters */ (asterisk, slash). To set off single-line comments, you can use these same characters, or you can begin the comment with a percent sign (%). /* This is an example of a comment */ % This is also a comment /***************************************/ /* and so are these three lines */ /***************************************/ /*You can also nest a Visual Prolog comment /*within a comment*/ like this */

In Visual Prolog 5.0 you can also use a comment after de decalratition of a domain. DOMAINS articles = book(STRING title, STRING author); horse(STRING name) PREDICATES conv(STRING uppercase,STRING lowercase)

The words title, author, name, uppercase and lowercase will be ignored by the compiler, but makes the program much more readable.

What Is a Match? In the previous sections of this chapter, we've talked about Prolog "matching answers to questions", "finding a match", "matching conditions with facts", "matching variables with constants", and so on. In this section we explain what we mean when we use the term "match." There are several ways Prolog can match one thing to another. Obviously, identical structures match each other; parent(joe,tammy) matches parent(joe,tammy).

However, a match usually involves one or more free variables. For example, with X free, parent(joe,X) matches parent(joe,tammy)

and X takes on (is bound to) the value tammy. If X is already bound, it acts exactly like a constant. Thus, if X is bound to the value tammy, then parent(joe,X)

matches

parent(joe,X)

would not match

parent(joe,tammy) but parent(joe,millie)

The second instance doesn't match because, once a variable becomes bound, its value can't change. How could a variable, bindings already be bound when Prolog tries to match it with something? Remember that variables don't store values--they only stay bound for the length of time needed to find (or try to find) one solution to one goal. So the only way a variable could be bound before trying a match is that the goal involves more than one step, and the variable became bound in a previous step. For example, parent(joe,X), parent(X,jenny)

is a legitimate goal; it means, "Find someone who is a child of Joe and a parent of Jenny." Here X will already be bound when the subgoal parent(X,jenny) is reached. If there is no solution to parent(X,jenny), Prolog will unbind X and go back and try to find another solution to parent(joe,X), then see if parent(X,jenny) will work with the new value of X. Two free variables can even match each other. For example, parent(joe,X) matches parent(joe,Y)

binding the variables X and Y to each other. As long as the binding lasts, X and Y are treated as a single variable, and if one of them gets a value the other one will immediately have the same value. When free variables are bound to each other like this, they're called pointers, shared free sharing variables. Some really powerful programming techniques involve binding together variables that were originally separate. In Prolog, variable bindings (values) are passed in two ways: in and out. The direction in which a value is passed is referred to as its flow pattern. When a variable is passed into a clause, it is an input , argument, signified by (i); when passed out of a clause, a variable is an output argument, signified by (o).

xxxix

Summary These are the ideas we've introduced in this chapter: 1. A Prolog program is made up of clauses, which conceptually are two types of phrases: facts and rules. Facts are relations or properties that you, the programmer, know to be true. Rules are dependent relations; they allow Prolog to infer one piece of information from another. 2. Facts have the general form: property(object1, object2, ..., objectN)

or relation(object1, object2, ..., objectN)

where a property is a property of the objects and a relation is a relation between the objects. As far as Prolog programming goes, the distinction doesn't exist and we will refer to both as relations in this book. 3. Each fact given in a program consists of either a relation that affects one or more objects or a property of one or more objects. For example, in the Prolog fact likes(tom, baseball).

the relation is likes, and the objects are tom and baseball; Tom likes baseball. Also, in the fact left_handed(benjamin)

the property is left_handed and the object is benjamin; in other words, Benjamin is left-handed. 4. Rules have the general form Head:-

, which looks like this in a program:

Body

relation(object,object,...,object):relation(object,...,object), . . relation(object,...,object).

5. You are free to choose names for the relations and objects in your programs, subject to the following constraints:

Object names must begin with a lower-case letter, followed by any number of characters; characters are upper-case or lower-case letters, digits, and underscores. Properties and relation names must start with a lower-case letter, followed by any combination of letters, digits, and underscore characters. 6. A predicate is the symbolic name (identifier) for a relation and a sequence of arguments. A Prolog program is a sequence of clauses and directives, and a procedure is a sequence of clauses defining a predicate. Clauses that belong to the same predicate must follow one another. 7. Variables enable you to write general facts and rules and ask general questions. Variable names in Visual Prolog must begin with a capital letter or an underscore character (_), after which you can use any number of letters (upper-case or lower-case), digits, or underscores. Variables in Prolog get their values by being matched to constants in facts or rules. Until it gets a value, a variable is said to be free; when it gets a value, it becomes bound. You can't store information globally by binding a value to a variable, because a variable is only bound within a clause. 8. If you only need certain information from a query, you can use anonymous variables to ignore the values you don't need. In Prolog, the anonymous variable is represented by a lone underscore (_). The anonymous variable can be used in place of any other variable; it matches anything. The anonymous variable will never get set to a value. 9. Asking Prolog questions about the facts in your program is known as querying the Prolog system; the query is commonly called a goal. Prolog tries to satisfy a goal (answer the query) by starting at the top of the facts, looking at each fact until it reaches the bottom. 10. A compound goal is a goal made up of two or more parts; each part of the compound goal is called a subgoal. Compound goals can be conjunctions (subgoal A and subgoal B) or disjunctions (subgoal A or subgoal B). 11. Comments make your programs easier to read; you can enclose a comment with delimiters /* like this */ or precede a single-line comment with a percent sign, % like this. 12. There are several ways Prolog can match one thing to another:

xli

Identical structures match each other. A free variable matches a constant or a previously-bound variable (and becomes bound to that value). Two free variables can match (and be bound to) each other. As long as the binding lasts, they are treated as a single variable; if one gets a value the other will immediately have the same value.

CHAPTER

Visual Prolog Programs 31Visual PrologThe syntax of Visual Prolog is designed to express knowledge about properties and relationships. You've already seen the basics of how this is done; in Chapter 32 you learned about clauses (facts and rules), predicates, variables, and goals. Unlike other versions of Prolog, Visual Prolog is a typed Prolog compiler; you declare the types of the objects that each predicate applies to. The type declarations allow Visual Prolog programs to be compiled right down to native machine code, giving execution speeds similar to those of compiled C and pascal. We discuss the four basic sections of a Visual Prolog program--where you declare and define the predicates and arguments, define rules, and specify the program's goal--in the first part of this chapter. In the second part of this chapter we take a closer look at declarations and rule syntax. Then, at the end of this chapter, we briefly introduce the other sections of a Visual Prolog program, including the facts, constants, and various global sections, and compiler directives.

Visual Prolog's Basic Program Sections Visual PrologGenerally, a Visual Prolog program includes four basic program sections. These are the clauses section, the predicates section, the domains section, and the goal section. The clauses section is the heart of a Visual Prolog program; this is where you put the facts and rules that Visual Prolog will operate on when trying to satisfy the program's goal.

The predicates section is where you declare your predicates and the domains (types) of the arguments to your predicates. (You don't need to declare Visual Prolog's built-in predicates.) The domains section is where you declare any domains you're using that aren't Visual Prolog's standard domains. (You don't need to declare standard domains.) The goal section is where you put the starting goal for a Visual Prolog program.

The Clauses Section The clauses section is where you put all the facts and rules that make up your program. Most of the discussion in Chapter 33 was centered around the clauses (facts and rules) in your programs; what they convey, how to write them, and so on. If you understand what facts and rules are and how to write them in Prolog, you know what goes in the clauses section. Clauses for a given predicate must be placed together in the clauses section; a sequence of clauses defining a predicate is called a procedure. When attempting to satisfy a goal, Visual Prolog will start at the top of the clauses section, looking at each fact and rule as it searches for a match. As Visual Prolog proceeds down through the clauses section, it places internal pointers next to each clause that matches the current subgoal. If that clause is not part of a logical path that leads to a solution, Visual Prolog returns to the set pointer and looks for another match (this is backtracking, which we mentioned in Chapter 34).

The Predicates Section If you define your own predicate in the clauses section of a Visual Prolog program, you must declare it in a predicates section, or Visual Prolog won't know what you're talking about. When you declare a predicate, you tell Visual Prolog which domains the arguments of that predicate belong to. Visual Prolog comes with a wealth of built-in predicates. You don't need to declare any of Visual Prolog's built-in predicates that you use in your program. The Visual Prolog on-line help gives a full explanation of the built-in predicates. Facts and rules define predicates. The predicates section of the program simply lists each predicate, showing the types (domains) of its arguments. Although the clauses section is the heart of your program, Visual Prolog gets much of its

xliii

efficiency from the fact that you also declare the types of objects (arguments) that your facts and rules refer to. How to Declare User-Defined Predicates A predicate declaration begins with the predicate name, followed by an open (left) parenthesis. After the predicate name and the open parenthesis come zero or more arguments to the predicate. predicateName(argument_type1, argument_type2, ..., argument_typeN)

Each argument type is followed by a comma, and the last argument type is followed by the closing (right) parenthesis. Note that, unlike the clauses in the clauses section of your program, a predicate declaration is not followed by a period. The argument types are either standard domains or domains that you've declared in the domains section. Predicate Names The name of a predicate must begin with a letter, followed by a sequence of letters, digits, and underscores. The case of the letters does not matter, but we strongly recommend using only a lower-case letter as the first letter in the predicate name. (Other versions of Prolog don't allow predicate names to begin with upper-case letters or underscores, and future versions of Visual Prolog might not, either.) Predicate names can be up to 250 characters long. You can't use spaces, the minus sign, asterisks, slashes, or other nonalphanumeric characters in predicate names. Valid naming charactersnaming characters in Visual Prolog consist of the following: Upper-case Letters Lower-case Letters Digits Underscore character

A, B, ... , Z

a, b, ... , z

0, 1, ... , 9

All predicate names and arguments can consist of combinations of these characters, as long as you obey the rules for forming both predicate and argument names. Below are a few examples of legal and illegal predicate names. Legal Predicate Names

Illegal Predicate Names

fact

[fact]

is_a

*is_a*

has_a

has/a

patternCheckList

pattern-Check-List

choose_Menu_Item

choose Menu Item

predicateName

predicate<Name>

first_in_10

>first_in_10

Predicate Arguments The arguments to the predicates must belong to known Visual Prolog domains. A domain can be a standard domain, or it can be one you declare in the domains section. Examples If you declare a predicate section, like this:

my_predicate(symbol,

integer)

in the predicates

PREDICATES my_predicate(symbol, integer)

you don't need to declare its arguments' domains in a domains section, because symbol and integer are standard domains. But if you declare a predicate my_predicate(name, number) in the predicates section, like this: PREDICATES my_predicate(name, number)

you will need to declare suitable domains for name and number. Assuming you want these to be symbol and integer respectively, the domain declaration looks like this: DOMAINS name = symbol number = integer PREDICATES my_predicate(name, number)

This program excerpt shows some more predicate and domain declarations: DOMAINS person, activity = symbol car, make, color = symbol mileage, years_on_road, cost = integer

xlv

PREDICATES likes(person, activity) parent(person, person) can_buy(person, car) car(make, mileage, years_on_road, color, cost) green(symbol) ranking(symbol, integer)

This excerpt specifies the following information about these predicates and their arguments: The predicate likes takes two arguments (person and activity), both of which belong to unique symbol domains (which means that their values are names rather than numbers). The predicate parent takes two person arguments, where person is a symbol type. The predicate can_buy takes two arguments, person and car, which are also both symbol types. The predicate car takes five arguments: make and color are of unique symbol domains, while mileage, years_on_road, and cost are of unique integer domains. The predicate green takes one argument, a symbol: there is no need to declare the argument's type, because it's of the standard domain symbol. The predicate ranking takes two arguments, both of which belong to standard domains (symbol and integer), so there is no need to declare the argument types. Chapter 35, "Simple and Compound Objects," gives more detail about domain declarations.

The Domains Section In traditional Prolog there is only one type; the term. We have the same in Visual Prolog, but we are declaring what the domains of the arguments to the predicates actually are. Domains enable you to give distinctive names to different kinds of data that would otherwise look alike. In a Visual Prolog program, objects in a relation (the arguments to a predicate) belong to domains; these can be pre-defined domains, or special domains that you specify.

The domains section serves two very useful purposes. First, you can give meaningful names to domains even if, internally, they are the same as domains that already exist. Second, special domain declarations are used to declare data structures that are not defined by the standard domains. It is sometimes useful to declare a domain when you want to clarify portions of the predicates section. Declaring your own domains helps document the predicates that you define by giving a useful name to the argument type. Examples Here's an example to illustrate how declaring domains helps to document your predicates: Frank is a male who is 45 years old.

With the pre-defined domains, you come up with the following predicate declaration: person(symbol, symbol, integer)

This declaration will work fine for most purposes. But suppose you want to maintain your code months after you've finished writing it. The preceding predicate declaration won't mean much to you in six months. Instead, the following declarations will help you understand what the arguments in the predicate declaration stand for: DOMAINS name, sex = symbol age = integer PREDICATES person(name, sex, age)

One of the main advantages of this declarations, is that Visual Prolog can catch type errors, like the following obvious mistake: same_sex(X, Y) :person(X, Sex, _), person(Sex, Y, _).

Even though name and sex are both defined as symbol, they are not equivalent to each other. This enables Visual Prolog to detect an error if you accidentally swap them. This feature is very useful when your programs get large and complex.

xlvii

You might be wondering why we don't use special domains for all argument declarations, since special domains communicate the meaning of the argument so much better. The answer is that once an argument is typed to a specific domain, that domain can't be mixed with another domain you have declared, even if the domains are the same! So, even though name and sex are of the same domain (symbol), they can't be mixed. However, all user-defined domains can be matched with the pre-defined domains. This next example program will yield a type error when run: /* Program ch036e01.pro */ DOMAINS product,sum = integer PREDICATES add_em_up(sum,sum,sum) multiply_em(product,product,product) CLAUSES add_em_up(X,Y,Sum):Sum=X+Y. multiply_em(X,Y,Product):Product=X*Y.

This program does two things: It adds and it multiplies. Given the goal add_em_up(32, 54, Sum).

Visual Prolog will come up with Sum=86 1 Solution

which is the sum of the two integers you supplied the program. On the other hand, this program will also multiply two arguments with the multiply_em predicate. Now experiment with this program. If you need to figure out what the product of 31 and 13 is, you could enter the goal: multiply_em(31, 13, Product).

Visual Prolog would then respond with the correct answer. Product=403 1 Solution

But suppose you need the sum of 42 and 17; the goal for this would be add_em_up(42, 17, Sum).

Now you need to double the product of 31 and 17, so you write the following goal:

multiply_em(31, 17, Sum), add_em_up(Sum, Sum, Answer).

You might expect Visual Prolog to return Sum=527, Answer=1054 1 Solution

But, instead, you get a type error. What happened is that you tried to pass the resulting value of multiply_em (that is, of domain product), into the first and second arguments in add_em_up, which have domains of sum. This yields a type error because product is a different domain than sum. Even though both domains are really of type integer, they are different domains, and are treated as such. So, if a variable is used in more than one predicate within a clause, it must be declared the same in each predicate. Be sure that you fully understand the concept behind the type error given here; knowing the concept will avoid frustrating compiler error messages. Later in this chapter we will describe the different automatic and explicit type-conversions Visual Prolog offers. To further understand how you can use domain declarations to catch type errors, consider the following program example: /* Program ch037e02.pro */ DOMAINS brand,color = symbol age = byte price, mileage = ulong PREDICATES nondeterm car(brand,mileage,age,color,price) CLAUSES car(chrysler,130000,3,red,12000). car(ford,90000,4,gray,25000). car(datsun,8000,1,black,30000).

Here, the car predicate declared in the predicates section takes five arguments. One belongs to the age domain, which is of byte type. On the 'x86 family of CPUs, a byte is an 8-bit unsigned integer, which can take on values between 0 and 255, both inclusive. Similarly, the domains mileage and price are of type ulong, which is a 32-bit unsigned integer, and the domains brand and color are of type symbol. We'll discuss the built-in domains in greater detail in a moment. For now, load and run this program and try each of the following goals in turn.

xlix

car(renault, 13, 40000, red, 12000). car(ford, 90000, gray, 4, 25000). car(1, red, 30000, 80000, datsun).

Each goal produces a domain error. In the first case, for example, it's because age must be a byte. Hence, Visual Prolog can easily detect if someone typing in this goal has reversed the mileage and age objects in predicate car. In the second case, age and color have been swapped, and in the third case you get to find out for yourself where the mixups are.

The Goal Section Essentially, the goal section is the same as the body of a rule: it's simply a list of subgoals. There are two differences between the goal section and a rule: The goal keyword is not followed by :-. Visual Prolog automatically executes the goal when the program runs. It's as if Visual Prolog makes a call to goal, and the program runs, trying to satisfy the body of the goal rule. If the subgoals in the goal section all succeed, then the program terminates successfully. If, while the program is running, a subgoal in the goal section fails, then the program is said to have failed. (Although, from an external point of view, there isn't necessarily any difference; the program simply terminates.)

A Closer Look at Declarations and Rules Visual Prolog has several built-in standard domains. You can use standard domains when declaring the types of a predicate's arguments. Standard domains are already known to Visual Prolog and should not be defined in the domains section. We'll first look at all the integral ones, shown in Table 38.1. Table 39.2:Integral Standard Domains403 Domain

Description and implementation

short

A small, signed, quantity. All platforms

16 bits,2s comp

32768 .. 32767

ushort

A small, unsigned, quantity. All platforms

long

unsigned

32 bits,2s comp

-2147483648 .. 2147483647

A large, unsigned quantity All platforms

integer

0 .. 65535

A large signed quantity All platforms

ulong

16 bits

32 bits

0 .. 4294967295

A signed quantity, having the natural machine/platform architecture in question.

size

16bit platforms

16 bits,2s comp

-32768 .. 32767

32bit platforms

32 bits,2s comp

-2147483648 .. 2147483647

for

the

An unsigned quantity, having the natural size for the machine/platform architecture in question. 16bit platforms 32bit platforms

16 bits 32 bits

0 .. 65535 0 .. 4294967295

byte All platforms

³ 8 bits

All platforms

16 bits

All platforms

32 bits

0 .. 255

word 0 .. 65535

dword 0 .. 4294967295

Syntactically, a value belonging in one of the integral domains is written as a sequence of digits, optionally preceded by a minus-sign for the signed domains, with no white-space. There are also octal and hexadecimal syntaces for the integral domains; these will be illustrated in chapter 41. The byte, word, and dword domains are most useful when dealing with machinerelated quantities, except perhaps for the byte; an 8-bit integral quantity can prove quite relevant, as we have already seen. For general use, the integer and unsigned quantities are the ones to use, augmented by the short and long (and their unsigned counterparts) for slightly more specialized applications. Generally, the most efficient code results from using what's natural for the machine; a short li

is not as efficient on a '386 platform as a long, and a long is not as efficient on a '286 platform as a short, hence the different implementations of integer and unsigned. In domain declarations, the signed, and unsigned keywords may be used in conjunction with the byte, word, and dword built-in domains to construct new integral domains, as in DOMAINS i8 = signed byte

creating a new integral domain having a range of -128 to +127. The other basic domains are shown in table 42.4. Visual Prolog recognizes several other standard domains, but we cover them in other chapters, after you've got a good grasp of the basics. Table 43.5: Basic Standard Domains446 Domain

Description and implementation

char

A character, implemented as an unsigned byte. Syntactically, it is written as a character surrounded by single quotation marks: 'a'.

real

A floating-point number, implemented as 8 bytes in accordance with IEEE conventions; equivalent to C's double. Syntactically, a real is written with an optional sign (+ or -) followed by some digits DDDDDDD, then an optional decimal point (.) followed by more digits DDDDDDD, and an optional exponential part (e(+ or -)DDD): <+|-> DDDDD <.> DDDDDDD <e <+|-> DDD> Examples of real numbers: 42705 9999 86.72 9111.929437521e238 79.83e+21 Here 79.83e+21 means 79.83 x 10^21, just as in other languages. The permitted number range is 1 ) 10-307 to 1 ) 10308 (1e-307 to 1e+308). Values from the integral domains are automatically converted to real numbers when necessary.

string

A sequnce of characters, implemented as a pointer to a zero-

terminated byte array, as in C. Two formats are permitted for strings: 1. a sequence of letters, numbers and underscores, provided the first character is lower-case; or 2. a character sequence surrounded by a pair of double quotation marks. Examples of strings: telephone_number

"railway ticket"

"Dorid Inc"

Strings that you write in the program can be up to 255 characters long. Strings that the Visual Prolog system reads from a file or builds up internally can be up to 64K characters long on 16-bit platforms, and (theoretically) up to 4G long on 32-bit platforms.

symbol

A sequence of characters, implemented as a pointer to an entry in a hashed symbol-table, containing strings. The syntax is the same as for strings.

Symbols and strings are largely interchangeable as far as your program is concerned, but Visual Prolog stores them differently. Symbols are kept in a lookup table, and their addresses, rather than the symbols themselves, are stored to represent your objects. This means that symbols can be matched very quickly, and if a symbol occurs repeatedly in a program, it can be stored very compactly. Strings are not kept in a look-up table; Visual Prolog examines them characterby-character whenever they are to be matched. You must determine which domain will give better performance in a particular program. The following table gives some examples of simple objects that belong to the basic standard domains. Table 45.7: Simple Objects46 "&&", caitlin, "animal lover", b_l_t

(symbol or string)

-1, 3, 5, 0

(integer)

3.45, 0.01, -30.5, 123.4e+5

(real)

'a', 'b', 'c'

(char)

'/', '&'

liii

Typing Arguments in Predicate Declarations Declaring the domain of an argument in the predicates section is called typing the argument. For example, suppose you have the following relationship and objects: Frank is a male who is 45 years old.

The Prolog fact that corresponds to this natural language relation might be person(frank, male, 45).

In order to declare person as a predicate with these three arguments, you could place the following declaration in the predicates section: person(symbol, symbol, unsigned)

Here, you have used standard domains for all three arguments. Now, whenever you use the predicate person, you must supply three arguments to the predicate; the first two must be of type symbol, while the third argument must be an integer. If your program only uses standard domains, it does not need a domains section; you have seen several programs of this type already. Or, suppose you want to define a predicate that will tell you the position of a letter in the alphabet. That is, alphabet_position(Letter, Position)

will have Position = 1 if Letter = a, Position = 2 if Letter = b, and so on. The clauses for this predicate would look like this: alphabet_position(A_character, N).

If standard domains are the only domains in the predicate declarations, the program does not need a domains section. Suppose you want to define a predicate so that the goal will be true if A_character is the Nth letter in the alphabet. The clauses for this predicate would look like this: alphabet_position('a', alphabet_position('b', alphabet_position('c', ... alphabet_position('z',

1). 2). 3). 26).

You can declare the predicate as follows: PREDICATES alphabet_position(char, unsigned)

and there is no need for a domains section. If you put the whole program together, you get PREDICATES alphabet_position(char, integer) CLAUSES alphabet_position('a', 1). alphabet_position('b', 2). alphabet_position('c', 3). /* ... other letters go here ... */ alphabet_position('z', 26).

Here are a few sample goals you could enter: alphabet_position('a', 1). alphabet_position(X, 3). alphabet_position('z', What).

Exercises Program 4 is a complete Visual Prolog program that functions as a mini telephone directorytelephone directory. The domains section is not needed here, since only standard domains are used. /* Program ch047e04.pro */ PREDICATES nondeterm phone_number(symbol,symbol) CLAUSES phone_number("Albert","EZY-3665"). phone_number("Betty","555-5233"). phone_number("Carol","909-1010"). phone_number("Dorothy","438-8400").

Load and run the program 4, then try each of these goals in turn: a . phone_number("Carol", Number). b . phone_number(Who, "438-8400"). c . phone_number("Albert", Number). d . phone_number(Who, Number).

Now update the clauses. Suppose that Kim shares a condo with Dorothy and so has the same phone number. Add this fact to the clauses section and try the goal phone_number(Who, "438-8400").

You should get two solutions to this query: Who=Dorothy Who=Kim 2 Solutions

To illustrate the char domain, program 5 defines isletter, which, when given the goals isletter('%'). isletter('Q').

will return No and Yes, respectively.

/* Program ch048e05.pro */

PREDICATES nondeterm isletter(char) CLAUSES /* When applied to characters, '<=' means */ /* "alphabetically precedes or is the same as" */ isletter(Ch):'a' <= Ch, Ch <= 'z'. isletter(Ch):'A' <= Ch, Ch <= 'Z'.

Load and run Program 5 and try each of these goals in turn: a . isletter('x'). b . isletter('2'). c . isletter("hello"). d . isletter(a). e . isletter(X).

Goals (c) and (d) will result in a type error message, and (e) will return an Free message, because you can't test whether an unidentified object follows a or precedes z. variable

Multiple Arity The arity of a predicate is the number of arguments that it takes. You can have two predicates with the same name but different arity. You must group different arity versions of a given predicate name together in both the predicates and clauses sections of your program; apart from this restriction, the different arities are treated as completely different predicates. /* Program ch049e06.pro */ DOMAINS person = symbol PREDICATES father(person) father(person, person)

% This person is a father % One person is the father of the other person

CLAUSES father(Man):father(Man,_). father(adam,seth). father(abraham,isaac).

Rule Syntax Rules are used in Prolog when a fact depends upon the success (truth) of another fact or group of facts. As we explained in Chapter 50, a Prolog rule has two parts: the head and the body. This is the generic syntax for a Visual Prolog rule: HEAD :- <Subgoal>, <Subgoal>, ..., <Subgoal>.

The body of the rule consists of one or more subgoals. Subgoals are separated by commas, specifying conjunction, and the last subgoal in a rule is terminated by a period. Each subgoal is a call to another Prolog predicate, which may succeed or fail. In effect, calling another predicate amounts to evaluating its subgoals, and, depending on their success or failure, the call will succeed or fail. If the current subgoal can be satisfied (proven true), the call returns, and processing continues on to the next subgoal. Once the final subgoal in a rule succeeds, the call returns successfully; if any of the subgoals fail, the rule immediately fails. To use a rule successfully, Prolog must satisfy all of the subgoals in it, creating a consistent set of variable bindings as it does so. If one subgoal fails, Prolog will back up and look for alternatives to earlier subgoals, then proceed forward again with different variable values. This is called backtracking. A full discussion of backtracking and how Prolog finds solutions is covered in Chapter 51.

lvii

Prolog if Symbol vs. IF in Other Languages As we have mentioned earlier, the :- separating the head and the body of a rule, is read "if". However, a Prolog if differs from the IF written in other languages, such as Pascal. In Pascal, for instance, the condition contained in the IF statement must be met before the body of the statement can be executed; in other words, "if HEAD is true, then BODY is true (or: then do BODY)" This type of statement is known as an if/then conditional. Prolog, on the other hand, uses a different form of logic in its rules. The head of a Prolog rule is concluded to be true if (after) the body of the rule succeeds; in other words, "HEAD is true if BODY is true (or: if BODY can be done)" Seen in this manner, a Prolog rule is in the form of a then/if conditional. Automatic Type Conversions When Visual Prolog matches two variables, it's not always necessary that they belong to the same domain. Also, variables can sometimes be bound to constants from other domains. This (selective) mixing is allowed because Visual Prolog performs automatic type conversion (from one domain to another) in the following circumstances: • Between strings and symbols. • Between all the integral domains, and also real. When a character is

converted to a numeric value, the number is the ASCII value for that character.

An argument from a domain my_dom declared in this form DOMAINS my_dom = <base domain>

/*<base domain> is a standard domain */

can mix freely with arguments from that base domain and all other standard domains that are compatible with that base domain. (If the base domain is string, arguments from the symbol domain are compatible; if the base domain is integer, arguments from the real, char, word, etc., domains are compatible. These type conversions mean, for example, that you can call a predicate that handles strings with a symbol argument, and vice versa call a predicate that handles reals with an integer argument call a predicate that handles characters with integer values

use characters in expressions and comparisons without needing to look up their ASCII values. There are a number of rules deciding what domain the result of the expression belongs to, when different domains are mixed. These will be detailed in chapter 52.

Other Program Sections Now that you're reasonably familiar with the clauses, predicates, domains, and goal sections of a Visual Prolog program, we'll tell you a little bit about some other commonly-used program sections: the facts section, the constants section, and the various global sections. This is just an introduction; as you work through the rest of the tutorials in this book you'll learn more about these sections and how to use them in your programs.

The Facts Section A Visual Prolog program is a collection of facts and rules. Sometimes, while the program is running, you might want to update (change, remove, or add) some of the facts the program operates on. In such a case, the facts constitute a dynamic or internal database; it can change while the program is running. Visual Prolog includes a special section for declaring the facts in the program that are to be a part of the dynamic (or changing) database; this is the facts section. The keyword facts declares the facts section. It is here that you declare the facts to be included in the dynamic facts section. Visual Prolog includes a number of built-in predicates that allow easy use of the dynamic facts section. The keyword facts is synonymous with database. Chapter 53 provides a complete discussion of the facts section and the predicates used along with it.

The Constants Section You can declare and use symbolic constants in your Visual Prolog programs. A constant declaration section is indicated by the keyword constants, followed by the declarations themselves, using the following syntax: <Id> =

<Id> is the name of your symbolic constant, and <Macro definition> is what you're assigning to that constant. Each <Macro definition> is terminated by a

lix

newline character, so there can only be one constant declaration per line. Constants declared in this way can then be referred to later in the program. Consider the following program fragment: CONSTANTS zero = 0 one = 1 two = 2 hundred = (10*(10-1)+10) pi = 3.141592653 ega = 3 slash_fill = 4 red = 4

Before compiling your program, Visual Prolog will replace each constant with the actual string to which it corresponds. For instance: ..., A = hundred*34, delay(A), setfillstyle(slash_fill, red), Circumf = pi*Diam, ...

will be handled by the compiler in exactly the same way as ..., A = (10*(10-1)+10)*34, delay(A), setfillstyle(4, 4), Circumf = 3.141592653*Diam, ...

There are a few restrictions on the use of symbolic constants: The definition of a constant can't refer to itself. For example: my_number = 2*my_number/2 /* Is not allowed

will generate the error message Recursion in constant definition. The system does not distinguish between upper-case and lower-case in a constants declaration. Consequently, when a constants identifier is used in the clauses section of a program, the first letter must be lower-case to avoid confusing constants with variables. So, for example, the following is a valid construction: CONSTANTS Two = 2

GOAL A=two, write(A).

There can be several constants declaration sections in a program, but constants must be declared before they are used. Declared constants are effective from their point of declaration to the end of the source file, and in any files included after the declaration. Constant identifiers can only be declared once. Multiple declarations of the same identifier will result in the error message This constant is already defined.

The Global Sections Visual Prolog allows you to declare some domains, predicates, and clauses in your program to be global (rather than local); you do this by setting aside separate global domains, global predicates, and global facts sections at the top of your program. These global sections are discussed in the chapter 54.

The Compiler Directives Visual Prolog provides several compiler directives you can add to your program to tell the compiler to treat your code in specified ways when compiling. You can also set most of the compiler directives from the Options | Project | Compiler Options menu item in the Visual Prolog system. Compiler directives are covered in detail in the chapter 55, but you'll want to know how to use a couple of them before you get to that chapter, so we introduce the basic ones here. The include Directive As you get more familiar with using Visual Prolog, you'll probably find that you use certain procedures over and over again in your programs. You can use the include directive to save yourself from having to type those procedures in again and again. Here's an example of how you could use it: You create a file (such as MYSTUFF.PRO) in which you declare your frequently-used predicates (using domains and predicates sections) and give the procedures defining those predicates in a clauses section. You write the source text for the program that will make use of these procedures. At a natural boundary in your source text, you place the line

lxi

include "mystuff.pro"

(A natural boundary is anywhere in your program that you can place a domains, facts, predicates, clauses, or goal section.) When you compile your source text, Visual Prolog will compile the contents of MYSTUFF.PRO right into the final compiled product of your source text. You can use the include directive to include practically any often-used text into your source text, and one included file can in turn include another (but a given file can only be included once in your program). The include directive can appear at any natural boundary in your source text. However, you must observe the restrictions on program structure when you include a file into your source text.

Summary These are the ideas we've introduced in this chapter: A Visual Prolog program has the following basic structure: DOMAINS /* ... domain declarations ... */ PREDICATES /* ... predicate declarations ... */ CLAUSES /* ... clauses (rules and facts) ... */ GOAL /* ... subgoal_1, subgoal_2, etc. */

The clauses section is where you put the facts and rules that Visual Prolog will operate on when trying to satisfy the program's goal. The predicates section is where you declare your predicates and the domains (types) of the arguments to your predicates. Predicate names must begin with a letter (preferably lower-case), followed by a sequence of

letters, digits, and underscores, up to 250 characters long. You can't use spaces, the minus sign, asterisks, or slashes in predicate names. Predicate declarations are of the form PREDICATES predicateName(argument_type1, argument_type2, ..., argument_typeN)

argument_type1, ..., argument_typeN are either standard domains or domains that you've declared in the domains section. Declaring the domain of an argument and defining the argument's type are the same thing. The domains section is where you declare any nonstandard domains you're using for the arguments to your predicates. Domains in Prolog are like types in other languages. Visual Prolog's basic standard domains are char, byte, short, ushort, word, integer, unsigned, long, ulong, dword, real, string, and symbol; the more specialized standard domains are covered in other chapters. The basic domain declarations are of the form DOMAINS argument_type1, ..., argument_typeN = <standard domain> argument_1, ..., argument_N) = <compound domain 1>; <compound domain 2>; < ... >; <compound domain N>;

Compound domains haven't been covered in this chapter; you'll see them in Chapter 56. The goal section is where you put your program's internal goal; this allows the program to run independent of the development environment. With an internal goal, Visual Prolog only searches for the first solution, and the values to which variables are bound are not displayed. If you don't use an internal goal, you'll enter an external goal in the Dialog window at run time. With an external goal, Visual Prolog searches for all solutions, and displays the values to which variables are bound. The arity of a predicate is the number of arguments that it takes; two predicates can have the same name but different arity. You must group a predicate's different arity versions together in both the predicates and clauses sections, but different arities are treated as completely different predicates. Rules are of the form HEAD :- <Subgoal1>, <Subgoal2>, ..., <SubgoalN>.

lxiii

For a rule to succeed, Prolog must satisfy all of its subgoals, creating a consistent set of variable bindings. If one subgoal fails, Prolog backs up and looks for alternatives to earlier subgoals, then proceeds forward with different variable values. This is called backtracking. The :- ("if") in Prolog should not be confused with the IF used in other languages; a Prolog rule is in the form of a then/if conditional, while IF statements in other languages are in the form of an if/then conditional.

CHAPTER

Unification and Backtracking 58This chapter is divided into four main parts. In the first part, we examine in detail the process Visual Prolog uses when trying to match a call (from a subgoal) with a clause (in the clauses section of the program). This search process includes a procedure known as unification, which attempts to match up the data-structures embodied in the call with those found in a given clause. In Prolog, unification implements several of the procedures you might know from other, more traditional languages--procedures such as parameter passing, case selection, structure building, structure access, and assignment. In the second part, we show you how Visual Prolog searches for solutions to a goal (through backtracking) and how to control a search. This includes techniques that make it possible for a program to carry out a task that would otherwise be impossible, either because the search would take too long (which is less likely with Visual Prolog than with other Prologs) or because the system would run out of free memory. In the third part of this chapter, we introduce a predicate you can use to encourage backtracking, and go into more detail about how you can control backtracking. We also introduce a predicate you can use to verify that a certain constraint in your program is (or is not) met. To shed more light on the subject, in the fourth part of this chapter we review the more important tutorial material (presented so far) from a procedural perspective. We show how you can understand the basic aspects of Prolog, a declarative language, by also looking at them as procedures.

Matching Things Up: Unification Consider Program 1 in terms of the external goal written_by(X, Y).

When Visual Prolog tries to fulfill the goal written_by(X, Y)., it must test each written_by clause in the program for a match. In the attempt to match the arguments X and Y with the arguments found in each written_by clause, Visual Prolog will search from the top of the program to the bottom. When it finds a clause that matches the goal, it binds values to free variables so that the goal and the clause are identical; the goal is said to unify with the clause. This matching operation is called unification. /* Program ch059e01.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS title,author = symbol pages = unsigned PREDICATES book(title, pages) nondeterm written_by(author, title) nondeterm long_novel(title) CLAUSES written_by(fleming, "DR NO"). written_by(melville, "MOBY DICK"). book("MOBY DICK", 250). book("DR NO", 310). long_novel(Title):written_by(_, Title), book(Title, Length), Length > 300.

Since X and Y are free variables in the goal, and a free variable can be unified with any other argument (even another free variable), the call (goal) can be unified with the first written_by clause in the program, as shown here: written_by(

X , Y ). | | written_by(fleming, "DR NO").

lxv

Visual Prolog makes a match, X becomes bound to fleming, and Y becomes bound to "DR NO." At this point, Visual Prolog displays X=fleming, Y=DR NO

Since Visual Prolog looks for all solutions when you use an external goal, the goal is also unified with the second written_by clause written_by(melville,"MOBY DICK").

and Visual Prolog displays the second solution: X=melville, Y=MOBY DICK 2 Solutions

If, on the other hand, you give the program the goal written_by(X, "MOBY DICK").

Visual Prolog will attempt a match with the first clause for written_by: written_by(

X ,"MOBY DICK"). | | written_by(fleming,"DR NO").

Since "MOBY DICK" and "DR NO" do not match, the attempt at unification fails. Visual Prolog then tries the next fact in the program: written_by(melville, "MOBY DICK").

This does unify, and X becomes bound to melville. Consider how Visual Prolog executes the following: long_novel(X).

When Visual Prolog tries to fulfill a goal, it investigates whether or not the call can match a fact or the head of a rule. In this case, the match is with long_novel(Title)

Visual Prolog looks at the clause for long_novel, trying to complete the match by unifying the arguments. Since X is not bound in the goal, the free variable X can be unified with any other argument. Title is also unbound in the head of the long_novel clause. The goal matches the head of the rule and unification is made. Visual Prolog will subsequently attempt to satisfy the subgoals to the rule.

long_novel(Title):written_by(_, Title), book(Title, Length), Length>300.

In attempting to satisfy the body of the rule, Visual Prolog will call the first subgoal in the body of the rule, written_by(_, Title). Notice that, since who authored the book is immaterial, the anonymous variable (_) appears in the position of the author argument. The call written_by(_, Title) becomes the current subgoal, and Prolog searches for a solution to this call. Prolog searches for a match with this subgoal from the top of the program to the bottom. In doing so, it achieves unification with the first fact for written_by as follows: written_by(_, Title), | | written_by(fleming,"DR NO").

The variable Title becomes bound to "DR NO" and the next subgoal, book(Title, Length), is called with this binding. Visual Prolog now begins its next search, trying to find a match with the call to book. Since Title is bound to "DR NO", the actual call resembles book("DR NO", Length). Again, the search starts from the top of the program. Notice that the first attempt to match with the clause book("MOBY DICK", 250) will fail, and Visual Prolog will go on to the second clause of book in search of a match. Here, the book title matches the subgoal and Visual Prolog binds the variable Length with the value 310. The third clause in the body of long_novel now becomes the current subgoal: Length > 300.

Visual Prolog makes the comparison and succeeds; 310 is greater than 300. At this point, all the subgoals in the body of the rule have succeeded and therefore the call long_novel(X) succeeds. Since the X in the call was unified with the variable Title in the rule, the value to which Title is bound when the rule succeeds is returned to the call and unified with the variable X. Title has the value "DR NO" when the rule succeeds, so Visual Prolog will output: X=DR NO 1 Solution

In the following chapters, we will show several advanced examples of unification. However, there are still a few basics that need to be introduced first, such as

lxvii

complex structures. In the next section of this chapter, we'll discuss how Prolog searches for its solutions.

Backtracking Often, when solving real problems, you must pursue a path to its logical conclusion. If this conclusion does not give the answer you were looking for, you must choose an alternate path. For instance, you might have played maze games when you were a child. One sure way to find the end of the maze was to turn left at every fork in the maze until you hit a dead end. At that point you would back up to the last fork, and try the right-hand path, once again turning left at each branch encountered. By methodically trying each alternate path, you would eventually find the right path and win the game. Visual Prolog uses this same backing-up-and-trying-again method, called backtracking, to find a solution to a given problem. As Visual Prolog begins to look for a solution to a problem (or goal), it might have to decide between two possible cases. It sets a marker at the branching spot (known as a backtracking point) and selects the first subgoal to pursue. If that subgoal fails (equivalent to reaching a dead end), Visual Prolog will backtrack to the back-tracking point and try an alternate subgoal. Here is a simple example: /* Program ch060e02.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ PREDICATES nondeterm likes(symbol,symbol) tastes(symbol,symbol) nondeterm food(symbol) CLAUSES likes(bill,X):food(X), tastes(X,good). tastes(pizza,good). tastes(brussels_sprouts,bad). food(brussels_sprouts). food(pizza).

This small program is made up of two sets of facts and one rule. The rule, represented by the relationship likes, simply states that Bill likes good-tasting food. To see how backtracking works, give the program the following goal to solve: likes(bill, What).

When Prolog begins an attempt to satisfy a goal, it starts at the top of the program in search of a match.

In this case, it will begin the search for a solution by looking from the top for a match to the subgoal likes(bill, What). It finds a match with the first clause in the program, and the variable What is unified with the variable X. Matching with the head of the rule causes Visual Prolog to attempt to satisfy that rule. In doing so, it moves on to the body of the rule, and calls the first subgoal located there: food(X). When a new call is made, a search for a match to that call also begins at the top of the program.

In the search to satisfy the first subgoal, Visual Prolog starts at the top, attempting a match with each fact or head of a rule encountered as processing goes down into the program. It finds a match with the call at the first fact representing the food relationship. Here, the variable X is bound to the value brussels_sprouts. Since there is more than one possible answer to the call food(X), Visual Prolog sets a backtracking point next to the fact food(brussels_sprouts). This backtracking point keeps track of where Prolog will start searching for the next possible match for food(X). When a call has found a successful match, the call is said to succeed, and the next subgoal in turn may be tried.

With X bound to brussels_sprouts, the next call made is

lxix

tastes(brussels_sprouts, good)

and Visual Prolog begins a search to attempt to satisfy this call, again starting from the top of the program. Since no clause is found to match, the call fails and Visual Prolog kicks in its automatic backtracking mechanism. When backtracking begins, Prolog retreats to the last backtracking point set. In this case, Prolog returns to the fact food(brussels_sprouts). Once a variable has been bound in a clause, the only way to free that binding is through backtracking.

When Prolog retreats to a backtracking point, it frees all the variables set after that point, and sets out to find another solution to the original call. The call was food(X), so the binding of brussels_sprouts for X is released. Prolog now tries to resolve this call, beginning from the place where it left off. It finds a match with the fact food(pizza)] and returns, this time with the variable X bound to the value pizza. Prolog now moves on to the next subgoal in the rule, with the new variable binding. A new call is made, tastes(pizza, good)], and the search begins at the top of the program. This time, a match is found and the goal returns successfully. Since the variable What in the goal is unified with the variable X in the likes rule, and the variable X is bound to the value pizza, the variable What is now bound to the value pizza and Visual Prolog reports the solution What=pizza 1 Solution

Visual Prolog's Relentless Search for Solutions As we've described earlier, with the aid of backtracking, Visual Prolog will not only find the first solution to a problem, but is actually capable of finding all possible solutions. Consider Program 3, which contains facts about the names and ages of some players in a racquet club. /* Program ch061e03.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */

DOMAINS child = symbol age = integer PREDICATES nondeterm player(child, age) CLAUSES player(peter,9). player(paul,10). player(chris,9). player(susan,9).

You'll use Visual Prolog to arrange a ping-pong tournament between the nineyear-olds in a racquet club. There will be two games for each pair of club players. Your aim is to find all possible pairs of club players who are nine years old. This can be achieved with the compound goal: player(Person1, 9), player(Person2, 9), Person1 <> Person2.

In natural language: Find Person1 (age 9) and Person2 (age 9) so that Person1 is different from Person2. Visual Prolog will try to find a solution to the first subgoal player(Person1, 9) and continue to the next subgoal only after the first subgoal is reached. The first subgoal is satisfied by matching Person1 with peter. Now Visual Prolog can attempt to satisfy the next subgoal: player(Person2, 9)

by also matching Person2 with peter. Now Prolog comes to the third and final subgoal Person1 <> Person2

Since Person1 and Person2 are both bound to peter, this subgoal fails. Because of this, Visual Prolog backtracks to the previous subgoal, and searches for another solution to the second subgoal: player(Person2, 9)

This subgoal is fulfilled by matching Person2 with chris. Now, the third subgoal: Person1 <> Person2

can succeed, since peter and chris are different. Here, the entire goal is satisfied by creating a tournament between the two players, chris and peter. lxxi

However, since Visual Prolog must find all possible solutions to a goal, it backtracks to the previous goal--hoping to succeed again. Since player(Person2, 9)

can also be satisfied by taking Person2 to be susan, Visual Prolog tries the third subgoal once again. It succeeds (since peter and susan are different), so another solution to the entire goal has been found. Searching for more solutions, Visual Prolog once again backtracks to the second subgoal, but all possibilities for this subgoal have been exhausted. Because of this, backtracking now continues back to the first subgoal. This can be satisfied again by matching Person1 with chris. The second subgoal now succeeds by matching Person2 with peter, so the third subgoal is satisfied, again fulfilling the entire goal. Here, another tournament has been scheduled, this time between chris and peter. Searching for yet another solution to the goal, Visual Prolog backtracks to the second subgoal in the rule. Here, Person2 is matched to chris and again the third subgoal is tried with these bindings. The third subgoal fails, since Person1 and Person2 are equal, so backtracking regresses to the second subgoal in search of another solution. Person2 is now matched with susan, and the third subgoal succeeds, providing another tournament for the racket club (chris vs. susan). Once again, searching for all solutions, Prolog backtracks to the second subgoal, but this time without success. When the second subgoal fails, backtracking goes back to the first subgoal, this time finding a match for Person1 with susan. In an attempt to fulfill the second subgoal, Prolog matches Person2 with peter, and subsequently the third subgoal succeeds with these bindings. A fifth tournament has been scheduled for the players. Backtracking again goes to the second subgoal, where Person2 is matched with chris. A sixth solution is found for the racquet club, producing a full set of tournaments. The final solution tried is with both Person1 and Person2 bound to susan. Since this causes the final subgoal to fail, Visual Prolog must backtrack to the second subgoal, but there are no new possibilities. Visual Prolog then backtracks to the first subgoal, but the possibilities for Person1 have also been exhausted. No more solutions can be found for the goal, so the program terminates. Type in this compound goal for the program:

player(Person1, 9), player(Person2, 9), Person1 <> Person2.

Verify that Visual Prolog responds with Person1=peter, Person1=peter, Person1=chris, Person1=chris, Person1=susan, Person1=susan, 6 Solutions

Person2=chris Person2=susan Person2=peter Person2=susan Person2=peter Person2=chris

Notice how backtracking might cause Visual Prolog to come up with redundant solutions. In this example, Visual Prolog does not distinguish that Person1 = peter is the same thing as Person2 = peter. We will show you later in this chapter how to control the search Visual Prolog generates. Exercise in Backtracking Using Program 3, decide what Visual Prolog will reply to the following goal: player(Person1, 9), player(Person2, 10).

Check your answer by typing in the exercise and the given goal when you run the program. A Detailed Look at Backtracking With this simple example under your belt, you can take a more detailed look at how Visual Prolog's backtracking mechanism works. Start by looking at Program 4 in light of the following goal, which consists of two subgoals: likes(X, wine) , likes(X, books)

When evaluating the goal, Visual Prolog notes which subgoals have been satisfied and which have not. This search can be represented by a goal tree:

likes (X, wine)

likes (X, books)

lxxiii

Before the goal evaluation begins, the goal tree consists of two unsatisfied subgoals. In the following goal tree diagrams, a subgoal satisfied in the goal tree is marked with an underline, and the corresponding clause is shown beneath that subgoal. /* Program ch062e04.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS name,thing = symbol PREDICATES likes(name, thing) reads(name) is_inquisitive(name) CLAUSES likes(john,wine):-!. likes(lance,skiing):-!. likes(lance,books):-!. likes(lance,films):-!. likes(Z,books):reads(Z), is_inquisitive(Z). reads(john). is_inquisitive(john).

The Four Basic Principles of Backtracking In this example, the goal tree shows that two subgoals must be satisfied. To do so, Visual Prolog follows the first basic principle of backtracking:

Subgoals must be satisfied in order, from top to bottom.

Visual Prolog determines which subgoal it will use when trying to satisfy the clause according to the second basic principle of backtracking:

Predicate clauses are tested in the order they appear in the program, from top to bottom.

When executing Program 4, Visual Prolog finds a matching clause with the first fact defining the likes predicate. Take a look at the goal tree now.

likes (X, wine)

likes (X, books)

likes (john, wine)

The subgoal likes(X, wine) matches the fact likes(john, wine) and binds X to the value john. Visual Prolog tries to satisfy the next subgoal to the right. The call to the second subgoal begins a completely new search with the binding X = john. The first clause likes(john, wine)

does not match the subgoal likes(X, books)

since wine is not the same as books. Visual Prolog must therefore try the next clause, but lance does not match the value X (because, in this case, X is bound to john), so the search continues to the third clause defining the predicate likes: likes(Z, books):- reads(Z), is_inquisitive(Z).

The argument Z is a variable, so it is able to match with X. The second arguments agree, so the call matches the head of the rule. When X matches Z, the arguments are unified. With the arguments unified, Visual Prolog will equate the value X has (which is john) with the variable Z. Because of this, now the variable Z also has the value john. The subgoal now matches the left side (head) of a rule. Continued searching is determined by the third basic principle of backtracking:

lxxv

When a subgoal matches the head of a rule, the body of that rule must be satisfied next. The body of the rule then constitutes a new set of subgoals to be satisfied.

This yields the following goal tree:

likes (X, wine)

likes (X, books)

likes (john, wine)

likes ( Z, books)

reads (Z)

is_inquisitive (Z)

The goal tree now includes the subgoals reads(Z) and is_inquisitive(Z)

where Z is bound to the value john. Visual Prolog will now search for facts that match both subgoals. This is the resulting final goal tree:

likes (X, wine)

likes (X, books)

likes (john, wine)

likes ( Z, books)

reads (Z)

is_inquisitive (Z)

reads (john)

is_inquisitive (john)

According to the fourth basic principle of backtracking:

A goal has been satisfied when a matching fact is found for each of the extremities (leaves) of the goal tree.

So now the initial goal is satisfied. Visual Prolog uses the result of the search procedure in different ways, depending on how the search was initiated. If the goal is a call from a subgoal in the body of a rule, Visual Prolog attempts to satisfy the next subgoal in the rule after the call has returned. If the goal is a query from the user, Visual Prolog replies directly: X=john 1 Solution

As you saw in Program 4, having once satisfied an external goal, Visual Prolog backtracks to find all alternate solutions. It also backtracks if a subgoal fails, hoping to re-satisfy a previous subgoal in such a way that the failed subgoal is satisfied by other clauses. To fulfill a subgoal, Visual Prolog begins a search with the first clause that defines the predicate. One of two things can then happen: It finds a matching clause, in which case the following occurs: If there is another clause that can possibly re-satisfy the subgoal, Visual Prolog places a pointer (to indicate a backtracking point) next to the matching clause. All free variables in the subgoal that match values in the clause are bound to the corresponding values. If the matching clause is the head of a rule, that rule's body is then evaluated; the body's subgoals must succeed for the call to succeed. It can't find a matching clause, so the goal fails. Visual Prolog backtracks as it attempts to re-satisfy a previous subgoal. When processing reaches the last backtracking point, Visual Prolog frees all variables that had been assigned new values since the backtracking point was set, then attempts to re-satisfy the original call. Visual Prolog begins a search from the top of the program. When it backtracks to a call, the new search begins from the last backtracking point set. If the search is unsuccessful, it backtracks again. If backtracking exhausts all clauses for all subgoals, the goal fails. lxxvii

Backtracking Here is another, slightly more complex, example, illustrating how backtracking takes place in Prolog. /* Program ch063e05.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ PREDICATES nondeterm type(symbol, symbol) nondeterm is_a(symbol, symbol) lives(symbol, symbol) nondeterm can_swim(symbol) CLAUSES type(ungulate,animal). type(fish,animal). is_a(zebra,ungulate). is_a(herring,fish). is_a(shark,fish). lives(zebra,on_land). lives(frog,on_land). lives(frog,in_water). lives(shark,in_water). can_swim(Y):type(X,animal), is_a(Y,X), lives(Y,in_water).

This example program uses an internal goal to illustrate how backtracking works. When the program is compiled and run, Visual Prolog will automatically begin executing the goal, attempting to satisfy all the subgoals in the goal section. Visual Prolog calls the can_swim predicate with a free variable, What. In trying to solve this call, Visual Prolog searches the program looking for a match. It finds a match with the clause defining can_swim, and the variable What is unified with the variable Y. Next, Visual Prolog attempts to satisfy the body of the rule. In doing so, Visual Prolog calls the first subgoal in the body of the rule, type(X, animal), and searches for a match to this call. It finds a match with the first fact defining the type relationship. At this point, X is bound to ungulate. Since there is more than one possible solution, Visual Prolog sets a backtracking point at the fact type(ungulate, animal).

With X bound to ungulate, Visual Prolog makes a call to the second subgoal in the rule (is_a(Y, ungulate)), and again searches for a match. It finds one with the first fact, is_a(zebra, ungulate). Y is bound to zebra and Prolog sets a backtracking point at is_a(zebra, ungulate). Now, with X bound to ungulate and Y bound to zebra, Prolog tries to satisfy the last subgoal, lives(zebra, in_water). Prolog tries each lives clause, but there is no lives(zebra, in_water) clause in the program, so the call fails and Prolog begins to backtrack in search of another solution. When Visual Prolog backtracks, processing returns to the last point where a backtracking point was placed. In this case, the last backtracking point was placed at the second subgoal in the rule, on the fact is_a(zebra, ungulate). When Visual Prolog reaches a backtracking point, it frees the variables that were assigned new values after the last backtracking point and attempts to find another solution to the call it made at that time. In this case, the call was is_a(Y, ungulate). Visual Prolog continues down into the clauses in search of another clause that will match with this one, starting from the point where it previously left off. Since there are no other clauses in the program that can match this one, the call fails and Visual Prolog backtracks again in an attempt to solve the original goal. From this position, the last backtracking point was set at animal).

type(ungulate,

Visual Prolog frees the variables set in the original call and tries to find another solution to the call type(X, animal). The search begins after the backtracking point. Visual Prolog finds a match with the next type fact in the program (type(fish, animal)); X is bound to fish, and a new backtracking point is set at that fact. Visual Prolog now moves down to the next subgoal in the rule; since this is a new call, the search begins at the top of the program with is_a(Y, fish). Visual Prolog finds a match to this call and Y is bound to herring. Since Y is now bound to herring, the next subgoal called is lives(herring, in_water). Again, this is a new call, and the search begins from the top of the program. Visual Prolog tries each lives fact, but fails to find a match and the subgoal fails. Visual Prolog now returns to the last backtracking point, fish).

lxxix

is_a(herring,

The variables that were bound by this matching are now freed. Starting at the point where it last left off, Visual Prolog now searches for a new solution to the call is_a(Y, fish). Visual Prolog finds a match with the next is_a clause, and Y becomes bound to the symbol shark. Visual Prolog tries the last subgoal again, with the variable Y bound to shark. It calls lives(shark, in_water); the search begins at the top of the program, since this is a new call. It finds a match and the last subgoal to the rule succeeds. At this point, the body of the can_swim(Y) rule is satisfied. Visual Prolog returns Y to the call can_swim(What). Since What is bound to Y, and Y is bound to shark, What is now bound to shark in the goal. Visual Prolog continues processing where it left off in the goal section, and calls the second subgoal in the goal. Visual Prolog completes the program by outputting A shark can swim.

and the program terminates successfully.

RULE:

What is

can_swim(What):-

unified with Y

type(X,animal), is_a(What,X), lives(What,in_water). CALL: MATCH:

CALL: MATCH

No match

lives(zebra,in_water) lives(zebra,in_water)

REDO: FAIL:

REDO: MATCH:

Y is bound to zebra

is_a(Y,ungulate) is_a(zebra,ungulate)

CALL: FAIL:

X is bound to ungulate

type(X,animal) type(ungulate,animal)

No more facts that match this call

is_a(Y,ungulate) is_a(Y,ungulate)

X is now bound to fish

type(X,animal) type(fish,animal)

CALL: MATCH:

CALL: FAIL:

REDO: MATCH:

CALL: MATCH:

is_a(Y, fish) is_a(herring, fish)

lives(herring,in_water) lives(herring,in_water)

is_a(Y, fish) is_a(shark, fish)

lives(shark,in_water) lives(shark,in_water)

Y is now bound to herring

No match

Y is now bound to shark

What is bound to shark

Figure 64.1: How the can_swim Program Works65

Controlling the Search for Solutions Prolog's built-in backtracking mechanism can result in unnecessary searching; because of this, inefficiencies can arise. For instance, there may be times when you want to find unique solutions to a given question. In other cases, it may be necessary to force Visual Prolog to continue looking for additional solutions even though a particular goal has been satisfied. In cases such as these, you must

lxxxi

control the backtracking process. In this section, we'll show you some techniques you can use to control Visual Prolog's search for the solutions to your goals. Visual Prolog provides two tools that allow you to control the backtracking mechanism: the fail predicate, which is used to force backtracking, and the cut (signified by !), which is used to prevent backtracking.

Using the fail Predicate Visual Prolog begins backtracking when a call fails. In certain situations, it's necessary to force backtracking in order to find alternate solutions. Visual Prolog provides a special predicate, fail, to force failure and thereby encourage backtracking. The effect of the fail predicate corresponds to the effect of the comparison 2 = 3 or any other impossible subgoal. Program 6 illustrates the use of this special predicate. /* Program ch066e06.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS name = symbol PREDICATES nondeterm father(name, name) everybody CLAUSES father(leonard,katherine). father(carl,jason). father(carl,marilyn). everybody:father(X,Y), write(X," is ",Y,"'s father\n"), fail. everybody.

Once an internal goal has completely succeeded, there is nothing that tells Visual Prolog to backtrack. Because of this, an internal call to father will come up with only one solution. However, the predicate everybody in Program 6 uses fail to force backtracking, and therefore finds all possible solutions. The object of the predicate everybody is to produce a cleaner response from program runs. Compare the answers to the two preceding goals: Goal father(X, Y).

X=leonard, Y=katherine X=carl, Y=jason X=carl, Y=marilyn 3 Solutions

and Goal everybody. leonard is katherine's father carl is jason's father carl is marilyn's father yes

The predicate everybody uses backtracking to generate more solutions for father(X, Y) by forcing Prolog to backtrack through the body of the everybody rule: father(X, Y), write(X," is ",Y,"'s father\n"), fail.

fail can never be satisfied (it always fails), so Visual Prolog is forced to backtrack. When backtracking takes place, Prolog backtracks to the last call that can produce multiple solutions. Such a call is labeled non-deterministic. A nondeterministic call contrasts with a call that can produce only one solution, which is a deterministic call. The write predicate can't be re-satisfied (it can't offer new solutions), so Visual Prolog must backtrack again, this time to the first subgoal in the rule. Notice that it's useless to place a subgoal after fail in the body of a rule. Since the predicate fail always fails, there would be no way of reaching a subgoal located after fail. Exercises Load and run Program 6 and evaluate the following goals: a.

father(X, Y).

everybody.

2. Edit the body of the rule defining everybody so that the rule ends with the call to the write predicate (delete the call to fail). Now compile and run the program, giving everybody. as the goal. Why doesn't Visual Prolog find all the solutions as it does with the query father(X, Y)?

lxxxiii

3. Relocate the call to fail at the end of the everybody rule. Again, give the query everybody as the goal. Why are the solutions to everybody terminated by no? For a clue, append everybody. as a second clause to the definition of predicate everybody and re-evaluate the goal.

Preventing Backtracking: The Cut Visual Prolog contains the cut, which is used to prevent backtracking; it's written as an exclamation mark (!). The effect of the cut is simple: It is impossible to backtrack across a cut. You place the cut in your program the same way you place a subgoal in the body of a rule. When processing comes across the cut, the call to cut immediately succeeds, and the next subgoal (if there is one) is called. Once a cut has been passed, it is not possible to backtrack to subgoals placed before the cut in the clause being processed, and it is not possible to backtrack to other predicates defining the predicate currently in process (the predicate containing the cut). There are two main uses of the cut: When you know in advance that certain possibilities will never give rise to meaningful solutions, it's a waste of time and storage space to look for alternate solutions. If you use a cut in this situation, your resulting program will run quicker and use less memory. This is called a green cut. When the logic of a program demands the cut, to prevent consideration of alternate subgoals. This is a red cut. How to Use the Cut In this section, we give examples that show how you can use the cut in your programs. In these examples, we use several schematic Visual Prolog rules (r1, r2, and r3), which all describe the same predicate r, plus several subgoals (a, b, c, etc.). Prevent Backtracking to a Previous Subgoal in a Rule r1 :- a, b, !, c.

This is a way of telling Visual Prolog that you are satisfied with the first solution it finds to the subgoals a and b. Although Visual Prolog is able to find multiple solutions to the call to c through backtracking, it is not allowed to backtrack across the cut to find an alternate solution to the calls a or b. It is also not allowed to backtrack to another clause that defines the predicate r1. As a concrete example, consider Program 7.

/* Program ch067e07.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ PREDICATES buy_car(symbol,symbol) nondeterm car(symbol,symbol,integer) colors(symbol,symbol) CLAUSES buy_car(Model,Color):car(Model,Color,Price), colors(Color,sexy),!, Price > 25000. car(maserati,green,25000). car(corvette,black,24000). car(corvette,red,26000). car(porsche,red,24000). colors(red,sexy). colors(black,mean). colors(green,preppy).

In this example, the goal is to find a Corvette with a sexy color and a price that's ostensibly affordable. The cut in the buy_car rule means that, since there is only one Corvette with a sexy color in the database, if its price is too high there's no need to search for another car. Given the goal buy_car(corvette, Y)

Visual Prolog calls car, the first subgoal to the buy_car predicate. It makes a test on the first car, the Maserati, which fails. It then tests the next car clauses and finds a match, binding the variable Color with the value black. It proceeds to the next call and tests to see whether the car chosen has a sexy color. Black is not a sexy color in the program, so the test fails. Visual Prolog backtracks to the call to car and once again looks for a Corvette to meet the criteria. It finds a match and again tests the color. This time the color is sexy, and Visual Prolog proceeds to the next subgoal in the rule: the cut. The cut immediately succeeds and effectively "freezes into place" the variable bindings previously made in this clause. lxxxv

Visual Prolog now proceeds to the next (and final) subgoal in the rule: the comparison Price < 25000.

This test fails, and Visual Prolog attempts to backtrack in order to find another car to test. Since the cut prevents backtracking, there is no other way to solve the final subgoal, and the goal terminates in failure. Prevent Backtracking to the Next Clause The cut can be used as a way to tell Visual Prolog that it has chosen the correct clause for a particular predicate. For example, consider the following code: r(1):r(2):r(3):r(_):-

! , a , b , c. ! , d. ! , c. write("This is a catchall clause.").

Using the cut makes the predicate r deterministic. Here, Visual Prolog calls r with a single integer argument. Assume that the call is r(1). Visual Prolog searches the program, looking for a match to the call; it finds one with the first clause defining r. Since there is more than one possible solution to the call, Visual Prolog places a backtracking point next to this clause. Now the rule fires and Visual Prolog begins to process the body of the rule. The first thing that happens is that it passes the cut; doing so eliminates the possibility of backtracking to another r clause. This eliminates backtracking points, increasing the run-time efficiency. It also ensures that the error-trapping clause is executed only if none of the other conditions match the call to r. Note that this type of structure is much like a "case" structure written in other programming languages. Also notice that the test condition is coded into the head of the rules. You could just as easily write the clauses like this: r(X) r(X) r(X) r(_)

::::-

X = 1 , ! , X = 2 , ! , X = 3 , ! , write("This

a , b , c. d. c. is a catchall clause.").

However, you should place the testing condition in the head of the rule as much as possible, as doing this adds efficiency to the program and makes for easier reading. As another example, consider the following program. Run this program and give the query friend(bill, Who) as the goal.

/* Program ch068e08.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ PREDICATES friend(symbol,symbol) girl(symbol) likes(symbol,symbol) CLAUSES friend(bill,jane):girl(jane), likes(bill,jane),!. friend(bill,jim):likes(jim,baseball),!. friend(bill,sue):girl(sue). girl(mary). girl(jane). girl(sue). likes(jim,baseball). likes(bill,sue).

Without cuts in the program, Visual Prolog would come up with two solutions: Bill is a friend of both Jane and Sue. However, the cut in the first clause defining friend tells Visual Prolog that, if this clause is satisfied, it has found a friend of Bill and there's no need to continue searching for more friends. A cut of this type says, in effect, that you are satisfied with the solution found and that there is no reason to continue searching for another friend. Backtracking can take place inside the clauses, in an attempt to satisfy the call, but once a solution is found, Visual Prolog passes a cut. The friend clauses, written as such, will return one and only one friend of Bill's (given that a friend can be found). Determinism and the Cut If the friend predicate (defined in the previous program) were coded without the cuts, it would be a non-deterministic predicate (one capable of generating multiple solutions through backtracking). In many implementations of Prolog, programmers must take special care with non-deterministic clauses because of the attendant demands made on memory resources at run time. However, Visual Prolog makes internal checks for non-deterministic clauses, reducing the burden on you, the programmer.

lxxxvii

However, for debugging (and other) purposes, it can still be necessary for you to intercede; the check_determ compiler directive is provided for this reason. If check_determ is inserted at the very beginning of a program, Visual Prolog will display a warning if it encounters any non-deterministic clauses during compilation. You can make non-deterministic clauses into deterministic clauses by inserting cuts into the body of the rules defining the predicate. For example, placing cuts in the clauses defining the friend predicate causes that predicate to be deterministic because, with the cuts in place, a call to friend can return one, and only one, solution. The not Predicate This program demonstrates how you can use the not predicate to identify an honor student: one whose grade point average (GPA) is at least 3.5 and who is not on probation. /* Program ch069e10.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS name = symbol gpa = real PREDICATES nondeterm honor_student(name) nondeterm student(name, gpa) probation(name) CLAUSES honor_student(Name):student(Name, GPA), GPA>=3.5, not(probation(Name)). student("Betty Blue", 3.5). student("David Smith", 2.0). student("John Johnson", 3.7). probation("Betty Blue"). probation("David Smith").

There is one thing to note when using not: The not predicate succeeds when the subgoal can't be proven true. This results in a situation that prevents unbound variables from being bound within a not. When a subgoal with free variables is called from within not, Visual Prolog will return the error message Free

variables not allowed in 'not' or 'retractall'.

This happens because, for Prolog to bind the free variables in a subgoal, that subgoal must unify with some other clause and the subgoal must succeed. The correct way to handle unbound variables within a not subgoal is with anonymous variables. Here are some examples of correct clauses and incorrect clauses. likes(bill, Anyone):likes(sue, Anyone), not(hates(bill, Anyone).

/* 'Anyone' is an output argument */

In this example, Anyone is bound by likes(sue, Anyone) before Visual Prolog finds out that hates(bill, Anyone) is not true. This clause works just as it should. If you rewrite this so that it calls not first, you will get an error message to the effect that free variables are not allowed in not. likes(bill, Anyone):not(hates(bill, Anyone)), likes(sue, Anyone).

/* This won't work right */

Even if you correct this (by replacing Anyone in not(hates(bill, Anyone)) with an anonymous variable) so that the clause does not return the error, it will still return the wrong result. likes(bill, Anyone):not(hates(bill, _)), likes(sue, Anyone).

/* This won't work right */

This clause states that Bill likes Anyone if nothing that Bill hates is known and if Sue likes Anyone. The original clause stated that Bill likes Anyone if there is some Anyone that Sue likes and that Bill does not hate. Example Always be sure that you think twice when using the not predicate. Incorrect use will result in an error message or errors in your program's logic. The following is an example of the proper way to use the not predicate. /* Program ch070e11.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ PREDICATES nondeterm likes_shopping(symbol) nondeterm has_credit_card(symbol,symbol) bottomed_out(symbol,symbol)

lxxxix

CLAUSES likes_shopping(Who):has_credit_card(Who,Card), not(bottomed_out(Who,Card)), write(Who," can shop with the ",Card, " credit card.\n"). has_credit_card(chris,visa). has_credit_card(chris,diners). has_credit_card(joe,shell). has_credit_card(sam,mastercard). has_credit_card(sam,citibank). bottomed_out(chris,diners). bottomed_out(sam,mastercard). bottomed_out(chris,visa).

Give the following at the goal likes_shopping(Who).

Exercises Suppose an average taxpayer in the USA is a married US citizen with two children who earns no less than $500 a month and no more than $2,000 per month. Define a special_taxpayer predicate that, given the goal special_taxpayer(fred)., will succeed only if fred fails one of the conditions for an average taxpayer. Use the cut to ensure that there is no unnecessary backtracking. Players in a certain squash club are divided into three leagues, and players may only challenge members in their own league or the league below (if there is one). Write a Visual Prolog program that will display all possible matches between club players in the form: tom versus bill marjory versus annette

Use the cut to ensure, for example, that tom versus bill

and bill versus tom

are not both displayed.

This is an exercise in backtracking, not a test of your ability to solve murder mysteries. Load and run the following program, then enter the goal killer(X). (Note: Bert is guilty because he has a motive and is smeared in the same stuff as the victim.) /* Program ch071e12.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS name,sex,occupation,object,vice,substance = symbol age=integer PREDICATES nondeterm person(name, age, sex, occupation) nondeterm had_affair(name, name) killed_with(name, object) killed(name) nondeterm killer(name) motive(vice) smeared_in(name, substance) owns(name, object) nondeterm operates_identically(object, object) nondeterm owns_probably(name, object) nondeterm suspect(name) /* * * Facts about the murder * * */ CLAUSES person(bert,55,m,carpenter). person(allan,25,m,football_player). person(allan,25,m,butcher). person(john,25,m,pickpocket). had_affair(barbara,john). had_affair(barbara,bert). had_affair(susan,john). killed_with(susan,club). killed(susan). motive(money). motive(jealousy). motive(righteousness).

xci

smeared_in(bert, blood). smeared_in(susan, blood). smeared_in(allan, mud). smeared_in(john, chocolate). smeared_in(barbara,chocolate). owns(bert,wooden_leg). owns(john,pistol). /* * * Background knowledge * * */ operates_identically(wooden_leg, club). operates_identically(bar, club). operates_identically(pair_of_scissors, knife). operates_identically(football_boot, club). owns_probably(X,football_boot):person(X,_,_,football_player). owns_probably(X,pair_of_scissors):person(X,_,_,hairdresser). owns_probably(X,Object):owns(X,Object). /* * * *

* * * * * * Suspect all which Susan * * * * * *

* * * those could * * *

* * * * * who own a have been * * * * *

* * * * * * * * weapon with * killed. * * * * * * * * */

suspect(X):killed_with(susan,Weapon) , operates_identically(Object,Weapon) , owns_probably(X,Object). /* * * * * * * * * * * * * * * * * * * * * * * * * * * Suspect men who have had an affair with Susan. * * * * * * * * * * * * * * * * * * * * * * * * * * */ suspect(X):motive(jealousy), person(X,_,m,_), had_affair(susan,X). /* * * *

* * * * * * * * * * Suspect females who affair with someone * * * * * * * * * *

* * * * * * * * * have had an that Susan knew. * * * * * * * * *

* * * */

suspect(X):motive(jealousy), person(X,_,f,_), had_affair(X,Man), had_affair(susan,Man). /* * * * * * * * * * * * * * * * * * * * * * * * * * * * Suspect pickpockets whose motive could be money. * * * * * * * * * * * * * * * * * * * * * * * * * * * */ suspect(X):motive(money), person(X,_,_,pickpocket). killer(Killer):person(Killer,_,_,_), killed(Killed), Killed <> Killer, /* It is not a suicide */ suspect(Killer), smeared_in(Killer,Goo), smeared_in(Killed,Goo).

Prolog from a Procedural Perspective Now that you've read chapters 72, 73, and the first three parts of this chapter, you should have a pretty good understanding of the basics of Prolog programming and using Visual Prolog. Remember, Prolog is a declarative language, which means that you describe a problem in terms of facts and rules and let the computer figure out how to find a solution. Other programming languages--such as Pascal, BASIC, and C--are procedural, which means that you must write subroutines and functions that tell the computer exactly what steps to go through in order to solve the problem. We're going to back up now and review of some of the material you've just learned about Prolog, but this time we're going to present it from a procedural perspective.

How Rules and Facts Are Like Procedures It's easy to think of a Prolog rule as a procedure definition. For instance, the rule likes(bill,Something):- likes(cindy,Something).

means,

xciii

"To prove that Bill likes something, prove that Cindy likes it." With this in mind, you can see how procedures like say_hello:- write("Hello"), nl.

and greet:write("Hello, Earthlings!"), nl.

correspond to subroutines and functions in other programming languages. You can even think of Prolog facts of as procedures; for instance, the fact likes(bill, pasta).

means "To prove that Bill likes pasta, do nothing--and by the way, if the arguments Who and What in your query likes(Who, What) are free variables, you can bind them to bill and pasta, respectively." Some programming procedures that you might be familiar with from other languages are case statements, boolean tests, goto statements, and computational returns. In the next sections, by reiterating what we've already covered from a different (procedural) point of view, we'll show you how Prolog rules can perform these same functions. Using Rules Like Case Statements One big difference between rules in Prolog and procedures in other languages is that Prolog allows you to give multiple alternative definitions of the same procedure. This came up with the "parent" program earlier on page 33; a person can be a parent by being a father or by being a mother, so the definition of "parent" is made up of two rules. You can use multiple definitions like you use a Pascal case statement by writing a different definition for each argument value (or set of argument values). Prolog will try one rule after another until it finds a rule that matches, then perform the actions that rule specifies, as in Program 13. /* Program ch074e13.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ PREDICATES nondeterm action(integer)

CLAUSES action(1):nl, write("You typed 1."),nl. action(2):nl, write("You typed two."),nl. action(3):nl, write("Three was what you typed."),nl. action(N):nl, N<>1, N<>2, N<>3, write("I don't know that number!"). GOAL write("Type a number from 1 to 3: "), readint(Num), action(Num).

If the user types 1, 2, or 3, action will be called with its argument bound to the appropriate value, and it will match only one of the first three rules. Performing Tests within the Rule Look more closely at the fourth clause for action. It will match whatever argument it's called with, binding X to that value. So you have to make sure that it doesn't print I don't know that number unless the number is indeed out of range. That's the purpose of the subgoals X<>1, X<>2, X<>3

where <> means not equal. In order to print I don't know that number, Prolog must first prove that X is not 1, 2, or 3. If any of these subgoals fail, Prolog will try to back up and find alternatives--but there aren't any alternatives, so the rest of the clause will never be executed. Notice that action relies on Choice being bound. If you call action with a free variable as an argument, the goal would match all of the clauses. The first three would return alternative solutions, and then the last one would raise an error because you can't test whether an unbound variable is not equal to a number.

xcv

The Cut as a GoTo Program 13 is somewhat wasteful because, after choosing and executing the correct rule, Prolog still keeps looking for alternatives and has to find out the hard way that the last rule doesn't apply. It would save time and memory if you could tell Prolog to stop looking for alternatives. And you can, by using the cut, which means, "If you get this far, don't do any backtracking within this rule, and don't look for any alternatives to this rule." In other words, "Burn your bridges behind you." Backtracking is still possible, but only at a higher level. If the current rule was called by another rule, and the higher rule has alternatives, they can still be tried. But the cut rules out alternatives within, and alternatives to, the present rule. Using cuts, the program can be rewritten as follows: /* Program ch075e14.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ PREDICATES action(integer) CLAUSES action(1):-!, nl, write("You typed 1."). action(2):-!, nl, write("You typed two."). action(3):-!, nl, write("Three was what you typed."). action(_):write("I don't know that number!"). GOAL write("Type a number from 1 to 3: "), readint(Num), action(Num),nl.

The cut has no effect unless it is actually executed. That is, in order to perform a cut, Prolog must actually get into the rule containing the cut and reach the point where the cut is located. The cut can be preceded by other tests, like this:

action(X) :- X>3, !, write("Too high.").

In this rule, the cut won't have any effect unless the subgoal X>3 succeeds first. Notice that the order of the rules is now significant. In 13, you could have written the rules in any order; only one of them will match any particular number. But in Program 14 you must make sure that the computer doesn't even try the rule that prints I don't know that number unless all of the preceding rules have been tried (and have not executed their cuts). The cuts in 14 are what some people call red cuts--cuts that change the logic of the program. If you had kept the tests X<>1, X<>2, and X<>3, changing the program only by inserting a cut in each clause, you would have been using green cuts-cuts that save time in a program that would be equally correct without them. The efficiency gained is not as great, but there is less risk of making an error in the program. The cut is a powerful, but messy, Prolog operation. In this respect it resembles the goto statement in other programming languages--you can do many things with it, but it can make your program really hard to understand. Returning Computed Values As we have seen, a Prolog rule or fact can return information to the goal that called it. This is done by binding arguments that were previously unbound. The fact likes(bill, cindy).

returns information to the goal likes(bill, Who).

by binding Who to cindy. A rule can return the results of a computation the same way. Here's a simple example: /* Program ch076e15.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ PREDICATES nondeterm classify(integer,symbol)

xcvii

CLAUSES classify(0,zero). classify(X,negative):X < 0. classify(X,positive):X > 0.

The first argument of classify must always be either a constant or a bound variable. The second argument can be either bound or unbound; it gets matched with the symbol zero, negative, or positive, depending on the value of the first argument. Here are some examples of how rules can return values: You can ask whether 45 is positive by giving the goal: Goal classify(45, positive). yes

Because 45 is greater than 0, only the third clause of classify can succeed. In doing so, it matches the second argument with positive. But the second argument is already positive, so the match succeeds, and you get the answer yes. Conversely, if the match fails, you get no: Goal classify(45, negative).

What happens is this: Prolog tries the first clause, but the first argument won't match 0 (nor does the second argument match zero). Then it tries the second clause, binding X to 45, but the test X<0 fails. So it backs out and tries the third clause, but this time the second arguments don't match. To get an actual answer, rather than just yes or no, you must call classify with the second argument free: Goal classify(45, What). What=positive 1 Solution

Here's what really takes place in this case:

The goal

What) won't match the head of the first clause, because 45 doesn't match 0. So the first clause can't be

classify(45,

classify(0, zero),

used. Again, the goal

matches the head of the second clause, binding X to 45 and negative to What. But then the text X<0 fails, because X is 45 and it is not true that 45<0. So Prolog backs out of this clause, undoing the variable bindings just created. classify(45, What)

classify(X, negative),

Finally, classify(45, What) matches classify(X, positive), binding X to 45 and What to positive. The test X>0 succeeds. Since this is a successful solution, Prolog doesn't backtrack; it returns to the calling procedure (which in this case is the goal that you typed). And since the variable X belongs to the calling procedure, that procedure can use its binding--in this case, to print out the value automatically.

Summary In this chapter we've introduced unification, backtracking, determinism, the predicates not and fail, and the cut (!), and we've reviewed the important parts of the tutorial information up to this point from a procedural perspective. Prolog facts and rules receive information by being called with arguments that are constants or bound variables; they return information to the calling procedure by binding variable arguments that were unbound. Unification is the process of matching two predicates and assigning free variables to make the predicates identical. This mechanism is necessary so Prolog can identify which clauses to call and bind values to variables. These are the major points about matching (unification) presented in this chapter: When Prolog begins an attempt to satisfy a goal, it starts at the top of the program in search of a match. When a new call is made, a search for a match to that call also begins at the top of the program. When a call has found a successful match, the call is said to return, and the next subgoal in turn can be tried. Once a variable has been bound in a clause, the only way to free that binding is through backtracking. Backtracking is the mechanism that instructs Prolog where to go to look for solutions to the program. This process gives Prolog the ability to search

xcix

through all known facts and rules for a solution. These are the four basic principles of backtracking given in this chapter: Subgoals must be satisfied in order, from top to bottom. Predicate clauses are tested in the order they appear in the program, from top to bottom. When a subgoal matches the head of a rule, the body of that rule must be satisfied next. The body of the rule then constitutes a new set of subgoals to be satisfied. A goal has been satisfied when a matching fact is found for each of the extremities (leaves) of the goal tree. A call that can produce multiple solutions is non-deterministic, while a call that can produce one and only one solution is deterministic. Visual Prolog provides three tools for controlling the course of your program's logical search for solutions: these are the two predicates fail and not, and the cut. The fail predicate always fails; it forces backtracking in order to find alternate solutions. The not predicate succeeds when its associated subgoal can't be proven true. The cut prevents backtracking. It's easy to think of a Prolog rule as a procedure definition. From a procedural perspective, rules can function as case statements, perform boolean tests, act like goto statements (using the cut), and return computed values.

CHAPTER

Simple and Compound Objects 78So far we've only shown you a few kinds of Visual Prolog data objects, such as numbers, symbols, and strings. In this chapter we discuss the whole range of data objects that Visual Prolog can create, from simple to compound objects. We also show the different types of data structures and data objects that a Visual Prolog program can contain. Because the standard domains do not cover some of the compound data structures, we explain how to declare these compound data structures in both the domains section and the predicates section of your programs.

Simple Data Objects A simple data object is either a variable or a constant. Don't confuse this use of the word "constant" with the symbolic constants you define in the constants section of a program. What we mean here by a constant, is anything identifying an object not subject to variation, such as a character (a char), a number (an integral value or a real), or an atom (a symbol or string).

Variables as Data Objects Variables, which we've discussed in chapter 79, must begin with an upper-case letter (A-Z) or an underscore (_). A single underscore represents an anonymous variable, which stands for a "don't care what it is" situation. In Prolog, a variable can bind with any legal Prolog argument or data object. Prolog variables are local, not global. That is, if two clauses each contain a variable called X, these Xs are two distinct variables. They may get bound to each other if they happen to be brought together during unification, but ordinarily they have no effect on each other.

Constants as Data Objects Constants include characters, numbers, and atoms. Again, don't confuse constants in this context with the symbolic constants defined in the constants section of a

Chapter 80, Repetition and recursion

101

program. A constant's value is its name. That is, the constant 2 can only stand for the number 2, and the constant abracadabra can only stand for the symbol abracadabra. Characters Characters are char type. The printable characters (ASCII 32-127) are the digits 0-9, upper-case letters A-Z, lower-case letters a-z, and the punctuation and familiar TTY characters. Characters outside this range may not be portable between different platforms; in particular, characters less than ASCII 32 (space) are control characters, traditionally used by terminals and communication equipment. A character constant is simply written as the character you want, enclosed by single quotes: 'a' '*' 'W'

'3' '{' 'A'

If, however, you want to specify a backslash or a single quote itself as the character, precede it by a backslash (\): '\\' backslash '\'' single quote. There are a few characters that perform a special function, when preceded by the escape character: '\n'

Newline (linefeed)

'\r'

Carriage return.

'\t'

Tab (horizontal)

Character constants can also be written as their ASCII codes, preceded by the escape character, like this: '\225' '\3'

ß %]

but the exact character displayed by more exotic ASCII values will vary depending on your video-card/terminal.

Numbers Numbers are either from one of the integral domains (see Table 81.8 on page 50), or the real domain. Real numbers are stored in the IEEE standard format and range from 1e-308 to 1e308 (10-308 to 10+308). Examples are: Integers

Real Numbers

3 -77 32034 -10 0

3. 34.96 -32769 4e27 -7.4e-296

Atoms An atom is either a symbol or a string. The distinction between these is largely a question about machine-representation and implementation, and is generally not syntactically visible. When an atom is used as an argument in a predicate call, it is the declaration for the predicate that determines if that argument should be implemented as a string or a symbol. Visual Prolog performs an automatic type conversion between the string domain and the symbol domain, so you can use symbol atoms for string domains and string atoms for the symbol domains. However, there is a loose convention stating that anything in double quotes should be considered a string, while anything not needing to be quoted to be syntactically valid is a symbol: Symbol atoms are names starting with a lower-case letter, and containing only letters, digits, and underscores. String atoms are bound within double quotes and can contain any combination of characters, except ASCII NULL (0, binary zero), which marks the end of the string. Symbol Atoms

String Atoms

food

"Jesse James"

rick_Jones_2nd

"123 Pike street"

fred_Flintstone_1000_Bc_Rd_B edrock

"jon"

"a"

new_york

"New York"

Chapter 80, Repetition and recursion

103

pdcProlog

"Visual Prolog, by Prolog Development Center"

As far as the string/symbol domain interchangeability goes, this distinction is not important. However, things such as predicate names and functors for compound objects, introduced below, must follow the syntactic conventions for symbols.

Compound Data Objects and Functors Compound data objects allow you to treat several pieces of information as a single item in such a way that you can easily pick them apart again. Consider, for instance, the date April 2, 1988. It consists of three pieces of information--the month, day, and year--but it's useful to treat the whole thing as a single object with a treelike structure: / October

DATE | 15

\ 1991

You can do this by declaring a domain containing the compound object date: DOMAINS date_cmp = date(string,unsigned,unsigned)

and then simply writing e.g. ..., D = date("October",15,1991), ...

This looks like a Prolog fact, but it isn't here--it's just a data object, which you can handle in much the same way as a symbol or number. It begins with a name, usually called a functor (in this case date), followed by three arguments. Note carefully that a functor in Visual Prolog has nothing to do with a function in other programming languages. A functor does not stand for some computation to be performed. It's just a name that identifies a kind of compound data object and holds its arguments together. The arguments of a compound data object can themselves be compound. For instance, you might think of someone's birthday as an information structure like this:

BIRTHDAY /

\ / \ person date / \ / | \ "Per" "Bilse" "Apr" 14 1960

In Prolog you would write this as: birthday(person("Per","Bilse"),date("Apr",14,1960))

In this example, there are two parts to the compound object birthday: the object person("Per", "Bilse") and the object date("Apr", 14, 1960). The functors of these data objects are person and date.

Unification of Compound Objects A compound object can unify either with a simple variable or with a compound object that matches it (perhaps containing variables as parts of its internal structure). This means you can use a compound object to pass a whole collection of items as a single object, and then use unification to pick them apart. For example, date("April",14,1960)

matches X and binds X to date("April",14,1960). Also date("April",14,1960)

matches date(Mo,Da,Yr) and binds Mo to "April", Da to 14, and Yr to 1960. Some examples of programming with compound objects follow in the next sections. Using the Equal Sign to Unify Compound Objects Visual Prolog performs unification in two places. The first is when a call or goal matches the head of a clause. The second is the across the equal (=) sign, which is actually an infix predicate (a predicate that is located between its arguments rather than before them). Visual Prolog will make the necessary bindings to unify the objects on both sides of the equal sign. This is useful for finding the values of arguments within a compound object. For example, the following code excerpt tests if two people

Chapter 80, Repetition and recursion

105

have the same last name, then gives the second person the same address as the first. /* Program ch082e01.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS person name address street city,state,street_name first,last number

= = = = = = =

person(name,address) name(first,last) addr(street,city,state) street(number,street_name) string string integer

GOAL P1 = person(name(jim,mos),addr(street(5,"1st st"),igo,"CA")), P1 = person(name(_,mos),Address), P2 = person(name(jane,mos),Address), write("P1=",P1),nl, write("P2=",P2),nl.

Treating Several Items as One Compound objects can be regarded and treated as single objects in your Prolog clauses, which greatly simplifies programming. Consider, for example, the fact owns(john, book("From Here to Eternity", "James Jones")).

in which you state that John owns the book From Here to Eternity, written by James Jones. Likewise, you could write owns(john, horse(blacky)).

which can be interpreted as John owns a horse named blacky.

The compound objects in these two examples are book("From Here to Eternity", "James Jones")

and horse(blacky)

If you had instead written two facts:

owns(john, "From Here to Eternity"). owns(john, blacky ).

you would not have been able to decide whether blacky was the title of a book or the name of a horse. On the other hand, you can use the first component of a compound object--the functor--to distinguish between different objects. This example used the functors book and horse to indicate the difference between the objects. Remember: Compound objects consist of a functor and the objects belonging to that functor, as follows: functor(object1, object2, ..., objectN)

An Example Using Compound Objects A important feature of compound objects allows you to easily pass a group of values as one argument. Consider a case where you are keeping a telephone database. In your database, you want to include your friends' and family members' birthdays. Here is a section of code you might have come up with: PREDICATES phone_list(symbol, symbol, symbol, symbol, integer, integer) /* ( First, Last, Phone, Month, Day, Year) */ CLAUSES phone_list(ed, willis, 422-0208, aug, 3, 1955). phone_list(chris, grahm, 433-9906, may, 12, 1962).

Examine the data, noticing the six arguments in the fact phone_list; five of these arguments can be broken down into two compound objects, like this: person / \ First Name Last Name

birthday / | \ Month Day Year

It might be more useful to represent your facts so that they reflect these compound data objects. Going back a step, you can see that person is a relationship, and the first and last names are the objects. Also, birthday is a relationship with three arguments: month, day, and year. The Prolog representation of these relationships is person(First_name, Last_name) birthday(Month, Day, Year)

Chapter 80, Repetition and recursion

107

You can now rewrite your small database to include these compound objects as part of your database. DOMAINS name = person(symbol, symbol) birthday = b_date(symbol, integer, integer) ph_num = symbo

/* (First, Last) */ /* (Month, Day, Year) */ /* Phone_number */

PREDICATES phone_list(name, ph_num, birthday) CLAUSES phone_list(person(ed, willis), "422-0208", b_date(aug, 3, 1955)). phone_list(person(chris, grahm), "433-9906", b_date(may, 12, 1962)).

In this program, two compound domains declarations were introduced. We go into more detail about these compound data structures later in this chapter. For now, we'll concentrate on the benefits of using such compound objects. The phone_list predicate now contains three arguments, as opposed to the previous six. Sometimes breaking up your data into compound objects will clarify your program and might help process the data. Now add some rules to your small program. Suppose you want to create a list of people whose birthdays are in the current month. Here's the program code to accomplish this task; this program uses the standard predicate date to get the current date from the computer's internal clock. The date predicate is discussed later in chapter 83. For now, all you need to know is that it will return the current year, month, and day from your computer's clock. /* Program ch084e03.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS name = person(symbol,symbol) birthday = b_date(symbol,integer,integer) ph_num = symbol PREDICATES nondeterm phone_list(name,symbol,birthday) get_months_birthdays convert_month(symbol,integer) check_birthday_month(integer,birthday) write_person(name)

/* (First, Last) */ /* (Month, Day, Year) */ /* Phone_number */

CLAUSES get_months_birthdays:write("************ This Month's Birthday List *************"),nl, write(" First name\t\t Last Name\n"), write("*****************************************************"),nl, date(_, This_month, _), /* Get month from system clock */ phone_list(Person, _, Date), check_birthday_month(This_month, Date), write_person(Person), fail. get_months_birthdays:write("\n\n Press any key to continue: "),nl, readchar(_). write_person(person(First_name,Last_name)):write(" ",First_name,"\t\t ",Last_name),nl. check_birthday_month(Mon,b_date(Month,_,_)):convert_month(Month,Month1), Mon = Month1. phone_list(person(ed, willis), "767-8463", b_date(jan, 3, 1955)). phone_list(person(benjamin, thomas), "438-8400", b_date(feb, 5, 1985)). phone_list(person(ray, william), "555-5653", b_date(mar, 3, 1935)). phone_list(person(thomas, alfred), "767-2223", b_date(apr, 29, 1951)). phone_list(person(chris, grahm), "555-1212", b_date(may, 12, 1962)). phone_list(person(dustin, robert), "438-8400", b_date(jun, 17, 1980)). phone_list(person(anna, friend), "767-8463", b_date(jun, 20, 1986)). phone_list(person(brandy, rae), "555-5653", b_date(jul, 16, 1981)). phone_list(person(naomi, friend), "767-2223", b_date(aug, 10, 1981)). phone_list(person(christina, lynn), "438-8400", b_date(sep, 25, 1981)). phone_list(person(kathy, ann), "438-8400", b_date(oct, 20, 1952)). phone_list(person(elizabeth, ann), "555-1212", b_date(nov, 9, 1984)). phone_list(person(aaron, friend), "767-2223", b_date(nov, 15, 1987)). phone_list(person(jennifer, caitlin), "438-8400", b_date(dec, 31, 1981)).

Chapter 80, Repetition and recursion

109

convert_month(jan, convert_month(feb, convert_month(mar, convert_month(apr, convert_month(may, convert_month(jun, convert_month(jul, convert_month(aug, convert_month(sep, convert_month(oct, convert_month(nov, convert_month(dec,

1). 2). 3). 4). 5). 6). 7). 8). 9). 10). 11). 12).

GOAL get_months_birthdays.

Load and run this program. How do compound data objects help in this program? This should be easy to see when you examine the code. Most of the processing goes on in the get_months_birthdays predicate. First, the program makes a window to display the results. After this, it writes a header in the window to help interpret the results. Next, in get_months_birthdays, the program uses the built-in predicate date to obtain the current month. After this, the program is all set to search the database and list the people who were born in the current month. The first thing to do is find the first person in the database. The call phone_list(Person, _, Date) binds the person's first and last names to the variable Person by binding the entire functor person to Person. It also binds the person's birthday to the variable Date. Notice that you only need to use one variable to store a person's complete name, and one variable to hold the birthday. This is the power of using compound data objects. Your program can now pass around a person's birthday simply by passing on the variable Date. This happens in the next subgoal, where the program passes the current month (represented by an integer) and the birthday (of the person it's processing) to the predicate check_birthday_month. Look closely at what happens. Visual Prolog calls the predicate check_birthday_month with two variables: The first variable is bound to an integer, and the second is bound to a birthday term. In the head of the rule that defines check_birthday_month, the first argument, This_month, is matched

with the variable Mon. The second argument, Date, is matched against b_date(Month, _,_). Since all you're concerned with is the month of a person's birthday, you have used the anonymous variable for both the day and the year of birth. The predicate check_birthday_month first converts the symbol for the month into an integer value. Once this is done, Visual Prolog can compare the value of the current month with the value of the person's birthday month. If this comparison succeeds, then the subgoal check_birthday_month succeeds, and processing can continue. If the comparison fails (the person currently being processed was not born in the current month), Visual Prolog begins to backtrack to look for another solution to the problem. The next subgoal to process is write_person. The person currently being processed has a birthday this month, so it's OK to print that person's name in the report. After printing the information, the clause fails, which forces backtracking. Backtracking always goes up to the most recent non-deterministic call and tries to re-satisfy that call. In this program, the last non-deterministic call processed is the call to phone_list. It is here that the program looks up another person to be processed. If there are no more people in the database to process, the current clause fails; Visual Prolog then attempts to satisfy this call by looking further down in the database. Since there is another clause that defines get_months_birthdays, Visual Prolog tries to satisfy the call to get_months_birthdays by satisfying the subgoals to this other clause. Exercise Modify the previous program so that it will also print the birth dates of the people listed. Next, add telephone numbers to the report.

Declaring Domains of Compound Objects In this section, we show you how domains for compound objects are defined. After compiling a program that contains the following relationships: owns(john, book("From Here to Eternity", "James Jones")).

and owns(john, horse(blacky)).

you could query the system with this goal:

Chapter 80, Repetition and recursion

111

owns(john, X)

The variable X can be bound to different types of objects: a book, a horse, or perhaps other objects you define. Because of your definition of the owns predicate, you can no longer employ the old predicate declaration of owns: owns(symbol, symbol)

The second argument no longer refers to objects belonging to the domain symbol. Instead, you must formulate a new declaration to the predicate, such as owns(name, articles)

You can describe the articles domain in the domains section as shown here: DOMAINS articles = book(title,author); horse(name) /* Articles are books or horses */ title, author, name = symbol

The semicolon is read as or. In this case, two alternatives are possible: A book can be identified by its title and author, or a horse can be identified by its name. The domains title, author, and name are all of the standard domain symbol. More alternatives can easily be added to the domains declaration. For example, articles could also include a boat, a house, or a bankbook. For a boat, you can make do with a functor that has no arguments attached to it. On the other hand, you might want to give a bank balance as a figure within the bankbook. The domains declaration of articles is therefore extended to: articles

= book(title, author) ; horse(name) ; boat ; bankbook(balance) title, author, name = symbol balance = real

Here is a full program that shows how compound objects from the domain articles can be used in facts that define the predicate owns. /* Program ch085e04.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS articles

= book(title, author) ; horse(name) ; boat ; bankbook(balance) title, author, name = symbol balance = real

PREDICATES nondeterm owns(name,articles) CLAUSES owns(john, owns(john, owns(john, owns(john,

book("A friend of the family", "Irwin Shaw")). horse(blacky)). boat). bankbook(1000)).

Now compile and run the program with the following goal: owns(john, Thing).

Visual Prolog responds with: Thing=book("A friend of the family","Irwin Shaw") Thing=horse("blacky") Thing=boat Thing=bankbook(1000) 4 Solutions

Writing Domain Declarations: a Summary This is a generic representation of how to write domain declarations for compound objects: domain =alternative1(D, D, ...); alternative2(D, D, ...); ...

Here, alternative1 and alternative2 are arbitrary (but different) functors. The notation (D, D, ...) represents a list of domain names that are either declared elsewhere or are one of the standard domain types (such as symbol, integer, real, etc). Note: The alternatives are separated by semicolons. Every alternative consists of a functor and, possibly, a list of domains for the corresponding arguments. If the functor has no arguments, you can write it as alternativeN or alternativeN( ) in your programs. In this book, we use the former syntax.

Chapter 80, Repetition and recursion

113

Multi-Level Compound Objects Visual Prolog allows you to construct compound objects on several levels. For example, in book("The Ugly Duckling", "Andersen")

instead of using the author's last name, you could use a new structure that describes the author in more detail, including both the author's first and last names. By calling the functor for the resulting new compound object author, you can change the description of the book to book("The Ugly Duckling", author("Hans Christian", "Andersen"))

In the old domain declaration book(title, author)

the second argument of the book functor is author. But the old declaration author = symbol

can only include a single name, so it's no longer sufficient. You must now specify that an author is also a compound object made up of the author`s first and last name. You do this with the domain statement: author = author(first_name, last_name)

which leads to the following declarations: DOMAINS articles = book(title, author); .. /* First level */ author = author(first_name, last_name)/* Second level */ title, first_name, last_name = symbol /* Third level */

When using compound objects on different levels in this way, it's often helpful to draw a "tree": book / \ title author / \ / \ firstname lastname

A domain declaration describes only one level of the tree at a time, and not the whole tree. For instance, a book can't be defined with the following domain declaration:

book = book(title,author(first_name,last_name))

/* Not allowed */

An Example That Illustrates Sentence Structure As another example, consider how to represent the grammatical structure of the sentence ellen owns the book.

using a compound object. The most simple sentence structure consists of a noun and a verb phrase: sentence = sentence(noun, verbphrase)

A noun is just a simple word: noun = noun(word)

A verb phrase consists of either a verb with a noun phrase or a single verb. verbphrase = verbphrase(verb, noun); verb(word) verb = verb(word)

Using these domain declarations (sentence, noun, verbphrase, and verb), the sentence ellen owns the book. becomes sentence(noun(ellen), verbphrase(verb(owns), noun(book)))

A data structure like this might be the output of a parser, which is a program that determines the grammatical structure of a sentence. Parsing is not built into Visual Prolog, but we have included a parser implementing simple sentence analysis with your Visual Prolog package. (Try to run the project VPI\PROGRAMS\SEN_AN when you're ready to tackle this subject.)

Chapter 80, Repetition and recursion

115

Exercises Write a suitable domains declaration using compound objects that could be used in a Visual Prolog catalog of musical shows. A typical entry in the catalog might be Show: West Side Story Lyrics: Stephen Sondheim Music: Leonard Bernstein

Using compound objects wherever possible, write a Visual Prolog program to keep a database of United States senators. Entries should include the senator's first and last name, affiliation (state and party), size of constituency, date of election, and voting record on ten bills. Or, if you're not familiar with United States senators, use any political (or other) organization that you're familiar with.

Compound Mixed-Domain Declarations In this section, we discuss three different types of domain declarations you can add to your programs. These declarations allow you to use predicates that take an argument, more than one type of more than one possible type take a variable number of arguments, each of a specified type take a variable number of arguments, some of which might be of more than one possible type Multiple-Type Arguments To allow a Visual Prolog predicate to accept an argument that gives information of different types, you must add a functor declaration. In the following example, the your_age clause will accept an argument of type age, which can be a string, a real, or an integer. DOMAINS age = i(integer); r(real); s(string) PREDICATES your_age(age) CLAUSES your_age(i(AGE)) :- write(Age). your_age(r(AGE)) :- write(Age). your_age(s(AGE)) :- write(Age).

Visual Prolog does not allow the following domain declaration: DOMAINS age = integer; real; string

/* Not permitted. */

Lists Suppose you are keeping track of the different classes a professor might teach. You might produce the following code: PREDICATES teacher(symbol, symbol, symbol)

/* First_name, Last_name, Class) */

CLAUSES teacher(ed, willis, english1). teacher(ed, willis, math1). teacher(ed, willis, history1). teacher(mary, maker, history2). teacher(mary, maker, math2). teacher(chris, grahm, geometry).

Here, you need to repeat the teacher's name for each class he or she teaches. For each class, you need to add another fact to the database. Although this is perfectly OK in this situation, you might find a school where there are hundreds of classes; this type of data structure would get a little tedious. Here, it would be helpful if you could create an argument to a predicate that could take on one or more values. A list in Prolog does just that. In the following code, the argument class is declared to be of a list type. We show here how a list is represented in Prolog, but list-handling predicates are covered in chapter 86. DOMAINS classes = symbol* PREDICATES teacher(symbol, symbol, classes)

/* declare a list domain */

/* (First, Last, Classes) */

CLAUSES teacher(ed, willis, [english1, math1, history1]). teacher(mary, maker, [history2, math2]). teacher(chris, grahm, [geometry]).

In this example, the code is more concise and easier to read than in the preceding one. Notice the domains declaration: DOMAINS classes = symbol*

Chapter 80, Repetition and recursion

117

The asterisk (*) means that classes is a list of symbols. You can just as easily declare a list of integers: DOMAINS integer_list = integer*

Once you declare a domain, it's easy to use it; just place it as an argument to a predicate declared in the predicates section. Here's an example of using an integer list: DOMAINS integer_list = integer* PREDICATES test_scores(symbol, symbol, integer_list)/* (First, Last, Test Scores) */ CLAUSES test_scores(lisa, lavender, [86, 91, 75]). test_scores(libby, dazzner, [79, 75]). test_scores(jeff, zheutlin, []).

In the case of Jeff Zheutlin, notice that a list doesn't need to contain any elements at all. Lists are discussed in greater detail in chapter 87.

Summary These are the important points covered in this chapter: A Visual Prolog program can contain many types of data objects: simple and compound, standard and user-defined. A simple data object is one of the following: a variable; such as X, MyVariable, _another_variable, or a single underscore ( _ ) for an anonymous variable a constant; a char, an integer or real number, or a symbol or string atom Compound data objects allow you to treat several pieces of information as a single item. A compound data object consists of a name (known as a functor) and one or more arguments. You can define a domain with several alternative functors.

A functor in Visual Prolog is not the same thing as a function in other programming languages. A functor does not stand for some computation to be performed. It's just a name that identifies a kind of compound data object and holds its arguments together. Compound objects can be regarded and treated as single objects; you use the functor to distinguish between different objects. Visual Prolog allows you to construct compound objects on several levels; the arguments of a compound data object can also be compound objects. With compound mixed domain declarations, you can use predicates that: take an argument of more than one possible type (functor declaration). take a variable number of arguments, each of a specified type (list declaration). take a variable number of arguments, some of which might be of more than one possible type.

Chapter 80, Repetition and recursion

119

CHAPTER

Repetition and Recursion 89Much of the usefulness of computers comes from the fact that they are good at doing the same thing over and over again. Prolog can express repetition both in its procedures and in its data structures. The idea of a repetitive data structure may sound strange, but Prolog allows you to create data structures whose ultimate size is not known at the time you create them. In this chapter, we discuss repetitive processes first (as loops and recursive procedures), then cover recursive data structures.

Repetitive Processes Pascal, BASIC, or C programmers who start using Visual Prolog are often dismayed to find that the language has no FOR, WHILE, or REPEAT statements. There is no direct way to express iteration. Prolog allows only two kinds of repetition--backtracking, in which it searches for multiple solutions in a single query, and recursion, in which a procedure calls itself. As it turns out, this lack doesn't restrict the power of the Prolog language. In fact, Visual Prolog recognizes a special case of recursion--called tail recursion --and compiles it into an iterative loop in machine language. This means that although the program logic is expressed recursively, the compiled code is as efficient as it would be in Pascal or BASIC. In this section, we explore the art of writing repetitive processes in Prolog. As you'll see, recursion is--in most cases--clearer, more logical, and less error-prone than the loops that conventional languages use. Before delving into recursion, however, take another look at backtracking.

Backtracking Revisited When a procedure backtracks, it looks for another solution to a goal that has already been satisfied. It does this by retreating to the most recent subgoal that has an untried alternative, using that alternative, then moving forward again. You can exploit backtracking as a way to perform repetitive processes.

Example Program ch06e01.pro demonstrates how to use backtracking to perform repetitive processes--it prints all solutions to a query. /* Program ch090e01.pro */ PREDICATES nondeterm country(symbol) print_countries CLAUSES country("England"). country("France"). country("Germany"). country("Denmark"). print_countries:country(X), write(X), nl, fail. print_countries.

/* write the value of X */ /* start a new line */

The predicate country simply lists the names of various countries, so that a goal such as country(X)

has multiple solutions. The predicate print_countries then prints out all of these solutions. It is defined as follows: print_countries :country(X), write(X), nl, fail. print_countries.

The first clause says: "To print countries, find a solution to line, then fail."

country(X),

then write X and start a new

In this case, "fail" means: "assume that a solution to the original goal has not been reached, so back up and look for an alternative." The built-in predicate fail always fails, but you could equally well force backtracking by using any other goal that would always fail, such as 5=2+2 or country(shangri_la).

Chapter 91, Repetition and recursion

121

The first time through, X is bound to england, which is printed. Then, when it hits fail, the computer backs up. There are no alternative ways to satisfy nl or write(X), so the computer looks for a different solution to country(X). The last time country(X) was executed, it bound a value to the previously free variable X. So, before retrying this step, the computer unbinds X (frees it). Then it can look for an alternative solution for country(X) and bind X to a different value. If it succeeds, processing goes forward again and the name of another country is printed. Eventually, the first clause runs out of alternatives. The only hope then is to try another clause for the same predicate. Sure enough, execution falls through to the second clause, which succeeds without doing anything further. In this way the goal print_countries terminates with success. Its complete output is england france germany denmark yes

If the second clause were not there, the print_countries goal would terminate with failure, and the final message would be no. Apart from that, the output would be the same. Exercise Modify ch06e01.pro so that country has two arguments, name and population, and only those countries with populations greater than 10 million (1e+7) are printed Pre- and Post-Actions Typically, a program that retrieves all the solutions to a goal will also want to do something beforehand and afterward. For instance, your program could Print Some delightful places to live are.... Print all solutions to country(X). Close by printing And

maybe others.

Note that print_countries, as defined in the preceding example, already includes clauses that print all solutions to country(X) and close by (potentially) printing a final message.

The first clause for print_countries corresponds to step 2 and prints all the solutions; its second clause corresponds to step 3 and simply terminates the goal successfully (because the first clause always fails). You could change the second clause in ch06e01.pro to print_countries :- write("And maybe others."), nl.

which would implement step 3 as specified. What about step 1? There's no reason why print_countries should have only two clauses. It can have three, like this: print_countries :write("Some delightful places to live are"), nl, fail. print_countries :country(X), write(X), nl, fail. print_countries :write("And maybe others."), nl.

The fail in the first clause is important--it ensures that, after executing the first clause, the computer backs up and tries the second clause. It's also important that the predicates write and nl do not generate alternatives; strictly speaking, the first clause tries all possible solutions before failing. This three-clause structure is more of a trick than an established programming technique. A more fastidious programmer might try to do things this way: print_countries_with_captions :write("Some delightful places to live are"), nl, print_countries, write("And maybe others."), nl. print_countries :country(X), write(X), nl, fail.

Chapter 91, Repetition and recursion

123

There's nothing essentially wrong here, but this hypothetical fastidious programmer has made a mistake. Exercise Don't look ahead--figure out what's wrong with this program, and fix it! You're right--the problem is that, as written in the latest example, print_countries will always fail, and print_countries_with_captions will never get to execute any of the subgoals that follow it. As a result, And maybe others. will never be printed. To fix this, all you need to do is restore the original second clause for print_countries. print_countries.

to its original position. If you want the goal print_countries_with_captions to succeed, it must have at least one clause that does not contain fail.

Implementing Backtracking with Loops Backtracking is a good way to get all the alternative solutions to a goal. But even if your goal doesn't have multiple solutions, you can still use backtracking to introduce repetition. Simply define the two-clause predicate repeat. repeat :- repeat.

This tricks Prolog's control structure into thinking it has an infinite number of different solutions. (Never mind how--after reading about tail recursion, you'll see how this works.) The purpose of repeat is to allow backtracking ad infinitum. /* Program ch092e02.pro */ /* Uses repeat to keep accepting characters and printing them until the user presses Enter. */ PREDICATES repeat typewriter CLAUSES repeat. repeat:-repeat.

typewriter:repeat, readchar(C), write(C), C = '\r',!.

/* Read a char, bind C to it */ /* Is it a carriage return? fail if not */

Program 2 shows how repeat works. The rule typewriter :- ... describes a procedure that accepts characters from the keyboard and prints them on the screen until the user presses the Enter (Return) key. typewriter works as follows: Execute repeat (which does nothing). Then read a character into the variable C. Then write C. Then check if C is a carriage return. If so, you're finished. If not, backtrack and look for alternatives. Neither write nor readchar generates alternative solutions, so backtrack all the way to repeat, which always has alternative solutions. Now processing can go forward again, reading another character, printing it, and checking whether it's a carriage return. Note, by the way, that C looses its binding when you backtrack past readchar(C), which bound it. This kind of unbinding is vital when you use backtracking to obtain alternative solutions to a goal, but it makes it hard to use backtracking for any other purpose. The reason is that, although a backtracking process can repeat operations any number of times, it can't "remember" anything from one repetition to the next. All variables loose their values when execution backtracks over the steps that established those values. There is no simple way for a repeat loop to keep a counter, a total, or any other record of its progress. Exercises Modify 2 so that, if the user types lower-case letters, they will be displayed as upper-case. If you'd like to play with file I/O now, look up the appropriate built-in predicates and write a program that uses a repeat loop to copy a file character-bycharacter. (Refer to chapter 93.)

Chapter 91, Repetition and recursion

125

Recursive Procedures The other way to express repetition is through recursion. A recursive procedure is one that calls itself. Recursive procedures have no trouble keeping records of their progress because counters, totals, and intermediate results can be passed from each iteration to the next as arguments. The logic of recursion is easy to follow if you forget, for the moment, how computers work. (Prolog is so different from machine language that ignorance of computers is often an asset to the Prolog programmer.) Forget for the moment that the computer is trekking through memory addresses one by one, and imagine a machine that can follow recipes like this one: To find the factorial of a number N: If N is 1, the factorial is 1. Otherwise, find the factorial of N-1, then multiply it by N.

This recipe says: To find the factorial of 3, you must find the factorial of 2, and, to find the factorial of 2, you must find the factorial of 1. Fortunately, you can find the factorial of 1 without referring to any other factorials, so the repetition doesn't go on forever. When you have the factorial of 1, you multiply it by 2 to get the factorial of 2, then multiply that by 3 to get the factorial of 3, and you're done. In Visual Prolog: factorial(1, 1) :- !. factorial(X, FactX) :Y = X-1, factorial(Y, FactY), FactX = X*FactY.

A complete program is as follows: /* Program ch094e03.pro */ /* Recursive program to compute factorials. Ordinary recursion, not tail recursion. */ PREDICATES factorial(unsigned,real) CLAUSES factorial(1,1):-!.

factorial(X,FactX):Y=X-1, factorial(Y,FactY), FactX = X*FactY.

What the Computer is Really Doing But wait a minute, you say. How does the computer execute factorial while it's in the middle of executing factorial? If you call factorial with X=3, factorial will then call itself with X=2. Will X then have two values, or will the second value just wipe out the first, or what? The answer is that the computer creates a new copy of factorial so that factorial can call itself as if it were a completely separate procedure. The executable code doesn't have to be duplicated, of course, but the arguments and internal variables do. This information is stored in an area called a stack frame, which is created every time a rule is called. When the rule terminates, the stack is reset (unless it was a non-deterministic return) and execution continues in the stack frame for the parent. Advantages of Recursion Recursion has three main advantages: It can express algorithms that can't conveniently be expressed any other way. It is logically simpler than iteration. It is used extensively in list processing. Recursion is the natural way to describe any problem that contains within itself another problem of the same kind. Examples include tree search (a tree is made up of smaller trees) and recursive sorting (to sort a list, partition it, sort the parts, and then put them together). Logically, recursive algorithms have the structure of an inductive mathematical proof. The preceding recursive factorial algorithm, in Program 3, describes an infinite number of different computations by means of just two clauses. This makes it easy to see that the clauses are correct. Further, the correctness of each clause can be judged independently of the other.

Chapter 91, Repetition and recursion

127

Tail Recursion Optimisation Recursion has one big drawback: It eats memory. Whenever one procedure calls another, the calling procedure's state of execution must be saved so that it (the calling procedure) can resume where it left off after the called procedure has finished. This means that, if a procedure calls itself 100 times, 100 different states of execution must be stored at once. (The saved state of execution is known as a stack frame.) The maximum stack size on 16bit platforms, such as the IBM PC running DOS, is 64K, which will accommodate, at most, 3000 or 4000 stack frames. On 32bit platforms, the stack may theoretically grow to several GigaBytes; here, other system limitations will set in before the stack overflows. Anyway, what can be done to avoid using so much stack space? It turns out that there's a special case in which a procedure can call itself without storing its state of execution. What if the calling procedure isn't going to resume after the called procedure finishes? Suppose the calling procedure calls a procedure as its very last step. When the called procedure finishes, the calling procedure won't have anything else to do. This means the calling procedure doesn't need to save its state of execution, because that information isn't needed any more. As soon as the called procedure finishes, control can go directly to wherever it would have gone when the calling procedure finished. For example, suppose that procedure A calls procedure B, and B calls procedure C as its very last step. When B calls C, B isn't going to do anything more. So, instead of storing the current state of execution for C under B, you can replace B's old stored state (which isn't needed any more) with C's current state, making appropriate changes in the stored information. When C finishes, it thinks it was called by A directly. Now suppose that, instead of calling C, procedure B calls itself as its very last step. The recipe says that, when B calls B, the stack frame for the calling B should be replaced by a stack frame for the called B. This is a particularly simple operation; only the arguments need to be set to new values, and then processing jumps back to the beginning of the procedure. So, from a procedural point of view, what happens is very similar to updating the control variables in a loop. This is called tail recursion optimization, or last-call optimization. Note that for technical reasons, recursive functions (predicates returning a value, described in chapter 95) cannot be tail recursive. Making Tail Recursion Work What does it mean to say that one procedure calls another "as its very last step?" In Prolog, this means that

The call is the very last subgoal of the clause. There are no backtrack points earlier in the clause. Here's an example that satisfies both conditions: count(N) :write(N), nl, NewN = N+1, count(NewN).

This procedure is tail recursive; it calls itself without allocating a new stack frame, so it never runs out of memory. As program 4 shows, if you give it the goal count(0)

count will print integers starting with 0 and never ending. Eventually, rounding errors will make it print inaccurate numbers, but it will never stop. /* Program ch096e04.pro */ /* Tail recursive program that never runs out of memory */ PREDICATES count(ulong) CLAUSES count(N):write('\r',N), NewN = N+1, count(NewN). GOAL nl, count(0).

Exercise Without looking ahead, modify 4 so that it is no longer tail recursive. How many iterations can it execute before running out of memory? Try it and see. (On 32bit platforms, this will take a considerable length of time, and the program will most likely not run out of stack space; it, or the system, will run out of memory in general. On 16bit platforms, the number of possible iterations is directly related to the stack size.

Chapter 91, Repetition and recursion

129

How Not to Do Tail Recursion Now that you've seen how to do tail recursion right, program 5 shows you three ways to do it wrong. If the recursive call isn't the very last step, the procedure isn't tail recursive. For example: badcount1(X) :write('\r',X), NewX = X+1, badcount1(NewX), nl.

Every time badcount1 calls itself, a stack frame has to be saved so that control can return to the calling procedure, which has yet to execute its final nl. So only a few thousand recursive calls can take place before the program runs out of memory. Another way to lose tail recursion is to leave an alternative untried at the time the recursive call is made. Then a stack frame must be saved so that, if the recursive call fails, the calling procedure can go back and try the alternative. For example: badcount2(X) :write('\r',X), NewX = X+1, badcount2(NewX). badcount2(X) :X < 0, write("X is negative.").

Here, the first clause of badcount2 calls itself before the second clause has been tried. Again, the program runs out of memory after a certain number of calls. The untried alternative doesn't need to be a separate clause for the recursive procedure itself. It can equally well be an alternative in some other clause that it calls. For example: badcount3(X) :write('\r',X), NewX = X+1, check(NewX), badcount3(NewX). check(Z) :- Z >= 0. check(Z) :- Z < 0.

Suppose X is positive, as it normally is. Then, when badcount3 calls itself, the first clause of check has succeeded, but the second clause of check has not yet been tried. So badcount3 has to preserve a copy of its stack frame in order to go back and try the other clause of check if the recursive call fails. /* Program ch097e05.pro */ /* In 32bit memory architectures, such as '386 UNIX, the examples here will run for a considerable length of time, occupying large amounts of memory and possibly reducing system performance significantly. */ PREDICATES badcount1(long) badcount2(long) badcount3(long) check(long) CLAUSES /* badcount1: The recursive call is not the last step. */ badcount1(X):write('\r',X), NewX = X+1, badcount1(NewX), nl. /* badcount2: There is a clause that has not been tried at the time the recursive call is made. */ badcount2(X):write('\r',X), NewX = X+1, badcount2(NewX). badcount2(X):X < 0, write("X is negative."). /* badcount3: There is an untried alternative in a predicate called before the recursive call. */

Chapter 91, Repetition and recursion

131

badcount3(X):write('\r',X), NewX = X+1, check(NewX), badcount3(NewX). check(Z):Z >= 0. check(Z):Z < 0.

Cuts to the Rescue By now, you may think it's impossible to guarantee that a procedure is tail recursive. After all, it's easy enough to put the recursive call in the last subgoal of the last clause, but how do you guarantee there are no alternatives in any of the other procedures that it calls? Fortunately, you don't have to. The cut (!) allows you to discard whatever alternatives may exist. You'll need to use the check_determ compiler directive to guide you through setting the cuts. (Compiler directives are described in the chapter 98.) You can fix up badcount3 as follows (changing its name in the process): cutcount3(X) :write('\r',X), NewX = X+1, check(NewX), !, cutcount3(NewX).

leaving check as it was. The cut means "burn your bridges behind you" or, more precisely, "once you reach this point, disregard alternative clauses for this predicate and alternative solutions to earlier subgoals within this clause." That's precisely what you need. Because alternatives are ruled out, no stack frame is needed and the recursive call can go inexorably ahead. A cut is equally effective in badcount2, by negating and moving the test from the second clause to the first:

cutcount2(X) :X >= 0, !, write('\r',X), NewX = X+1, cutcount2(NewX). cutcount2(X) :write("X is negative.").

A cut is really all about making up ones mind. You set a cut whenever you can look at non-deterministic code, and say "Yes! Go ahead!" -- whenever it's obvious that alternatives are of no interest. In the original version of the above example, which tries to illustrate a situation where you have to decide something about X (the test X < 0 in the second clause), the second clause had to remain an option as the code in the first clause didn't test X. By moving the test to the first clause and negating it, a decision can be reached already there and a cut set in accordance: "Now I know I don't want to write that X is negative.". The same applies to cutcount3. The predicate check illustrates a situation where you want to do some additional processing of X, based on its sign. However, the code for check is, in this case for illustration, non-deterministic, and the cut after the call to it is all about you having made up your mind. After the call to check, you can say "Yes! Go ahead!". However, the above is slightly artificial -- it would probably be more correct for check to be deterministic: check(Z) :- Z >= 0, !, ... % processing using Z check(Z) :- Z < 0, ... %processing using Z

And, since the test in the second clause of check is the perfect negation of the test in the first, check can be further rewritten as: check(Z) :- Z >= 0, !, % processing using Z check(Z) :- ... % processing using Z

When a cut is executed, the computer assumes there are no untried alternatives and does not create a stack frame. Program 6 contains modified versions of badcount2 and badcount3: /* Program ch099e06.pro */ /* Shows how badcount2 and badcount3 can be fixed by adding cuts to rule out the untried clauses. These versions are tail recursive. */ PREDICATES cutcount2(long) cutcount3(long) nondeterm check(long)

Chapter 91, Repetition and recursion

133

CLAUSES /* cutcount2: There is a clause that has not been tried at the time the recursive call is made. */ cutcount2(X):X>=0,!, write('\r',X), NewX = X + 1, cutcount2(NewX). cutcount2(_):write("X is negative."). /* cutcount3: There is an untried alternative in a clause called before the recursive call. */ cutcount3(X):write('\r',X), NewX = X+1, check(NewX), !, cutcount3(NewX). check(Z):-Z >= 0. check(Z):-Z < 0.

Unfortunately, cuts won't help with badcount1, whose need for stack frames has nothing to do with untried alternatives. The only way to improve badcount1 would be to rearrange the computation so that the recursive call comes at the end of the clause.

Using Arguments as Loop Variables Now that you've mastered tail recursion, what can you do about loop variables and counters? To answer that question, we'll do a bit of Pascal-to-Prolog translation, assuming that you're familiar with Pascal. Generally, the results of direct translations between two languages, whether natural or programming, are poor. The following isn't too bad and serves as a reasonable illustration of strictly imperative programming in Prolog, but you should never write Prolog programs by blind translation from another language. Prolog is a very powerful and expressive language, and properly written Prolog programs will display a programming style and problem focus quite different from what programs in other languages do.

In the "Recursion" section, we developed a recursive procedure to compute factorials; in this section we'll develop an iterative one. In Pascal, this would be: P := 1; for I := 1 to N do P := P*I; FactN := P;

If you're unfamiliar with Pascal, the :- is the assignment, read as "becomes". There are four variables here. N is the number whose factorial will be calculated; FactN is the result of the calculation; I is the loop variable, counting from 1 to N; and P is the variable in which the product accumulates. A more efficient Pascal programmer might combine FactN and P, but in Prolog it pays to be fastidiously tidy. The first step in translating this into Prolog is to replace for with a simpler loop statement, making what happens to I in each step more explicit. Here is the algorithm recast as a while loop: P := 1; I := 1; while I <= N do begin P := P*I; I := I+1 end; FactN := P;

/* Initialize P and I */ /* Loop test */ /* Update P and I */

/* Return result */

shows the Prolog translation constructed from this Pascal while loop. /* Program ch0100e07.pro */ PREDICATES factorial(unsigned,long) factorial_aux(unsigned,long,unsigned,long) /* Numbers likely to become large are declared as longs. */ CLAUSES factorial(N, FactN):factorial_aux(N,FactN,1,1). factorial_aux(N,FactN,I,P):I <= N,!, NewP = P * I, NewI = I + 1, factorial_aux(N, FactN, NewI, NewP).

Chapter 91, Repetition and recursion

135

factorial_aux(N, FactN, I, P) :I > N, FactN = P.

Let's look at this in greater detail. The factorial clause has only N and FactN as arguments; they are its input and output, from the viewpoint of someone who is using it to find a factorial. A second clause, factorial_aux(N, FactN, I, P), actually performs the recursion; its four arguments are the four variables that need to be passed along from each step to the next. So factorial simply invokes factorial_aux, passing to it N and FactN, along with the initial values for I and P, like so: factorial(N, FactN) :factorial_aux(N, FactN, 1, 1).

That's how I and P get initialized. How can factorial "pass along" FactN? It doesn't even have a value yet! The answer is that, conceptually, all Visual Prolog's doing here is unifying a variable called FactN in one clause with a variable called FactN in another clause. The same thing will happen whenever factorial_aux passes FactN to itself as an argument in a recursive call. Eventually, the last FactN will get a value, and, when this happens, all the other FactN's, having been unified with it, will get the same value. We said "conceptually" above, because in reality there is only one FactN. Visual Prolog can determine from the source code that FactN is never really used before the second clause for factorial_aux, and just shuffles the same FactN around all the time. Now for factorial_aux. Ordinarily, this predicate will check that I is less than or equal to N--the condition for continuing the loop--and then call itself recursively with new values for I and P. Here another peculiarity of Prolog asserts itself. In Prolog there is no assignment statement such as P = P + 1

which is found in most other programming languages. You can't change the value of a Prolog variable. In Prolog, the above is as absurd as in algebra, and will fail. Instead, you have to create a new variable and say something like NewP = P + 1

So here's the first clause:

factorial_aux(N, FactN, I, P) :I <= N, !, NewP = P*I, NewI = I+1, factorial_aux(N, FactN, NewI, NewP).

As in cutcount2, the cut enables last-call optimization to take effect, even though the clause isn't the last in the predicate. Eventually I will exceed N. When it does, processing should unify the current value of P with FactN and stop the recursion. This is done in the second clause, which will be reached when the test I <= N in the first clause fails: factorial_aux(N, FactN, I, P) :I > N, FactN = P.

But there is no need for FactN = P to be a separate step; the unification can be performed in the argument list. Putting the same variable name in the positions occupied by FactN and P requires the arguments in these positions to be matched with each other. Moreover, the test I > N is redundant since the opposite has been tested for in the first clause. This gives the final clause: factorial_aux(_, FactN, _, FactN).

Exercises The following is a more elegant version of factorial. /* Program ch0101e08.pro */ PREDICATES factorial(unsigned,real) factorial(unsigned,real,unsigned,real) /* Numbers likely to become large are declared as reals. */ CLAUSES factorial(N,FactN):factorial(N,FactN,1,1). factorial(N,FactN,N,FactN):-!. factorial(N,FactN,I,P):NewI = I+1, NewP = P*NewI, factorial(N, FactN, NewI, NewP).

Chapter 91, Repetition and recursion

137

Load and run this program. Carefully look at the code in the second clause of factorial/4. It takes advantage of the fact that the first time it's called the counter variable I always has the value 1. This allows the multiplication step to be carried out with the incremented counter variable NewI rather than I, saving one recursion/iteration. This is reflected in the first clause. Write a tail recursive program that behaves like 2 but doesn't use backtracking. Write a tail recursive program that prints a table of powers of 2, like this: N -1 2 3 4 ... 10

2^N ----2 4 8 16 ... 1024

Make it stop at 10 as shown here. Write a tail recursive program that accepts a number as input and can end in either of two ways. It will start multiplying the number by itself over and over until it either reaches 81 or reaches a number greater than 100. If it reaches 81, it will print yes; if it exceeds 100, it will print no.

Recursive Data Structures Not only can rules be recursive; so can data structures. Prolog is the only widely used programming language that allows you to define recursive data types. A data type is recursive if it allows structures to contain other structures like themselves. The most basic recursive data type is the list, although it doesn't immediately look recursively constructed. A lot of list-processing power is built into Prolog, but we won't discuss it here; lists are such an important part of Prolog that there is a whole chapter devoted to them, chapter 102. In this chapter, we invent a recursive data type, implement it, and use it to write a very fast sorting program. The structure of this invented recursive data type is a tree (Figure 103.1). Crucially, each branch of the tree is itself a tree; that's why the structure is recursive.

Cathy

Michael

Charles

Melody

Hazel

Jim

Eleanor

Figure 104.2: Part of a Family Tree1053

Trees as a Data Type Recursive types were popularized by Niklaus Wirth in Algorithms + Data Structures = Programs. Wirth derived Pascal from ALGOL60 and published this work in the early 70's. He didn't implement recursive types in Pascal, but he did discuss what it would be like to have them. If Pascal had recursive types, you would be able to define a tree as something like this: tree = record name: string[80]; left, right: tree end;

/* Not correct Pascal! */

This code, translated into natural language, means "A tree consists of a name, which is a string, and the left and right subtrees, which are trees." The nearest approach to this in Pascal is to use pointers and say treeptr = ^tree; tree = record name: string[80]; left, right: treeptr end;

Chapter 91, Repetition and recursion

139

But notice a subtle difference: This code deals with the memory representation of a tree, not the structure of the tree itself. It treats the tree as consisting of cells, each containing some data plus pointers to two more cells. Visual Prolog allows truly recursive type definitions in which the pointers are created and maintained automatically. For example, you can define a tree as follows: DOMAINS treetype = tree(string, treetype, treetype)

This declaration says that a tree will be written as the functor, tree, whose arguments are a string and two more trees. But this isn't quite right; it provides no way to end the recursion, and, in real life, the tree does not go on forever. Some cells don't have links to further trees. In Pascal, you could express this by setting some pointers equal to the special value nil, but pointers are an implementation issue that ordinarily doesn't surface in Prolog source code. Rather, in Prolog we define two kinds of trees: ordinary ones and empty ones. This is done by allowing a tree to have either of two functors: tree, with three arguments, or empty, with no arguments. DOMAINS treetype = tree(string, treetype, treetype) ; empty

Notice that the names tree (a functor that takes three arguments) and empty (a functor taking no arguments) are created by the programmer; neither of them has any pre-defined meaning in Prolog. You could equally well have used xxx and yyy. This is how the tree in Figure 106.4 could appear in a Prolog program: tree("Cathy", tree("Michael" tree("Charles", empty, empty) tree("Hazel", empty, empty)) tree("Melody" tree("Jim", empty, empty) tree("Eleanor", empty, empty)))

This is indented here for readability, but Prolog does not require indentation, nor are trees indented when you print them out normally. Another way of setting up this same data structure is:

tree("Cathy" tree("Michael", tree("Charles", empty, empty), tree("Hazel", empty, empty)) tree("Melody", tree("Jim", empty, empty), tree("Eleanor", empty, empty)))

Note that this is not a Prolog clause; it is just a complex data structure. Traversing a Tree Before going on to the discussion of how to create trees, first consider what you'll do with a tree once you have it. One of the most frequent tree operations is to examine all the cells and process them in some way, either searching for a particular value or collecting all the values. This is known as traversing the tree. One basic algorithm for doing so is the following: If the tree is empty, do nothing. Otherwise, process the current node, then traverse the left subtree, then traverse the right subtree. Like the tree itself, the algorithm is recursive: it treats the left and right subtrees exactly like the original tree. Prolog expresses it with two clauses, one for empty and one for nonempty trees: traverse(empty).

/* do nothing */

traverse(tree(X, Y, Z)) :do something with X, traverse(Y), traverse(Z).

Chapter 91, Repetition and recursion

141

Cathy 2

Michael

Melody

Charles 4

Hazel

Jim

Eleanor 7

Figure 107.5: Depth-First Traversal of the Tree in Figure 108.6109110 This tree traversal algorithm is known as depth-first search because it goes as far as possible down each branch before backing up and trying another branch (Figure 111.7). To see it in action, look at program 9, which traverses a tree and prints all the elements as it encounters them. Given the tree in Figures 112.8 and 113.9, 9 prints Cathy Michael Charles Hazel Melody Jim Eleanor

Of course, you could easily adapt the program to perform some other operation on the elements, rather than printing them. /* Program ch0114e09.pro */ /* Traversing a tree by depth-first search and printing each element as it is encountered */ DOMAINS treetype = tree(string, treetype, treetype) ; empty()

PREDICATES traverse(treetype) CLAUSES traverse(empty). traverse(tree(Name,Left,Right)):write(Name,'\n'), traverse(Left), traverse(Right). GOAL traverse(tree("Cathy", tree("Michael", tree("Charles", empty, empty), tree("Hazel", empty, empty)), tree("Melody", tree("Jim", empty, empty), tree("Eleanor", empty, empty)))).

Depth-first search is strikingly similar to the way Prolog searches a knowledge base, arranging the clauses into a tree and pursuing each branch until a query fails. If you wanted to, you could describe the tree by means of a set of Prolog clauses such as: father_of("Cathy", "Michael"). mother_of("Cathy", "Melody"). father_of("Michael", "Charles"). mother_of("Michael", "Hazel"). ...

This is preferable if the only purpose of the tree is to express relationships between individuals. But this kind of description makes it impossible to treat the whole tree as a single complex data structure; as you'll see, complex data structures are very useful because they simplify difficult computational tasks. Creating a Tree One way to create a tree is to write down a nested structure of functors and arguments, as in the preceding example (Program 9). Ordinarily, however, Prolog creates trees by computation. In each step, an empty subtree is replaced by a nonempty one through Prolog's process of unification (argument matching). Creating a one-cell tree from an ordinary data item is trivial: create_tree(N, tree(N, empty, empty)).

Chapter 91, Repetition and recursion

143

This says: "If N is a data item, then containing it."

tree(N, empty, empty)

is a one-cell tree

Building a tree structure is almost as easy. The following procedure takes three trees as arguments. It inserts the first tree as the left subtree of the second tree, giving the third tree as the result: insert_left(X, tree(A, _, B), tree(A, X, B)).

Notice that this rule has no body--there are no explicit steps in executing it. All the computer has to do is match the arguments with each other in the proper positions, and the work is done. Suppose, for example, you want to insert tree("Michael", empty, empty) as the left subtree of tree("Cathy", empty, empty). To do this, just execute the goal insert_left(tree("Michael", empty, empty), tree("Cathy", empty, empty), T)

and T immediately takes on the value tree("Cathy", tree("Michael", empty, empty), empty).

This gives a way to build up trees step-by-step. Program 10 demonstrates this technique. In real life, the items to be inserted into the tree could come from external input. /* Program ch0115e10.pro */ /* * * * * * *

* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * Simple tree-building procedures * create_tree(A, B) puts A in the data field of a one-cell tree * giving B insert_left(A, B, C) inserts A as left subtree of B * giving C insert_right(A, B, C) inserts A as right subtree of B * giving C * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */

DOMAINS treetype = tree(string,treetype,treetype) ; empty() PREDICATES create_tree(string,treetype) insert_left(treetype,treetype,treetype) insert_right(treetype, treetype, treetype)

CLAUSES create_tree(A,tree(A,empty,empty)). insert_left(X,tree(A,_,B),tree(A,X,B)). insert_right(X,tree(A,B,_),tree(A,B,X)). GOAL /* First create some one-cell trees... */ create_tree("Charles",Ch), create_tree("Hazel",H), create_tree("Michael",Mi), create_tree("Jim",J), create_tree("Eleanor",E), create_tree("Melody",Me), create_tree("Cathy",Ca), /* ...then link them up... */ insert_left(Ch, Mi, Mi2), insert_right(H, Mi2, Mi3), insert_left(J, Me, Me2), insert_right(E, Me2, Me3), insert_left(Mi3, Ca, Ca2), insert_right(Me3, Ca2, Ca3), /* ...and print the result. */ write(Ca3,'\n').

Notice that there is no way to change the value of a Prolog variable once it is bound. That's why 10 uses so many variable names; every time you create a new value, you need a new variable. The large number of variable names here is unusual; more commonly, repetitive procedures obtain new variables by invoking themselves recursively, since each invocation has a distinct set of variables.

Binary Search Trees So far, we have been using the tree to represent relationships between its elements. Of course, this is not the best use for trees, since Prolog clauses can do the same job. But trees have other uses. Trees provide a good way to store data items so that they can be found quickly. A tree built for this purpose is called a search tree; from the user's point of view, the tree structure carries no information--the tree is merely a faster alternative to a list or array. Recall that, to traverse an ordinary tree, you look at the current cell and then at both of its subtrees. To find a particular item, you might have to look at every cell in the whole tree.

Chapter 91, Repetition and recursion

145

The time taken to search an ordinary tree with N elements is, on the average, proportional to N. A binary search tree is constructed so that you can predict, upon looking at any cell, which of its subtrees a given item will be in. This is done by defining an ordering relation on the data items, such as alphabetical or numerical order. Items in the left subtree precede the item in the current cell and, in the right subtree, they follow it. Figure 116.10 shows an example. Note that the same names, added in a different order, would produce a somewhat different tree. Notice also that, although there are ten names in the tree, you can find any of them in--at most-five steps. Grasso

Blackwell

Anthony

Rankin

Tubmann

Mott

Chrisholm

Lovelace

OKeeffe

Stanton

Figure 117.11: Binary Search Tree11812 Every time you look at a cell in a binary search tree during a search, you eliminate half the remaining cells from consideration, and the search proceeds very quickly. If the size of the tree were doubled, then, typically, only one extra step would be needed to search it. The time taken to find an item in a binary search tree is, on the average, proportional to log2 N (or, in fact, proportional to log N with logarithms to any base).

To build the tree, you start with an empty tree and add items one by one. The procedure for adding an item is the same as for finding one: you simply search for the place where it ought to be, and insert it there. The algorithm is as follows: If the current node is an empty tree, insert the item there. Otherwise, compare the item to be inserted and the item stored in the current node. Insert the item into the left subtree or the right subtree, depending on the result of the comparison. In Prolog, this requires three clauses, one for each situation. The first clause is insert(NewItem, empty, tree(NewItem, empty, empty) :- !.

Translated to natural language, this code says "The result of inserting NewItem into empty is tree(NewItem, empty, empty)." The cut ensures that, if this clause can be used successfully, no other clauses will be tried. The second and third clauses take care of insertion into nonempty trees: insert(NewItem, tree(Element, Left, Right), tree(Element, NewLeft, Right) :NewItem < Element, !, insert(NewItem, Left, NewLeft). insert(NewItem, tree(Element, Left, Right), tree(Element, Left, NewRight) :insert(NewItem, Right, NewRight).

If NewItem < Element, you insert it into the left subtree; otherwise, you insert it into the right subtree. Notice that, because of the cuts, you get to the third clause only if neither of the preceding clauses has succeeded. Also notice how much of the work is done by matching arguments in the head of the rule. Tree-Based Sorting Once you have built the tree, it is easy to retrieve all the items in alphabetical order. The algorithm is again a variant of depth-first search: If the tree is empty, do nothing. Otherwise, retrieve all the items in the left subtree, then the current element, then all the items in the right subtree. Or, in Prolog: retrieve_all(empty).

Chapter 91, Repetition and recursion

/* Do nothing */

147

retrieve_all(tree(Item, Left, Right)) :retrieve_all(Left), do_something_to(Item), retrieve_all(Right).

You can sort a sequence of items by inserting them into a tree and then retrieving them in order. For N items, this takes time proportional to N log N, because both insertion and retrieval take time proportional to log N, and each of them has to be done N times. This is the fastest known sorting algorithm. Example Program 11 uses this technique to alphabetize character input. In this example we use some of Visual Prolog's standard predicates we haven't introduced before. These predicates will be discussed in detail in later chapters. /* Program ch0119e11.pro */ DOMAINS chartree = tree(char, chartree, chartree); end PREDICATES nondeterm do(chartree) action(char, chartree, chartree) create_tree(chartree, chartree) insert(char, chartree, chartree) write_tree(chartree) nondeterm repeat CLAUSES do(Tree):repeat,nl, write("*****************************************************"),nl, write("Enter 1 to update tree\n"), write("Enter 2 to show tree\n"), write("Enter 7 to exit\n"), write("*****************************************************"),nl, write("Enter number - "), readchar(X),nl, action(X, Tree, NewTree), do(NewTree).

action('1',Tree,NewTree):write("Enter characters or # to end: "), create_Tree(Tree, NewTree). action('2',Tree,Tree):write_Tree(Tree), write("\nPress a key to continue"), readchar(_),nl. action('7', _, end):exit. create_Tree(Tree, NewTree):readchar(C), C<>'#',!, write(C, " "), insert(C, Tree, TempTree), create_Tree(TempTree, NewTree). create_Tree(Tree, Tree). insert(New,end,tree(New,end,end)):-!. insert(New,tree(Element,Left,Right),tree(Element,NewLeft,Right)):New<Element,!, insert(New,Left,NewLeft). insert(New,tree(Element,Left,Right),tree(Element,Left,NewRight)):insert(New,Right,NewRight). write_Tree(end). write_Tree(tree(Item,Left,Right)):write_Tree(Left), write(Item, " "), write_Tree(Right). repeat. repeat:-repeat. GOAL write("*************** Character tree sort *******************"),nl, do(end).

Load and run Program 11 and watch how Visual Prolog does tree-based sorting on a sequence of characters. Exercises Program 12 is similar to 11, but more complex. It uses the same sorting technique to alphabetize any standard text file, line by line. Typically it's more than five times faster than "SORT.EXE", the sort program provided by DOS and OS/2,

Chapter 91, Repetition and recursion

149

but it's beaten by the highly optimized "sort" on UNIX. Nevertheless, treebased sorting is remarkably efficient. In this example we use some of the predicates from Visual Prolog's file system, to give you a taste of file redirection. To redirect input or output to a file, you must tell the system about the file; you use openread to read from the file or openwrite to write to it. Once files are open, you can switch I/O between an open file and the screen with writedevice, and between an open file and the keyboard with readdevice. These predicates are discussed in detail later in chapter 120. Load and run Program 12. When it prompts File to read type in the name of an existing text file; the program will then alphabetize that file, line by line. /* Program ch0121e12.pro */ DOMAINS treetype = tree(string, treetype, treetype) ; empty file = infile ; outfile PREDICATES main read_input(treetype) read_input_aux(treetype, treetype) insert(string, treetype, treetype) write_output(treetype) CLAUSES main :write("PDC Prolog Treesort"),nl, write("File to read: "), readln(In),nl, openread(infile, In), /* open the specified file write("File to write: "), readln(Out),nl, openwrite(outfile, Out), readdevice(infile), /* redirect all read operations to the read_input(Tree), writedevice(outfile), /* redirect all write operations to the write_output(Tree), closefile(infile), /* close the file opened closefile(outfile).

for reading */

opened file */

opened file */ for reading */

/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * read_input(Tree) * reads lines from the current input device until EOF, then * instantiates Tree to the binary search tree built * therefrom * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

* * * * * */

read_input(Tree):read_input_aux(empty,Tree). /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * read_input_aux(Tree, NewTree) * reads a line, inserts it into Tree giving NewTree, * and calls itself recursively unless at EOF. * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

* * * * */

read_input_aux(Tree, NewTree):readln(S), !, insert(S, Tree, Tree1), read_input_aux(Tree1, NewTree). read_input_aux(Tree, Tree). /* The first clause fails at EOF. */ /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * insert(Element, Tree, NewTree) * * inserts Element into Tree giving NewTree. * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ insert(NewItem, empty, tree(NewItem,empty,empty)):-!. insert(NewItem,tree(Element,Left,Right),tree(Element,NewLeft, Right)):NewItem < Element, !, insert(NewItem, Left, NewLeft). insert(NewItem,tree(Element,Left,Right),tree(Element,Left,NewRight)):insert(NewItem, Right, NewRight). /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * write_output(Tree) * * writes out the elements of Tree in alphabetical order. * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ write_output(empty). /* Do nothing */ write_output(tree(Item,Left,Right)):write_output(Left), write(Item), nl, write_output(Right).

Chapter 91, Repetition and recursion

151

GOAL main,nl.

Use recursive data structures to implement hypertext. A hypertext is a structure in which each entry, made up of several lines of text, is accompanied by pointers to several other entries. Any entry can be connected to any other entry; for instance, you could get to an entry about Abraham Lincoln either from "Presidents" or from "Civil War." To keep things simple, use one-line entries (strings) and let each of them contain a pointer to only one other entry. Hint: Start with DOMAINS entrytype = empty() ; entry(string, entry)

Build a linked structure in which most of the entries have a nonempty second argument. Now, take your hypertext implementation and redo it using Prolog clauses. That is, use clauses (rather than recursive data structures) to record which entry follows which.

Summary These are the major points covered in this chapter: In Prolog there are two ways to repeat the same clause; through backtracking and recursion. By failing, Prolog will backtrack to find a new piece of data and repeat the clause until there are no more options. Recursion is the process of a clause calling itself. Backtracking is very powerful and memory efficient, but variables are freed after each iteration, so their values are lost. Recursion allows variables to be incremented, but it is not memory efficient. However, Visual Prolog does tail recursion elimination, which relieves the memory demands of recursion. For Visual Prolog to achieve tail recursion elimination, the recursive call must be the last subgoal in the clause body.

CHAPTER

122

Lists and Recursion 123List processing--handling objects that contain an arbitrary number of elements--is a powerful technique in Prolog. In this chapter, we explain what lists are and how to declare them, then give several examples that show how you might use list processing in your own applications. We also define two wellknown Prolog predicates--member and append--while looking at list processing from both a recursive and a procedural standpoint. After that, we introduce findall, a Visual Prolog standard predicate that enables you to find and collect all solutions to a single internal goal. We round out this chapter with a discussion of compound lists--combinations of different types of elements--and an example of parsing by difference lists.

What Is a List? In Prolog, a list is an object that contains an arbitrary number of other objects within it. Lists correspond roughly to arrays in other languages, but, unlike an array, a list does not require you to declare how big it will be before you use it. There are other ways to combine several objects into one, of course. If the number of objects to be combined is known in advance, you can make them the arguments of a single compound data structure. And even if the number of objects is unpredictable, you can use a recursive compound data structure, such as a tree. But lists are usually easier to use because the language provides a concise notation for them. A list that contains the numbers 1, 2, and 3 is written as [ 1, 2, 3 ]

Each item contained in the list is known as an element. To form a list data structure, you separate the elements of a list with commas and then enclose them in square brackets. Here are some examples: [dog, cat, canary] ["valerie ann", "jennifer caitlin", "benjamin thomas"]

Chapter 91, Repetition and recursion

153

Declaring Lists To declare the domain for a list of integers, you use the domains declaration, like this: DOMAINS integerlist = integer*

The asterisk means "list of"; that is, integer* means "list of integers." Note that the word list has no special meaning in Visual Prolog. You could equally well have called your list domain zanzibar. It's the asterisk, not the name, that signifies a list domain. The elements in a list can be anything, including other lists. However, all elements in a list must belong to the same domain, and in addition to the declaration of the list domain there must be a domains declaration for the elements: DOMAINS elementlist = elements* elements = ....

Here elements must be equated to a single domain type (for example: integer, real, or symbol) or to a set of alternatives marked with different functors. Visual Prolog does not allow you to mix standard types in a list. For example, the following declarations would not properly indicate a list made up of integers, reals, and symbols: elementlist = elements* elements = integer; real; symbol

/* Incorrect */

The way to declare a list made up of integers, reals, and symbols is to define a single domain comprising all three types, with functors to show which type a particular element belongs to. For example: elementlist = elements* elements = i(integer); r(real); s(symbol) /* the functors are i, r, and s */

(For more information about this, refer to "Compound Lists" later in this chapter.) Heads and Tails A list is really a recursive compound object. It consists of two parts: the head, of list which is the first element, and the tail, which is a list comprising all the

subsequent elements. The tail of a list is always a list; the head of a list is an element. For example, the head of [a, b, c] is a the tail of [a, b, c] is [b,

What happens when you get down to a one-element list? The answer is that the head of [c] is c the tail of [c] is [] If you take the first element from the tail of a list enough times, you'll eventually get down to the empty list ([ ]). The empty list can't be broken into head and tail. This means that, conceptually, lists have a tree structure just like other compound objects. The tree structure of [a, b, c, d] is list

\ list / \ b list / \ c list / \ d []

Further, a one-element list such as [a] is not the same as the element that it contains because, simple as it looks, [a] is really the compound data structure shown here: list

\ []

List Processing Prolog provides a way to make the head and tail of a list explicit. Instead of separating elements with commas, you can separate the head and tail with a vertical bar (|). For instance, [a, b, c]

is equivalent to [a|[b, c]]

and, continuing the process,

Chapter 91, Repetition and recursion

155

is equivalent to [a|[b|[c]]] which is equivalent to [a|[b|[c|[]]]] [a|[b, c]]

You can even use both kinds of separators in the same list, provided the vertical bar is the last separator. So, if you really want to, you can write [a, b, c, d] as [a, b|[c, d]]. Table 124.1 gives more examples. Table 125.2: Heads and Tails of Lists1263 List

Head

Tail

['a', 'b', 'c']

'a'

['b', 'c']

[ 'a' ]

'a'

[] /* an empty list */

[ ]

undefined

[[1, 2, 3], [2, 3, 4], []]

[1, 2, 3]

[[2, 3, 4], []]

Table 127.4 gives several examples of list unification. Table 128.5: Unification of Lists1296 List 1

List 2

Variable Binding

[X, Y, Z]

[egbert, eats, icecream]

X=egbert, Y=eats, Z=icecream

[7]

[X | Y]

X=7, Y=[]

[1, 2, 3, 4]

[X, Y | Z]

X=1, Y=2, Z=[3,4]

[1, 2]

[3 | X]

fail

Using Lists Because a list is really a recursive compound data structure, you need recursive algorithms to process it. The most basic way to process a list is to work through it, doing something to each element until you reach the end. An algorithm of this kind usually needs two clauses. One of them says what to do with an ordinary list (one that can be divided into a head and a tail). The other says what to do with an empty list.

Writing Lists For example, if you just want to print out the elements of the list, here's what you do: /* Program ch0130e01.pro */ DOMAINS list = integer*

/* or whatever type you wish to use */

PREDICATES write_a_list(list) CLAUSES write_a_list([]).

/* If the list is empty, do nothing more. */

write_a_list([H|T]):/* Match the head to H and the tail to T, then... */ write(H),nl, write_a_list(T). GOAL write_a_list([1, 2, 3]).

Here are the two write_a_list clauses described in natural language: To write an empty list, do nothing. Otherwise, to write a list, write its head (which is a single element), then write its tail (a list).

The first time through, the goal is: write_a_list([1, 2, 3]).

This matches the second clause, with H=1 and T=[2, calls write_a_list recursively with the tail of the list: write_a_list([2, 3]).

3];

this writes 1 and then

/* This is write_a_list(T). */

This recursive call matches the second clause, this time with H=2 and T=[3], so it writes 2 and again calls write_a_list recursively: write_a_list([3]).

Chapter 91, Repetition and recursion

157

Now, which clause will this goal match? Recall that, even though the list [3] has only one element, it does have a head and tail; the head is 3 and the tail is []. So again the goal matches the second clause, with H=3 and T=[]. Hence, 3 is written and write_a_list is called recursively like this: write_a_list([]).

Now you see why this program needs the first clause. The second clause won't match this goal because [] can't be divided into head and tail. So, if the first clause weren't there, the goal would fail. As it is, the first clause matches and the goal succeeds without doing anything further. Exercise Is write_a_list tail-recursive? Would it be if the two clauses were written in the opposite order?

Counting List Elements Now consider how you might find out how many elements are in a list. What is the length of a list, anyway? Here's a simple logical definition: The length of [] is 0. The length of any other list is 1 plus the length of its tail. Can you implement this? In Prolog it's very easy. It takes just two clauses: /* Program ch0131e02.pro */ DOMAINS list = integer*

/* or whatever type you want to use */

PREDICATES length_of(list,integer) CLAUSES length_of([], 0). length_of([_|T],L):length_of(T,TailLength), L = TailLength + 1.

Take a look at the second clause first. Crucially, [_|T] will match any nonempty list, binding T to the tail of the list. The value of the head is unimportant; as long as it exists, it can be counted it as one element. So the goal:

length_of([1, 2, 3], L)

will match the second clause, with T=[2, 3]. The next step is to compute the length of T. When this is done (never mind how), TailLength will get the value 2, and the computer can then add 1 to it and bind L to 3. So how is the middle step executed? That step was to find the length of [2, 3] by satisfying the goal length_of([2, 3], TailLength).

In other words, length_of calls itself recursively. This goal matches the second clause, binding [3]

in the goal to T in the clause and

TailLength

in the goal to L in the clause.

Recall that TailLength in the goal will not interfere with TailLength in the clause, because each recursive invocation of a clause has its own set of variables. If this is unclear, review the section on recursion in chapter 132. So now the problem is to find the length of [3], which will be 1, and then add 1 to that to get the length of [2, 3], which will be 2. So far, so good. Likewise, length_of will call itself recursively again to get the length of [3]. The tail of [3] is [], so T is bound to [], and the problem is to get the length of [], then add 1 to it, giving the length of [3]. This time it's easy. The goal length_of([], TailLength)

matches the first clause, binding TailLength to 0. So now the computer can add 1 to that, giving the length of [3], and return to the calling clause. This, in turn, will add 1 again, giving the length of [2, 3], and return to the clause that called it; this original clause will add 1 again, giving the length of [1, 2, 3]. Confused yet? We hope not. In the following brief illustration we'll summarize the calls. We've used subscripts to indicate that similarly-named variables in different clauses--or different invocations of the same clause--are distinct. length_of([1, 2, 3], L1). length_of([2, 3], L2). length_of([3], L3). length_of([], 0). L3 = 0+1 = 1. L2 = L3+1 = 2. Chapter 91, Repetition and recursion

159

L1 = L2+1 = 3. Exercises What happens when you satisfy the following goal? length_of(X, 3), !.

Does the goal succeed, and if so, what is bound to X? Why? (Work through carefully by hand to see how this works.) Write a predicate called sum_of that works exactly like length_of, except that it takes a list of numbers and adds them up. For example, the goal: sum_of([1, 2, 3, 4], S).

should bind S to 10. What happens if you execute this goal? sum_of(List, 10).

This goal says, "Give me a list whose elements add up to 10." Can Visual Prolog do this? If not, why not? (Hint: It's not possible to do arithmetic on unbound variables in Prolog.)

Tail Recursion Revisited You probably noticed that length_of is not, and can't be, tail-recursive, because the recursive call is not the last step in its clause. Can you create a tail-recursive list-length predicate? Yes, but it will take some effort. The problem with length_of is that you can't compute the length of a list until you've already computed the length of the tail. It turns out there's a way around this. You'll need a list-length predicate with three arguments. One is the list, which the computer will whittle away on each call until it eventually becomes empty, just as before. Another is a free argument that will ultimately contain the result (the length). The third is a counter that starts out as 0 and increments on each call. When the list is finally empty, you'll unify the counter with the (up to then) unbound result.

/* Program ch0133e03.pro */ DOMAINS list = integer*

/* or whatever type you want to use */

PREDICATES length_of(list,integer,integer) CLAUSES length_of([], Result, Result). length_of([_|T],Result,Counter):NewCounter = Counter + 1, length_of(T, Result, NewCounter). GOAL length_of([1, 2, 3], L, 0), write("L=",L), nl.

/* start with Counter = 0 */

This version of the length_of predicate is more complicated, and in many ways less logical, than the previous one. We've presented it merely to show you that, by devious means, you can often find a tail-recursive algorithm for a problem that seems to demand a different type of recursion. Exercises Try both versions of length_of on enormous lists (lists with perhaps 200 to 500 elements). What happens? On long lists, how do they compare in speed? What happens with the new version of length_of if you give the following goal? length_of(MyList, 5, 0).

Hint: You are discovering a very important property of Prolog called interchangeability of unknowns. Not all Prolog predicates have it. Rewrite sum_of to work like the new version of length_of. Another Example -- Modifying the List Sometimes you want to take a list and create another list from it. You do this by working through the list element by element, replacing each element with a computed value. For example, here is a program that takes a list of numbers and adds 1 to each of them: /* Program ch0134e04.pro */ DOMAINS list = integer*

Chapter 91, Repetition and recursion

161

PREDICATES add1(list,list) CLAUSES add1([], []). /* boundary condition */ add1([Head|Tail],[Head1|Tail1]):Head1= Head+1 add1(Tail,Tail1).

/* separate the head /* from the rest of the list /* add 1 to the first element /* call element with the rest of the list

*/ */ */ */

To paraphrase this in natural language: To add 1 to all the elements of the empty list, just produce another empty list. To add 1 to all the elements of any other list, add 1 to the head and make it the head of the result, and then add 1 to each element of the tail and make that the tail of the result.

Load the program, and enter the goal add1([1,2,3,4],

NewList).

Visual Prolog will return NewList=[2,3,4,5] 1 Solution

Tail Recursion Again Is add1 tail-recursive? If you're accustomed to using Lisp or Pascal, you might think it isn't, because you think of it as performing the following operations: Split the list into Head and Tail. Add 1 to Head, giving Head1. Recursively add 1 to all the elements of Tail, giving Tail1. Combine Head1 and Tail1, giving the resulting list. This isn't tail-recursive, because the recursive call is not the last step. But--and this is important--that is not how Prolog does it. In Visual Prolog, add1 is tail-recursive, because its steps are really the following: Bind the head and tail of the original list to Head and Tail. Bind the head and tail of the result to Head1 and Tail1. (Head1 and Tail1 do not have values yet.)

Add 1 to Head, giving Head1. Recursively add 1 to all the elements of Tail, giving Tail1. When this is done, Head1 and Tail1 are already the head and tail of the result; there is no separate operation of combining them. So the recursive call really is the last step. More on Modifying Lists Of course you don't actually need to put in a replacement for every element. Here's a program that scans a list of numbers and copies it, leaving out the negative numbers: /* Program ch0135e05.pro */ DOMAINS list = integer* PREDICATES discard_negatives(list, list) CLAUSES discard_negatives([], []). discard_negatives([H|T],ProcessedTail):H < 0, /* If H is negative, just skip it */ !, discard_negatives(T, ProcessedTail). discard_negatives([H|T],[H|ProcessedTail]):discard_negatives(T, ProcessedTail).

For example, the goal discard_negatives([2, -45, 3, 468], X)

gives X=[2, 3, 468]. And here's a predicate that copies the elements of a list, making each element occur twice: doubletalk([], []). doubletalk([H|T], [H, H|DoubledTail]) :doubletalk(T, DoubledTail).

Chapter 91, Repetition and recursion

163

List Membership Suppose you have a list with the names John, Leonard, Eric, and Frank and would like to use Visual Prolog to investigate if a given name is in this list. In other words, you must express the relation "membership" between two arguments: a name and a list of names. This corresponds to the predicate member(name, namelist).

/* "name" is a member of "namelist" */

In Program 6, the first clause investigates the head of the list. If the head of the list is equal to the name you're searching for, then you can conclude that Name is a member of the list. Since the tail of the list is of no interest, it is indicated by the anonymous variable. Thanks to this first clause, the goal member(john, [john, leonard, eric, frank])

is satisfied. /* Program ch0136e06.pro */ DOMAINS namelist = name* name = symbol PREDICATES nondeterm member(name, namelist) CLAUSES member(Name, [Name|_]). member(Name, [_|Tail]):member(Name,Tail).

If the head of the list is not equal to Name, you need to investigate whether Name can be found in the tail of the list. In English: Name is a member of the list if Name is the first element of the list, or Name is a member of the list if Name is a member of the tail.

The second clause of member relates to this relationship. In Visual Prolog: member(Name, [_|Tail]) :- member(Name, Tail).

Exercises Load Program 6 and try the following goal:

member(susan, [ian, susan, john]).

Add domain and predicate statements so you can use member to investigate if a number is a member of a list of numbers. Try several goals, including member(X, [1, 2, 3, 4]).

to test your new program. Does the order of the two clauses for the member predicate have any significance? Test the behavior of the program when the two rules are swapped. The difference appears if you test the goal member(X, [1, 2, 3, 4, 5])

in both situations.

Appending One List to Another: Declarative and Procedural Programming As given, the member predicate of Program 6 works in two ways. Consider its clauses once again: member(Name, [Name|_]). member(Name, [_|Tail]) :- member(Name, Tail).

You can look at these clauses from two different points of view: declarative and procedural. From a declarative viewpoint, the clauses say Name

is a member of a list if the head is equal to Name; if not, Name is a member of the list if it is a member of the tail.

From a procedural viewpoint, the two clauses could be interpreted as saying: To find a member of a list, find its head; otherwise, find a member of its tail.

These two points of view correspond to the goals member(2, [1, 2, 3, 4]).

and member(X, [1, 2, 3, 4]).

In effect, the first goal asks Visual Prolog to check whether something is true; the second asks Visual Prolog to find all members of the list [1,2,3,4]. Don't be

Chapter 91, Repetition and recursion

165

confused by this. The member predicate is the same in both cases, but its behavior may be viewed from different angles. Recursion from a Procedural Viewpoint The beauty of Prolog is that, often, when you construct the clauses for a predicate from one point of view, they'll work from the other. To see this duality, in this next example you'll construct a predicate to append one list to another. You'll define the predicate append with three arguments: append(List1, List2, List3)

This combines List1 and List2 to form List3. Once again you are using recursion (this time from a procedural point of view). If List1 is empty, the result of appending List1 and List2 will be the same as List2. In Prolog: append([], List2, List2).

If List1 is not empty, you can combine List1 and List2 to form List3 by making the head of List1 the head of List3. (In the following code, the variable H is used as the head of both List1 and List3.) The tail of List3 is L3, which is composed of the rest of List1 (namely, L1) and all of List2. In Prolog: append([H|L1], List2, [H|L3]) :append(L1, List2, L3).

The append predicate operates as follows: While List1 is not empty, the recursive rule transfers one element at a time to List3. When List1 is empty, the first clause ensures that List2 hooks onto the back of List3. Exercise The predicate append is defined in Program 7. Load the program. /* Program ch0137e07.pro */ DOMAINS integerlist = integer* PREDICATES append(integerlist,integerlist,integerlist) CLAUSES append([],List,List). append([H|L1],List2,[H|L3]):append(L1,List2,L3).

Now run it with the following goal: append([1, 2, 3], [5, 6], L).

Now try this goal: append([1, 2], [3], L), append(L, L, LL).

One Predicate Can Have Different Uses Looking at append from a declarative point of view, you have defined a relation between three lists. This relation also holds if List1 and List3 are known but List2 isn't. However, it also holds true if only List3 is known. For example, to find which two lists could be appended to form a known list, you could use a goal of the form append(L1, L2, [1, 2, 4]).

With this goal, Visual Prolog will find these solutions: L1=[], L2=[1,2,4] L1=[1], L2=[2,4] L1=[1,2], L2=[4] L1=[1,2,4], L2=[] 4 Solutions

You can also use append to find which list you could append to [3,4] to form the list [1,2,3,4]. Try giving the goal append(L1, [3,4], [1,2,3,4]).

Visual Prolog finds the solution L1=[1,2].

This append predicate has defined a relation between an input set and an output set in such a way that the relation applies both ways. Given that relation, you can ask Which output corresponds to this given input?

or Which input corresponds to this given output?

The status of the arguments to a given predicate when you call that predicate is referred to as a flow pattern. An argument that is bound or instantiated at the time

Chapter 91, Repetition and recursion

167

of the call is an input argument, signified by (i); a free argument is an output argument, signified by (o). The append predicate has the ability to handle any flow pattern you provide. However, not all predicates have the capability of being called with different flow patterns. When a Prolog clause is able to handle multiple flow patterns, it is known as an invertible clause. When writing your own Visual Prolog clauses, keep in mind that an invertible clause has this extra advantage and that creating invertible clauses adds power to the predicates you write. Exercise Amend the clauses defining member in Program 6 and construct the clauses for a predicate even_member that will succeed if you give the goal even_member(2, [1, 2, 3, 4, 5, 6]).

The predicate should also display the following result: X=2 X=4 X=6 3 Solutions

given the goal even_member(X, [1, 2, 3, 4, 5, 6]).

Finding All the Solutions at Once In chapter 138, we compared backtracking and recursion as ways to perform repetitive processes. Recursion won out because, unlike backtracking, it can pass information (through arguments) from one recursive call to the next. Because of this, a recursive procedure can keep track of partial results or counters as it goes along. But there's one thing backtracking can do that recursion can't do--namely, find all the alternative solutions to a goal. So you may find yourself in a quandary: You need all the solutions to a goal, but you need them all at once, as part of a single compound data structure. What do you do? Fortunately, Visual Prolog provides a way out of this impasse. The built-in predicate findall takes a goal as one of its arguments and collects all of the solutions to that goal into a single list. findall takes three arguments:

The first argument, VarName, specifies which argument in the specified predicate is to be collected into a list. The second, mypredicate, indicates the predicate from which the values will be collected. The third argument, ListParam, is a variable that holds the list of values collected through backtracking. Note that there must be a user-defined domain to which the values of ListParam belong. Program 8 uses findall to print the average age of a group of people. /* Program ch0139e08.pro */ DOMAINS name,address = string age = integer list = age* PREDICATES nondeterm person(name, address, age) sumlist(list, age, integer) CLAUSES sumlist([],0,0). sumlist([H|T],Sum,N):sumlist(T,S1,N1), Sum=H+S1, N=1+N1. person("Sherlock Holmes", "22B Baker Street", 42). person("Pete Spiers", "Apt. 22, 21st Street", 36). person("Mary Darrow", "Suite 2, Omega Home", 51). GOAL findall(Age,person(_, _, Age),L), sumlist(L,Sum,N), Ave = Sum/N, write("Average=", Ave),nl.

The findall clause in this program creates a list, L, that is a collection of all the ages obtained from the predicate person. If you wanted to collect a list of all the people who are 42 years old, you could give the following subgoal: findall(Who, person(Who, _, 42), List)

Before trying this, please note that it requires the program to contain a domain declaration for the resulting list: slist = string*

Chapter 91, Repetition and recursion

169

Compound Lists A list of integers can be simply declared as integerlist = integer*

The same is true for a list of real numbers, a list of symbols, or a list of strings. However, it is often valuable to store a combination of different types of elements within a list, such as: [2, 3, 5.12, ["food", "goo"], "new"] /* Not correct Visual Prolog*/

Compound lists are lists that contain more than one type of element. You need special declarations to handle lists of multiple-type elements, because Visual Prolog requires that all elements in a list belong to the same domain. The way to create a list in Prolog that stores these different types of elements is to use functors, because a domain can contain more than one data type as arguments to functors. The following is an example of a domain declaration for a list that can contain an integer, a character, a string, or a list of any of these: DOMAINS /* the functors are l, i, c, and s */ llist = l(list); i(integer); c(char); s(string) list = llist*

The list [ 2, 9, ["food", "goo"], "new" ]

/* Not correct Visual Prolog */

would be written in Visual Prolog as [i(2), i(9), l([s("food"), s("goo")]), s("new")] /* Correct Visual Prolog */

The following example of append shows how to use this domain declaration in a typical list-manipulation program. /* Program ch0140e09.pro */ DOMAINS llist = l(list); i(integer); c(char); s(string) list = llist* PREDICATES append(list,list,list)

CLAUSES append([],L,L). append([X|L1],L2,[X|L3]):append(L1, L2, L3). GOAL append([s(likes), l([s(bill), s(mary)])],[s(bill), s(sue)],Ans), write("FIRST LIST: ", Ans,"\n\n"), append([l([s("This"),s("is"),s("a"),s("list")]),s(bee)], [c('c')],Ans2), write("SECOND LIST: ", Ans2, '\n').

Exercises Write a predicate, oddlist, that takes two arguments. The first argument is a list of integers, while the second argument returns a list of all the odd numbers found in the first list. Write a predicate, real_average, that calculates the average value of all the elements in a list of reals. Write a predicate that takes a compound list as its first argument and returns a second argument that is the list with all the sub-lists removed. This predicate is commonly known as flatten, as it flattens a list of lists into a single list. For example, the call flatten([s(ed), i(3), l([r(3.9), l([s(sally)])])], r(4.21), X)

returns X = [s(ed), i(3), r(3.9), s(sally), r(4.21)] 1 Solution

which is the original list, flattened. Parsing by Difference Lists Program 10 demonstrates parsing by difference lists. The process of parsing by difference lists works by reducing the problem; in this example we transform a string of input into a Prolog structure that can be used or evaluated later. The parser in this example is for a very primitive computer language. Although this example is very advanced for this point in the tutorial, we decided to put it here because parsing is one of the areas where Visual Prolog is very powerful. If you do not feel ready for this topic, you can skip this example and continue on in the tutorial without any loss of continuity.

Chapter 91, Repetition and recursion

171

/* Program ch0141e10.pro */ DOMAINS toklist

= string*

PREDICATES tokl(string,toklist) CLAUSES tokl(Str,[H|T]):fronttoken(Str,H,Str1),!, tokl(Str1,T). tokl(_,[]). /* * * * * * * * * * * * * * * * * * * * * * * * * * This second part of the program is the parser * * * * * * * * * * * * * * * * * * * * * * * * * */ DOMAINS program = program(statementlist) statementlist = statement* /* * * * * * * * * * * * * * * * * * * * * * * * * Definition of what constitutes a statement * * * * * * * * * * * * * * * * * * * * * * * * */ statement = if_Then_Else(exp,statement,statement); if_Then(exp,statement); while(exp,statement); assign(id,exp) /* * * * * * * * * * * * * * * * Definition of expression * * * * * * * * * * * * * * * */ exp

= plus(exp,exp); minus(exp,exp); var(id); int(integer)

= string

PREDICATES s_program(toklist,program) s_statement(toklist,toklist,statement) s_statementlist(toklist,toklist,statementlist) s_exp(toklist,toklist,exp) s_exp1(toklist,toklist,exp,exp) s_exp2(toklist,toklist,exp)

CLAUSES s_program(List1,program(StatementList)):s_statementlist(List1,List2,StatementList), List2=[]. s_statementlist([],[],[]):-!. s_statementlist(List1,List4,[Statement|Program]):s_statement(List1,List2,Statement), List2=[";"|List3], s_statementlist(List3,List4,Program). s_statement(["if"|List1],List7,if_then_else(Exp,Statement1, Statement2)):s_exp(List1,List2,Exp), List2=["then"|List3], s_statement(List3,List4,Statement1), List4=["else"|List5],!, s_statement(List5,List6,Statement2), List6=["fi"|List7]. s_statement(["if"|List1],List5,if_then(Exp,Statement)):-!, s_exp(List1,List2,Exp), List2=["then"|List3], s_statement(List3,List4,Statement), List4=["fi"|List5]. s_statement(["do"|List1],List4,while(Exp,Statement)):-!, s_statement(List1,List2,Statement), List2=["while"|List3], s_exp(List3,List4,Exp). s_statement([ID|List1],List3,assign(Id,Exp)):isname(ID), List1=["="|List2], s_exp(List2,List3,Exp). s_exp(LIST1,List3,Exp):s_exp2(List1,List2,Exp1), s_exp1(List2,List3,Exp1,Exp). s_exp1(["+"|List1],List3,Exp1,Exp):-!, s_exp2(List1,List2,Exp2), s_exp1(List2,List3,plus(Exp1,Exp2),Exp). s_exp1(["-"|List1],List3,Exp1,Exp):-!, s_exp2(List1,List2,Exp2), s_exp1(List2,List3,minus(Exp1,Exp2),Exp). s_exp1(List,List,Exp,Exp).

Chapter 91, Repetition and recursion

173

s_exp2([Int|Rest],Rest,int(I)):str_int(Int,I),!. s_exp2([Id|Rest],Rest,var(Id)):isname(Id).

Load and run this program, then enter the following goal: Goal tokl("b=2; if b then a=1 else a=2 fi; do a=a-1 while a;",Ans), s_program(Ans,Res).

Visual Prolog will return the program structure: Ans=["b","=","2",";","if","b","then","a","=","1", "else","a","=","2","fi",";","do","a","=","a", "-","1","while","a",";" ], Res=program([assign("b",int(2)), if_then_else(var("b"),assign("a",int(1)), assign("a",int(2))), while(var("a"),assign("a",minus(var("a"),int(1)))) ]) 1 Solution

The transformation in this example is divided into two stages: scanning and parsing. The tokl predicate is the scanner; it accepts a string and converts it into a list of tokens. All the predicates with names beginning in s_ are parser predicates. In this example the input text is a Pascal-like program made up of Pascal-like statements. This programming language only understands certain statements: IF THEN ELSE, IF THEN, DO WHILE, and ASSIGNMENT. Statements are made up of expressions and other statements. Expressions are addition, subtraction, variables, and integers. Here's how this example works: The first scanner clause, s_program, takes a list of tokens and tests if it can be transformed into a list of statements. The predicate s_statementlist takes this same list of tokens and tests if the tokens can be divided up into individual statements, each ending with a semicolon. The predicate s_statement tests if the first tokens of the token list make up a legal statement. If so, the statement is returned in a structure and the remaining tokens are returned back to s_statementlist.

The four s_statement clauses correspond to the four types of statements the parser understands. If the first s_statement clause is unable to transform the list of tokens into an IF THEN ELSE statement, the clause fails and backtracks to the next s_statement clause, which tries to transform the list of tokens into an IF THEN statement. If that clause fails, the next one tries to transform the list of tokens into a DO WHILE statement. If the first three s_statement clauses fail, the last clause for that predicate tests if the statement does assignment. This clause tests for assignment by testing if the first term is a symbol, the second term is "=", and the next terms make up a simple math expression. The s_exp, s_exp1, and s_exp2 predicates work the same way, by testing if the first terms are expressions and--if so--returning the remainder of the terms and an expression structure back to s_statement. See the Sentence Analyzer VPI\PROGRAMS\SEN_AN program on your disk for a more detailed example of parsing natural-language.

Summary These are the important points covered in this chapter: Lists are objects that can contain an arbitrary number of elements; you declare them by adding an asterisk at the end of a previously defined domain. A list is a recursive compound object that consists of a head and a tail. The head is the first element and the tail is the rest of the list (without the first element). The tail of a list is always a list; the head of a list is an element. A list can contain zero or more elements; the empty list is written []. The elements in a list can be anything, including other lists; all elements in a list must belong to the same domain. The domain declaration for the elements must be of this form: DOMAINS elementlist = elements* elements = ....

where elements = one of the standard domains (integer, real, etc.) or a set of alternatives marked with different functors (int(integer); rl(real); smb(symbol); etc.). You can only mix types in a list in Visual Prolog by enclosing them in compund objects/functors. You can use separators (commas, [, and |) to make the head and tail of a list explicit; for example, the list

Chapter 91, Repetition and recursion

175

[a, b, c, d]

can be written as [a|[b, c, d]] or [a, b|[c, d]] or [a, b, c|[d]] or [a|[b|[c, d]]] or [a|[b|[c|[d]]]] or even [a|[b|[c|[d|[]]]]]

List processing consists of recursively removing the head of the list (and usually doing something with it) until the list is an empty list. The classic Prolog list-handling predicates member and append enable you to check if an element is in a list and check if one list is in another (or append one list to another), respectively. A predicate's flow pattern is the status of its arguments when you call it; they can be input parameters (i)--which are bound or instantiated--or output parameters (o), which are free. Visual Prolog provides a built-in predicate, findall, which takes a goal as one of its arguments and collects all of the solutions to that goal into a single list. Because Visual Prolog requires that all elements in a list belong to the same domain, you use functors to create a list that stores different types of elements. The process of parsing by difference lists works by reducing the problem; the example in this chapter transforms a string of input into a Prolog structure that can be used or evaluated later.

CHAPTER

142

Visual Prologs fact sections 143Visual PrologVisual PrologIn this chapter, we describe how you declare facts sections and how you can modify the contents of your fact section. A facts-section is composed of facts that you can add directly into and remove from your Visual Prolog program at run time. You declare the predicates describing the facts section in the facts section of your program, and you use these predicates the same way you use the ones declared in the predicates section.

In Visual Prolog, you use the the predicates assert, asserta, assertz to add new facts to the facts section, and the predicates retract and retractall to remove existing facts. You can modify the contents of your facts section by first retracting a fact and then asserting the new version of that fact (or a different fact altogether). The predicate consult reads facts from a file and asserts them into the internal facts, and save saves the contents of an internal facts section to a file. Visual Prolog treats facts belonging to facts sections differently from the way it treats normal predicates. Facts for the facts section predicates are kept in tables, which are easy to modify, while the normal predicates are compiled to binary code for maximum speed.

Declaring the facts-sections The keyword facts or (database) marks the beginning of a sequence of declarations for predicates describing an facts-section. You can add facts--but not rules--to a facts-section from the keyboard at run time with asserta and assertz. Or, by calling the standard predicate consult, you can retrieve the added facts from a disk file. The facts section looks something like the following example. DOMAINS name, address = string age = integer gender = male ; female FACTS person(name, address, age, gender) PREDICATES male(name, address, age) female(name, address, age) child(name, age, gender) CLAUSES male(Name, Address, Age) :person(Name, Address, Age, male).

In this example, you can use the predicate person the same way you use the other predicates, (male, female, child); the only difference is that you can insert and remove facts for person while the program is running. There are two restrictions on using predicates in facts sections: 1. You can add them into the factssection as facts only--not as rules. 2. Facts in factssections may not have free variables.

Chapter 91, Repetition and recursion

177

It is possible to have several facts sections, but in order to do this, you must explicitly name each facts section. FACTS - mydatabase myFirstRelation(integer) mySecondRelation(real, string) myThirdRelation(string) /* etc. */

This declaration creates a factssection with the name mydatabase. If you don't supply a name for an facts database, it defaults to the standard name dbasedom. The names of predicates in a facts section must be unique within a module (source file), you can't use the same predicate name in two different facts sections. However, the predicates in the named facts sections are private to the module in which they're declared, and won't interfere with predicates in other modules. Modules are explained in the chapter 144.

Using the facts sections Because Visual Prolog represents a relational facts sections as a collection of facts, you can use it as a powerful query language for databases. Visual Prolog's unification algorithm automatically selects facts with the correct values for the known arguments and assigns values to any unknown arguments, while its backtracking algorithm can give all the solutions to a given query.

Accessing the facts sections Predicates belonging to a facts section are accessed in exactly the same way as other predicates. The only visible difference in your program is that the declarations for the predicates are in a facts section instead of a predicates section. Given for instance the following: DOMAINS name = string sex = char FACTS person(name,sex) CLAUSES person("Helen",'F'). person("Maggie",'F'). person("Suzanne",'F'). person("Per",'M').

you can call person with the goal person(Name,'F') to find all women, or person("Maggie",'F') to verify that there is a woman called Maggie in your facts section. You should be aware that, by their very nature, predicates in factssections are always nondeterministic. Because facts can be added anytime at run time, the compiler must always assume that it's possible to find alternative solutions during backtracking. If you have a predicate in a factssection for which you'll never have more than one fact, you can override this by prefacing the declaration with the compiler directive determ to the declaration: FACTS determ daylight_saving(integer)

You will get an error if you try to add a fact for a deterministic database predicate which already has a fact.

Updating the facts section Facts for database predicates can be specified at compile time in the clauses section, as in the example above. At run time, facts can be added and removed by using the predicates described below. Note that facts specified at compile time in the clauses section can be removed too, they're not different from facts added at run time. Visual Prolog's standard database predicates assert, asserta, assertz, retract, retractall, consult, and save will all take one or two arguments. The optional second argument is the name of an facts section. We describe these predicates in the following pages. The notation "/1" and "/2" after each predicate name refers to the number of arguments that arity version of the predicate takes. The comments after the formats (such as /* (i) */ and /* (o,i) */ show the flow pattern(s) for that predicate. Adding Facts at Run Time At run time, facts can be added to the facts sections with the predicates: assert, asserta and assertz, or by loading a file of facts with consult. There are three predicates to add a single fact at runtime: asserta(< the

fact > )

asserta(< the

fact > ,

assertz(< the

fact > )

assertz(< the

fac t > ,

/* (i) */ facts_sectionName)

/* (i, i) */ /* (i) */

facts_sectionName)

Chapter 91, Repetition and recursion

/* (i, i) */

179

assert(< the

fact > )

assert(< the

fact > ,

/* (i) */ facts_sectionName)

/* (i, i) */

asserta asserts a new fact into the facts section before the existing facts for the given predicate, while assertz asserts a new fact after the existing facts for that predicate. assert behaves like assertz. The assertion predicates always know which facts section to insert the fact in, because the names of the facts section predicates are unique within a program or module. However, you can use the optional second argument for type-checking purposes in order to ensure that you are working on the correct facts section. The first of the following goals inserts a fact about Suzanne for the person predicate, after all the facts for person currently stored in the facts section. The second inserts a fact about Michael before all the currently-stored facts for person. The third inserts a fact about John after all the other likes facts in the facts section likesDatabase, while the fourth inserts a fact about Shannon in the same facts section, before all the other likes facts. assertz(person("Suzanne", "New Haven", 35)). asserta(person("Michael", "New York", 26)). assertz(likes("John", "money"), likesDatabase). asserta(likes("Shannon", "hard work"), likesDatabase).

After these calls the facts sections look as if you'd started with the following facts: /* Facts section -- dbasedom */ person("Michael", "New York", 26). /* ... other person facts ... */ person("Suzanne", "New Haven", 35). /* Facts section -- likesDatabase */ likes("Shannon", "hard work"). /* ... other likes facts ... */ likes("John", "money").

Be careful that you don't accidentally write code asserting the same fact twice. The facts sections don't impose any kind of uniqueness, and the same fact may appear many times in a facts section. However, a uniqueness-testing version of assert is very easy to write: FACTS - people person(string,string) PREDICATES uassert(people)

CLAUSES uassert(person(Name,Address)):person(Name,Address), ! ; assert(person(Name,Address)).

% OR

Loading Facts from a File at Run Time consult reads in a file, fileName, containing facts declared in a facts section and asserts them at the end) of the appropriate facts section. consult takes one or two arguments: consult(fileName) /* (i) */ consult(fileName, databaseName) /* (i, i) */

Unlike assertz, if you call consult with only one argument (no facts section name), it will only read facts that were declared in the default dbasedom facts section section. If you call consult with two arguments (the file name and a facts section name), it will only consult facts from that named facts section. If the file contains anything other than a fact belonging to the specified facts section, consult will return an error when it reaches that part of the file. Keep in mind that the consult predicate reads one fact at a time; if the file has ten facts, and the seventh fact has some syntax error, consult will insert the first six facts into the facts section--then issue an error message. Note that consult is only able to read a file in exactly the same format as save generates (in order to insert facts as fast as possible). There can be no upper-case characters except in double-quoted strings no spaces except in double-quoted strings no comments no empty lines no symbols without double quotes You should be careful when modifying or creating such a file of facts with an editor. Removing Facts at Run Time retract unifies facts and removes them from the facts sections. It's of the following form: retract(<the

fact > [,

databaseName])

Chapter 91, Repetition and recursion

/* (i, i) */

181

retract will remove the first fact in your facts section that matches <the fact>, instantiating any free variables in <the fact> in the process. Retracting facts from a facts section is exactly like accessing it, with the side-effect that the matched fact is removed. Unless the facts section predicate accessed by retract was declared to be deterministic, retract is nondeterministic and will, during backtracking, remove and return the remaining matching facts, one at a time. When all matching facts have been removed, retract fails. Suppose you have the following facts sections in your program: DATABASE person(string, string, integer) FACTS - likesDatabase likes(string, string) dislikes(string, string) CLAUSES person("Fred", "Capitola", 35). person("Fred", "Omaha", 37). person("Michael", "Brooklyn", 26). likes("John", "money"). likes("Jane", "money"). likes("Chris", "chocolate"). likes("John", "broccoli"). dislikes("Fred", "broccoli"). dislikes("Michael", "beer").

Armed with these facts sections, you give Visual Prolog the following goals: retract(person("Fred", _, _)). retract(likes(_, "broccoli")). retract(likes(_, "money"), likesDatabase). retract(person("Fred", _, _), likesDatabase)

/* /* /* /*

1 2 3 4

*/ */ */ */

The first goal retracts the first fact for person about Fred from the default dbasedom facts section. The second goal retracts the first fact matching likes(X, "broccoli") from the facts section likesDatabase. With both of these goals, Visual Prolog knows which facts section to retract from because the names of the facts section predicates are unique: person is only in the default facts section, and likes is only in the facts section likesDatabase. The third and fourth goals illustrate how you can use the optional second argument for type-checking. The third goal succeeds, retracting the first fact that matches likes(_, "money") from likesDatabase, but the fourth cannot be

compiled because there are (and can be) no person facts in the facts section likesDatabase. The error message given by the compiler is: 506 Type error: The functor does not belong to the domain.

The following goal illustrates how you can obtain values from retract: GOAL retract(person(Name, Age)), write(Name, ", ", Age), fail.

If you supply the name of a facts section section as the second argument to retract, you don't have to specify the name of the facts section predicate you're retracting from. In this case, retract will find and remove all facts in the specified facts section. Here's an example: GOAL retract(X, mydatabase), write(X), fail.

Removing Several Facts at Once retractall removes all facts that match <the fact> from your facts section, and is of the following form: retractall(<the

fact > [,

databaseName])

retractall behaves as if defined by retractall(X):- retract(X), fail. retractall(_).

but it's considerably faster than the above. As you can gather, retractall always succeeds exactly once, and you can't obtain output values from retractall. This means that, as with not, you must use underscores for free variables. As with assert and retract, you can use the optional second argument for typechecking. And, as with retract, if you call retractall with an underscore, it can remove all the facts from a given facts section. The following goal removes all the facts about males from a database of person facts: retractall(person(_, _, _, male)).

Chapter 91, Repetition and recursion

183

The next goal removes all the facts from the facts section mydatabase. retractall(_, mydatabase).

Facts determiner-keywords Facts can be declared with several optional keywords: NONDETERM determines that any number of instances of a fact fact_N can exist. This is default. DETERM time.

determines that only one instance of a fact fact_N can exist at any

GLOBAL

determines, that the facts section is global in the project.

SINGLE

determines, only one instance of a fact fact_N should always exist. Fact_N is the functors for the facts (predicates) belonging to this facts section.

NOCOPY

determins, that dates are not copied from the heap to the Visual Prolog Global Stack (GStack), when the fact is referenced. Normally, when calling a fact to bind a variable to a string or a compound object, the string or object is copied to the Gstack.

Discussions Facts declared with the keyword nondeterm. The keyword nondeterm is the default type for facts (database predicates) declared in facts sections. If none of the keywords determ or single are used in a fact declaration, the compiler applies nondeterm keyword. Normally, by their very nature, database predicates are non-deterministic. Because facts can be added anytime at runtime, the compiler must always assume that it's possible to find alternative solutions during backtracking. If you have a database predicate of which you'll never have more than one fact, you can override this by adding the keyword determ or single to the declaration. Facts declared with the keyword determ. The keyword determ determins that the facts database can only contain one instance of a fact (database predicate) fact_N(...) declared with this keyword. So if you try to assert one and then a second such fact into the database, the Visual Prolog engine will generate runtime error. (1041 Assert to a fact declared as

determ, but fact already exists). In such cases, programmer must take special care about this. Preceding a fact with determ enables the compiler to produce better code, and you will not get non-deterministic warnings for calling such a predicate. This is useful for flags, counters, and other things that are essentially global variables. Particularly note that when retracting a fact that is declared to be determ, the call to non-deterministic predicates retract/1 and retract/2 will be deterministic. So if you know that at any moment the facts section contains no more then one fact counter() then you can write: FACTS determ counter(integer CounterValue) GOAL ... retract(counter(CurrentCount)), /* here Prolog will not set backtracking point */ Count= CurrentCount + 1, assert(counter(Count)),

... instead of FACTS counter(integer CounterValue) PREDICATES determ my_retract(dbasedom) CLAUSES my_retract(X): - retract(X),!. % deterministic predicate GOAL ... my_retract(counter(CurrentCount)), /* here Prolog will not set backtracking point */ Count= CurrentCount + 1, asserta(counter(Count)), ...

Facts declared with the keyword single. The keyword single before a fact fact_N declaration determines that one and only one instance of a fact must always exist: Since single facts must be already known when the program calls Goal; therefore, single facts must be initialized in a clauses section in the program's source code. For example:

Chapter 91, Repetition and recursion

185

FACTS single singleFact(STRING, STRING) CLAUSES singleFact("","").

·Single facts cannot be retracted. If one try to apply any retract predicate to a single fact then the compiler will generates the error 249 "Attempt to retract a fact declared as single".·Since one instance of a single fact always exists, a single fact never fails if it is called with free arguments. For example, a following call singleFact(X,Y),

never fails if X and Y are free variables. Therefore, it is convenient to use single facts in procedures.assert, asserta, assertz, and consult predicates applied to a single fact act similarly to couples of retract and assert predicates. That is, assert (consult) predicates change the existing instance of a fact to the specified one. Preceding a fact with single enables the compiler to produce optimized code for accessing and updating of a fact. For example, for assert predicate applied to a single fact the compiler generates a code that works more effectively that a couple of retract and assert predicates applied to a determ fact (and all the more so then retract and assert predicates applied to a nondeterm fact).

Saving a database of facts at runtime save saves facts from a given facts section to a file. save takes one or two arguments: save(fileName) save(fileName, databaseName)

/* (i) */ /* (i, i) */

If you call save with only one argument (no facts section name), it will save the facts from the default dbasedom database to the file fileName. If you call save with two arguments (the file name and a facts section name), it will save all facts of the facts section databaseName to the named file.

Examples 1. This is a simple example of how to write a classification expert system using the facts section. The important advantage of using the facts section in this

example is that you can add knowledge to (and delete it from) the program at run time. /* Program ch0145e01.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS thing = string conds = cond* cond = string FACTS is_a(thing, thing, conds) type_of(thing, thing, conds) false(cond) PREDICATES nondeterm run(thing) nondeterm ask(conds) update CLAUSES run(Item):is_a(X, Item, List), ask(List), type_of(ANS, X, List2), ask(List2), write("The ", Item," you need is a/an ", Ans),nl. run(_):write("This program does not have enough "), write("data to draw any conclusions."), nl. ask([]). ask([H|T]):not(false(H)), write("Does this thing help you to "), write(H," (enter y/n)"), readchar(Ans), nl, Ans='y', ask(T). ask([H|_]):assertz(false(H)), fail. is_a(language, tool, ["communicate"]). is_a(hammer, tool, ["build a house", "fix a fender", "crack a nut"]). is_a(sewing_machine, tool, ["make clothing", "repair sails"]). is_a(plow, tool, ["prepare fields", "farm"]).

Chapter 91, Repetition and recursion

187

type_of(english, language, ["communicate with people"]). type_of(prolog, language, ["communicate with a computer"]). update:retractall(type_of(prolog, language, ["communicate with a computer"])), asserta(type_of("PDC Prolog", language, ["communicate with a personal computer"])), asserta(type_of(prolog, language, ["communicate with a mainframe computer"])).

The following database facts could have been asserted using asserta or assertz, or consulted from a file using consult. In this example, however, they're listed in the clauses section. is_a(language, tool, ["communicate"]). is_a(hammer, tool, ["build a house", "fix a fender", "crack a nut"]). is_a(sewing_machine, tool, ["make clothing", "repair sails"]). is_a(plow, tool, ["prepare fields", "farm"]). type_of(english, language, ["communicate with people"]). type_of(prolog, language, ["communicate with a computer"]).

As the goal enter: run(tool).

Respond to each question as if you were looking for some tool to communicate with a personal computer. Now enter the following goal: update, run(tool).

The update predicate is included in the source code for the program, to save you a lot of typing, and will remove the fact type_of(prolog, language, ["communicate with a computer"])

from the facts section and add two new facts into it: type_of(prolog, language, ["communicate with a mainframe computer"]). type_of("Visual Prolog", language, ["communicate with a personal computer"]).

Now respond to each question once again as if you were looking for some tool to communicate with a personal computer.

You can save the whole facts database in a text file by calling the predicate save with the name of the text file as its argument. For example, after the call to save("mydata.dba")

the file mydata.dba will resemble the clauses section of an ordinary Visual Prolog program, with a fact on each line. You can read this file into memory later using the consult predicate: consult("mydata.dba")

2. You can manipulate facts describing database predicates (facts declared in the facts section of your program) as though they were terms. When you declare a facts section, Visual Prolog will internally generate a domain definition corresponding to the facts declaration. As an example, consider the declarations FACTS - dba1 /* dba1 is the domain for these predicates */ person(name, telno) city(cno, cname)

Given these declarations, the Visual Prolog system internally generates the corresponding dba1 domain: DOMAINS dba1 = person(name, telno) ; city(cno, cname)

This dba1 domain can be used like any other predefined domain. For example, you could use the standard predicate readterm (which is covered in chapter 146) to construct a predicate my_consult, similar to the standard predicate consult. Program 2 illustrates one practical way you might use the facts section in an application. This example uses a screen handler, which places text on the screen in predefined locations. A screen layout for the current screen display can be stored in the field and textfield facts that are defined in the screen facts section. Several screen names can be stored in the screens facts section. At run time, the shiftscreen predicate can copy one of these stored screens to the screen facts section by first retracting all current data from the screen facts section, calling the screen predicate to get the layout information for the upcoming screen, then asserting the new screen's form into the screen facts section. /* Program ch0147e02.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */

Chapter 91, Repetition and recursion

189

DOMAINS screenname, fname, type = symbol row,col,len = integer FACTS screen(symbol,dbasedom) FACTS - screen field(fname, type, row, col, len) txtfield(row, col, len, string) windowsize(row,col)

/* Saving different screens */

/* Definitions of field on screen */ /* Showing textfields */

PREDICATES shiftscreen(symbol) CLAUSES shiftscreen(_):retract(field(_,_,_,_,_)), fail. shiftscreen(_):retract(txtfield(_,_,_,_)), fail. shiftscreen(_):retract(windowsize(_,_)), fail. shiftscreen(Name):screen(Name,Term), assert(Term), fail. shiftscreen(_). GOAL shiftscreen(person).

Summary 1. Visual Prolog's facts section is composed of the facts in your program that are grouped into facts sections. You declare the user-defined predicates used in these groups of facts with the keyword facts. 2. You can name facts sections (which creates a corresponding internal domain); the default domain for (unnamed) facts sections is dbasedom. Your

program can have multiple facts sections, but each one must have a unique name. You can declare a given facts predicate in only one facts section. 3. With the standard predicates assert, asserta, assertz, and consult, you can add facts to the facts section at run time. You can remove such facts at run time with the standard predicates retract and retractall. 4. The save predicate saves facts from a facts section to a file (in a specific format). You can create or edit such a fact file with an editor, then insert facts from the file into your running program with consult. 5. You can call database predicates in your program just like you call other predicates. 6. You can handle facts as terms when using the domain internally generated for a database section.

CHAPTER

148

Arithmetic and Comparison 149Visual PrologVisual Prolog's arithmetic and comparison capabilities are similar to those provided in programming languages such as BASIC, C, and Pascal. Visual Prolog includes a full range of arithmetic functions; you have already seen some simple examples of Visual Prolog's arithmetic capabilities. In this chapter we summarize Visual Prolog's built-in predicates and functions for performing arithmetic and comparisons, as well as a two arity versions of a standard predicate used for random number generation. We'll also discuss comparison of strings and characters.

Arithmetic Expressions Arithmetic expressions consist of operands (numbers and variables), operators (+, -, *, /, div, and mod), and parentheses. The symbols on the right side of the equal sign (which is the = predicate) in the following make up an arithmetic expression. A = 1 + 6 / (11 + 3) * Z

Leading "0x" or "0o" signify hexadecimal and octal numbers, respectively, e.g.

Chapter 91, Repetition and recursion

191

0xFFF = 4095 86 = 0o112 + 12

The value of an expression can only be calculated if all variables are bound at the time of evaluation. The calculation then occurs in a certain order, determined by the priority of the arithmetic operators; operators with the highest priority are evaluated first.

Operations Visual Prolog can perform all four basic arithmetic operations (addition, subtraction, multiplication, and division) between integral and real values; the type of the result is determined according to Table 150.1. Table 151.1 Arithmetic Operations1522 Operand 1

Operator

Operand 2

Result

integral

+, -, *

integral

real

+, -, *

integral

real

integral

+, -, *

real

+, -, *

real

integral or real

real

integral

div

integral

mod

integral

In case of mixed integral arithmetic, involving both signed and unsigned quantities, the result is signed. The size of the result will be that of the larger of the two operands. Hence, if a ushort and a long are involved the result is long; if a ushort and a ulong are involved the result is ulong.

Order of Evaluation Arithmetic expressions are evaluated in this order: If the expression contains sub-expressions in parentheses, the sub-expressions are evaluated first.

If the expression contains multiplication (*) or division (/, div or mod), these operations are carried out next, working from left to right through the expression. Finally, addition (+) and subtraction (-) are carried out, again working from left to right. Hence, these are the operator precedence: Table 153.3 Operator Precedence154 Operator

Priority

* / mod div

- + (unary

In the expression A = 1 + 6/(11+3)*Z, assume that Z has the value 4, since variables must be bound before evaluation. (11 + 3) is the first sub-expression evaluated, because it's in parentheses; it evaluates to 14. Then 6/14 is evaluated, because / and * are evaluated left to right; this gives 0.428571. Next, 0.428571 * 4 gives 1.714285. Finally, evaluating 1 + 1.714285 gives the value of the expression as 2.714285. A will then be bound to 2.714285 which makes it a real value. However, you should exercise some care when handling floating point (real) quantities. In most cases they're not represented accurately and small errors can accumulate, giving unpredictable results. An example follows later in the chapter.

Functions and Predicates Visual Prolog has a full range of built-in mathematical functions and predicates that operate on integral and real values. The complete list is given in Table 155.4

Chapter 91, Repetition and recursion

193

Table 156.5: Visual Prolog Arithmetic Predicates and Functions1576Visual Prolog Name

Description

X mod Y

Returns the remainder (modulos) of X divided by Y.

X div Y

Returns the quotient of X divided by Y.

abs(X)

If X is bound to a positive value val, abs(X) returns that value; otherwise, it returns -1 * val.

cos(X)

The trigonometric functions require that X be bound to

sin(X)

a value representing an angle in radians.

tan(X)

Returns the tangent of its argument.

arctan(X)

Returns the arc tangent of the real value to which X is bound.

exp(X)

e raised to the value to which X is bound.

ln(X)

Logarithm of X, base e.

log(X)

Logarithm of X, base 10.

sqrt(X)

Square root of X.

random(X)

Binds X to a random real; 0 <= X < 1.

random(X, Y)

Binds Y to a random integer; 0 <= Y < X.

round(X)

Returns the rounded value of X. The result still being a real

trunc(X)

Truncates X. The result still being a real

val(domain,X)

Explicit conversion between numeric domains.

Generating Random Numbers Visual Prolog provides two standard predicates for generating random numbers. One returns a random real between 0 and 1, while the other returns a random integer between 0 and a given integer. Additionally, the random numbering sequence may be re-initialized.

random/1 This version of random returns a random real number that satisfies the constraints 0 <= RandomReal < 1.

random/1 takes this format: random(RandomReal)

/* (o) */

random/2 This version of random takes two arguments, in this format: random(MaxValue, RandomInt)

/* (i, o) */

It binds RandomInt to a random integer that satisfies the constraints 0 <= RandomInt < MaxValue

random/2 is much faster than random/1 because random/2 only uses integer arithmetic. randominit/1 randominit will initialize the random number generator and is of the following form: randominit(Seed)

/* (i) */

The random number seed is initially 1, and the Seed argument to randominit sets this seed value. The main use for randominit is to provide repeatable sequences of pseudo random numbers for statistical testing. Note that both the integer and floating point versions of random use the same seed and basic number generator. Example Program 1 uses random/1 to select three names from five at random. /* Program ch158e01.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ PREDICATES person(integer, symbol) rand_int_1_5(integer) rand_person(integer)

Chapter 91, Repetition and recursion

195

CLAUSES person(1,fred). person(2,tom). person(3,mary). person(4,dick). person(5,george). rand_int_1_5(X):random(Y), X=Y*4+1. rand_person(0):-!. rand_person(Count):rand_int_1_5(N), person(N,Name), write(Name),nl, NewCount=Count-1, rand_person(NewCount). GOAL rand_person(3).

Integer and Real Arithmetic Visual Prolog provides predicates and functions for modulos function, integer division, square roots and absolute values, trigonometry, transcendental functions, rounding (up or down), and truncation. They are summarized in Table 159.7 and explained in the following pages. mod/2 mod performs the function X modulo Y (where X and Y are integers). X mod Y

The expression Z Z = 7 mod 4 Y = 4 mod 7

/* (i, i) */ = X mod Y

binds Z to the result. For example, /* Z will equal 3 */ /* Y will equal 4 */

div/2 div performs the integer division X/Y (where X and Y are integers). X div Y

/* (i, i) */

The expression Z

= X

div Y binds Z to the integer part of the result. For example,

Z = 7 div 4 Y = 4 div 7

/* Z will equal 1 */ /* Y will equal 0 */

abs/1 abs returns the absolute value of its argument. abs(X)

/* (i) */

The expression Z = abs(X) binds Z (if it's free) to the result, or succeed/fail if Z is already bound. For example, Z = abs(-7)

/* Z will equal 7 */

cos/1 cos returns the cosine of its argument. cos(X)

/* (i) */

The expression Z = cos(X) binds Z (if it's free) to the result, or succeed/fail if Z is already bound. For example, Pi = 3.141592653, Z = cos(Pi)

/* Z will equal -1 */

sin/1 sin returns the sine of its argument. sin(X)

/* (i) */

The expression Z = sin(X) binds Z (if it's free) to the result, or succeed/fail if Z is already bound. For example: Pi = 3.141592653, Z = sin(Pi)

/* Z will almost equal 0 */

tan/1 tan returns the tangent of its argument. tan(X)

Chapter 91, Repetition and recursion

/* (i) */

197

The expression Z = tan(X) binds Z (if it's free) to the result, or succeed/fail if Z is already bound. For example, Pi = 3.141592653, Z = tan(Pi)

/* Z will almost equal 0 */

arctan/1 arctan returns the arc tangent of the real value to which X is bound. arctan(X)

/* (i) */

The expression Z = arctan(X) binds Z (if it's free) to the result, or succeed/fail if Z is already bound. For example, Pi = 3.141592653, Z = arctan(Pi)

/* Z will equal 1.2626272556 */

exp/1 exp returns e raised to the value to which X is bound. exp(X)

/* (i) */

The expression Z = exp(X) binds Z (if it's free) to the result, or succeed/fail if Z is already bound. For example, Z = exp(2.5)

/* Z will equal 12.182493961 */

ln/1 ln returns the natural logarithm of X (base e). ln(X

/* (i) */

The expression Z = ln(X) binds Z (if it's free) to the result, or succeed/fail if Z is already bound. For example, Z = ln(12.182493961)

/* Z will equal 2.5 */

log/1 log returns the base 10 logarithm of X. log(X)

/* (i) */

The expression Z = log(X) binds Z (if it's free) to the result, or succeed/fail if Z is already bound. For example, Z = log(2.5)

/* Z will equal 0.39794000867 */

sqrt/1 sqrt returns the positive square root of X. sqrt(X)

/* (i) */

The expression Z = sqrt(X) binds Z (if it's free) to the result, or succeed/fail if Z is already bound. For example, Z = sqrt(25)

/* Z will equal 5 */

round/1 round returns the rounded value of X. round(X)

/* (i) */

round rounds X up or down to the nearest integral value of X, but performs no type conversion. For example, Z1 = round(4.51) Z2 = round(3.40)

/* Z1 will equal 5 */ /* Z2 will equal 3 */

Both Z1 and Z2 are floating point values following the above; only the fractional parts of the arguments to round have been rounded up or down. trunc/1 trunc truncates X to the right of the decimal point, discarding any fractional part. Just like round, trunc performs no type conversion. trunc(X)

/* (i) */

For example, Z = trunc(4.7)

/* Z will equal 4 */

Again, Z is a floating point number.

Chapter 91, Repetition and recursion

199

val/2 val provides general purpose conversion between the numeric domains, in cases where you want full control over the result of the operation. val observes any possible overflow condition. The format is Result = val(returndom,Expr)

where Expr will be evaluated (if it's an expression), the result converted to returndom and unified with Result. Visual Prolog also has a cast function that will convert uncritically between any domains; this is described in chapter 160. Exercise Use the trigonometric functions in Visual Prolog to display a table of sine, cosine, and tangent values on the screen. The left column of the table should contain angle values in degrees, starting at 0 degrees and continuing to 360 degrees in steps of 15 degrees. Note: Because the trigonometric functions take values expressed in radians, you must convert radians to angles to obtain entries for the left column. Degrees = Radians * 180/3.14159265...

Comparisons Visual Prolog can compare arithmetic expressions as well as characters, strings, and symbols. The following statement is the Visual Prolog equivalent of "The total of X and 4 is less than 9 minus Y." X + 4 < 9 - Y

The less than (<) relational operator indicates the relation between the two expressions, X + 4 and 9 - Y. Visual Prolog uses infix notation, which means that operators are placed between the operands (like this: X+4) instead of preceding them (like this: +(X,4)). The complete range of relational operators allowed in Visual Prolog is shown in Table 161.8. Table 162.9: Relational Operators16310 Symbol

Relation

less than

less than or equal to

equal

greater than

greater than or equal to

<> or ><

not equal

Equality and the equal (=) Predicate In Visual Prolog, statements like N = N1 - 2 indicate a relation between three objects (N, N1, and 2), or a relation between two objects (N and the value of N1 2). If N is still free, the statement can be satisfied by binding N to the value of the expression N1 - 2. This corresponds roughly to what other programming languages call an assignment statement. Note that N1 must always be bound to a value, since it is part of an expression to be evaluated. When using the equal predicate (=) to compare real values, you must take care to ensure that the necessarily approximate representation of real numbers does not lead to unexpected results. For example, the goal 7/3 * 3 = 7

will frequently fail (the exact outcome depends on the accuracy of the floating point calculations in use on your particular platform). Program 2 illustrates another example: /* Program ch164e02.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ PREDICATES test(real,real) run CLAUSES test(X,X):-!, write("ok\n"). test(X,Y):Diff = X-Y, write(X,"<>",Y,"\nX-Y = ",Diff,'\n').

Chapter 91, Repetition and recursion

201

GOAL X=47, Y=4.7*10, test(X,Y).

Except when running Prolog on the UNIX platform, where it behaves as one might expect, this prints: 47<>47 X-Y = 7.1054273576E-15

Therefore, when comparing two real values for equality you should always check that the two are within a certain range of one another. Example Program 3 shows how to handle approximate equality; this is an iterative procedure for finding the square root in order to calculate the solutions to the quadratic equation: A*X*X + B*X + C = 0

The existence of solutions depends on the value of the discriminant D, defined as follows: D = B*B - 4*A*C.

D > 0 implies that there are two unique solutions. D = 0 implies there is only one solution. D < 0 implies that there are no solutions if X is to take a real value (there can be one or two complex solutions). /* Program ch165e03.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ PREDICATES solve(real, real, real) reply(real, real, real) mysqrt(real, real, real) equal(real, real) CLAUSES solve(A,B,C):D=B*B-4*A*C, reply(A, B, D), nl.

reply(_,_,D):D < 0, write("No solution"),!. reply(A,B,D):D=0, X=-B/(2*A),write("x=", X),!. reply(A,B,D):mysqrt(D,D,SqrtD), X1=(-B+SqrtD)/(2*A), X2 = (-B - SqrtD)/(2*A), write("x1 = ", X1," and x2 = ", X2). mysqrt(X,Guess,Root):NewGuess = Guess-(Guess*Guess-X)/2/Guess, not(equal(NewGuess,Guess)), !, mysqrt(X,NewGuess,Root). mysqrt(_,Guess,Guess). equal(X,Y):X/Y >0.99999, X/Y < 1.00001.

To solve the quadratic equation, this program calculates the square root of the discriminant, D. The program calculates square roots with an iterative formula where a better guess (NewGuess) for the square root of X can be obtained from the previous guess (Guess): NewGuess = Guess-(Guess*Guess-X)/2/Guess

Each iteration gets a little closer to the square root of X. Once the condition equal(X, Y) is satisfied, no further progress can be made, and the calculation stops. Once this calculation stops, the program can solve the quadratic using the values X1 and X2, where X1 = (-B + sqrtD)/(2*A) X2 = (-B - sqrtD)/(2*A)

Exercises Load Program 3 and try the following goals: solve(1, 2, 1). solve(1, 1, 4). solve(1, -3, 2).

The solutions should be

Chapter 91, Repetition and recursion

203

x = -1 No solution x1 = 2 and x2 = 1

respectively. The object of this exercise is to experiment with the mysqrt predicate in Program 3. To ensure that temporary calculations are monitored, add the following as the first subgoal in the first mysqrt clause: write(Guess).

To see the effect of this amendment, try this goal: mysqrt(8, 1, Result).

Next, replace the equal clause with this fact: equal(X, X).

and retry the goal. Experiment a little more with the properties of equal. For instance, try equal(X, Y) :X/Y < 1.1 , X/Y > 0.9.

Visual Prolog has a built-in square root function, sqrt. For example, X = sqrt(D)

will bind X to the square root of the value to which D is bound. Rewrite Program 3 using sqrt and compare the answers with those from the original version.

Comparing Characters, Strings, and Symbols Besides numeric expressions, you can also compare single characters, strings and symbols. Consider the following comparisons: 'a' < 'b' "antony" > "antonia" P1 = peter, P2 = sally, P1 > P2

/* Characters */ /* Strings */ /* Symbols */

Characters Visual Prolog converts the 'a' < 'b' to the equivalent arithmetic expression 97 < 98, using the corresponding ASCII code value for each character. You should be aware that only 7 bit ASCII comparisons should be relied upon (i.e. upper and lower case letters a-z, digits, etc.). 8 bit characters, used for a number of national characters, are not necessarily portable between the different platforms.

Strings When two strings or symbols are compared, the outcome depends on a characterby-character comparison of the corresponding positions. The result is the same as you'd get from comparing the initial characters, unless those two characters are the same. If they are, Visual Prolog compares the next corresponding pair of characters and returns that result, unless those characters are also equal, in which case it examines a third pair, and so on. Comparison stops when two differing characters are found or the end of one of the strings is reached. If the end is reached without finding a differing pair of characters, the shorter string is considered smaller. The comparison "antony" > "antonia" evaluates to true, since the two symbols first differ at the position where one contains the letter y (ASCII value 79) and the other the letter i (ASCII value 69). In the same vein, the character comparison "aa" > "a" is true. Similarly, the expression "peter" > "sally" would be false--as determined by comparing the ASCII values for the characters that make up peter and sally, respectively. The character p comes before s in the alphabet, so p has the lower ASCII value. Because of this, the expression evaluates to false. Symbols Symbols can't be compared directly because of syntax. In the preceding P1 = peter, P2 ... example, the symbol peter can't be compared directly to the symbol sally; they must be bound to variables to be compared, or written as strings

Chapter 91, Repetition and recursion

205

CHAPTER

166

Classes and objects Visual Prolog contains a powerful object-mechanism, that melts together the logic programming and object oriented programming (OOP) paradigms. You will find some small examples of Visual Prolog programs, that uses OOPtechnlogy on the CD in the directory \OOP\EXAMPLES. Four criterias have to be fulfilled, before a system can be considered to be object orientated: encapsulation, classes, inheritance, and identity.

Encapsulation The importance of encapsulation and modularity are well known. Encapsulated objects can help building more structured and readable programs because objects are treated like blackboxes. Look at complex problems, find a part, which you can declare and describe. Encapsulate it in an object, construct an interface and continue so, until you have declared all the sub problems. When you have encapsulated the objects of the problem, and ensured that they work correctly, you can abstract from them. OOP is also sometimes known as data-driven programming. You can actually let the objects themselves do the work for you. They contain methods, which are invoked, when they are created, deleted and whenever you call them. Methods can call methods in other objects.

Objects and classes The way data is stored in traditional programming languages is usually hard to grasp for humans and not suited for modelling. Objects are much easier to work with, because it is closer to the way humans understand real-world objects and in fact a tool for modeling in it self. Object is a far more complex data structure than lists. An object is at the basic level a declaration of coherent data. This declaration can also contain predicates, which work on these data. In OOP-terminology these are called methods. Each class type represents a unique set of objects and the operations (methods) available to create, manipulate, and destroy such objects.

A class is a definition of an object. An instance is an actual occurrence of this object. Normally you can define as many instances as you like of a class. Example class automobile owner string brand string endclass Actual instance 1 Owner Beatrice brand Morris Mascot End Actual instance 2 Owner John brand Rolls Royce End

Inheritance OOP is a powerful modeling tool. Objects can be defined on the abstraction level that suits best. From this level child-objects can be defined on lower levels, or parent-objects on higher levels. An object can inherit data and methods from objects at higher levels. Objects are thus an easy way to make very modular programs.

Identity A very important characteristic of objects is, that it identity remains, even though it attributes changes.

Visual Prolog Classes Defining a Class in Visual Prolog requires two things: a class declaration, and a class implementation. The class declaration specifies the interface to the class, what can be seen from the outside. The class implementation contains the Prolog clauses for defining the actual functionallity for the class. The declaration of the interface to a class and the actual definition of the clauses for a class are separated. The Class declarations will often be placed in header files which can be included in places which uses the class. Chapter 167 Classes and objects

207

Class declarations A simplified "first-look" syntax for a class declaration is: CLASS class-name [: parentclass-list ] PREDICATES predicatedeclaration-list FACTS factdeclaration-list ENDCLASS

The optional parentclass-list specifies the parent class or classes from which the class class-name will derive (or inherit) predicates and facts (methods and objects). If any parent classes are specified, the class class-name is called a derived class.

Class implementation A simplified "first-look" syntax for a class declaration is: IMPLEMENT class-name [: parentclass-list ] PREDICATES predicatedeclaration-list FACTS factdeclaration-list CLAUSES Clause-list ENDCLASS

The definition of the clauses for a class is done in a section starting with the IMPLEMENT keyword and ending with the ENDCLASS keyword. Inside the class implementation can be multiple Predicates, Facts and Clauses sections. Unless the Predicates and Facts sections are preceded with the keyword STATIC, the declarations work like they where given in the class declaration, meaning that the facts will belong to the instance, and the predicates will carry the invisible object pointer. However things declared in the implementation will be entirely private to the class implementation. Note also, that it is possible to inherit classes down in the implementation, thus encapsulating details on the class implementation.

Class instances When a class declaration and definition has been made a multiple number of instances of this class can be created. A new instance for a class is made with a call to new for that class. New will return a reference to the instance, which can then be used to perform operations on the object. Example: CLASS counter PREDICATES inc() dec() INTEGER getval() ENDCLASS IMPLEMENT counter FACTS single count(INTEGER) CLAUSES count(0). inc:count(X), X1=X+1, assert(count(X1)). dec:count(X), X1=X-1, assert(count(X1)). getval(VAL):count(Val). ENDCLASS GOAL O = counter::new, Initial = O:getval(), O:inc, NewVal = O:getval(), O:delete.

The output from the program in goal mode will be:

Chapter 167 Classes and objects

209

O=145464, Initial=0, NewVal=1 1 Solution

Objects will survive a fail. Each instance of the class will have its own copy of the facts database. Such a database can be manipulated as usual by retract, assert, save, consult etc. Note that specific new-predicates can be defined. These are called constructors and will be explained later.

Destroying Objects Error! Bookmark not defined.Objects have to be deleted explicitly by calling delete for the object. O = customers::new, O:change_account, O:delete.

Deleting an object will cause an automatic retracting of all facts in the database for that instance. Note that specific delete-predicates can be defined. These are called destructors and will be explained later.

Class Domains The declaration of a class generates a domain with the name of the class. This domain can be used to declare parameters for predicates that should handle a reference to the object. CLASS customer ..... ENDCLASS PREDICATES p(customer)

Passing an object in a parameter means just passing a pointer to the object as in normal OOP-programmingstyle.

Sub-classing and inheritance In a class, both parent class predicates and global predicates can be redefined. For example can the global predicates beep, concat etc. be overridden inside a class. When using the child class, it is possible to use predicates and facts from both the parent class and from the child class, however, the child class might choose to redefine some of the predicates or facts. CLASS person FACTS name( STRING ) father( person ) mother( person ) PREDICATES write_info() ENDCLASS

CLASS employe : person FACTS company(STRING Name) PREDICATES write_info() ENDCLASS

IMPLEMENT person CLAUSES write_info():name(X),write("Name=",X),nl,fail. write_info():father(F),write("Father:\n"), F:person::write_info(),fail. write_info():mother(M),write("Mother:\n"), M:person::write_info(),fail. write_info(). ENDCLASS

Chapter 167 Classes and objects

211

IMPLEMENT employe CLAUSES write_info():this(O), O:person::write_info(),fail. write_info():company(X),write("Company=",X),nl,fail. write_info(). ENDCLASS

GOAL F = person::new(), assert(F:name("Arne")), O = employe::new(), assert(O:name("Leo")), assert(O:father(F)), assert(O:company("PDC")), O:write_info(), O:delete().

The formal syntax for using the members of an object is: [ObjectVariable:] [name_of_class:] name_of_member[( list_of_arguments

)

]

The object can be omitted inside the implementation of a class or for calling the members, which were declared as static. It will be considered as a call to the corresponding member of that class (or its parent) in case it exists. Otherwise, if there is no member with given name, it will be considered as a call to the predicate with same name which must be previously declared in some PREDICATES- or DATABASE section. Names for members may be redefined in the hierarchy of classes. So, in order to refer to names of members in the previous scope, the name of class defined in call to some member may be used for explicit qualification of class in use.

Virtual Predicates In Visual Prolog all Class predicates are what is in the C++ terminology called Virtual methods. Virtual methods allow derived classes to provide different versions of a parent class method. You can declare a method in a parent class and then redefine it in any derived class.

Assume, that a parent class P contains a who_am_i, and class D, derived from P, has definitions for the predicate who_am_i. If who_am_i is called for an object of D, the call made is D: who_am_i, even if the access is via a reference to P. For example: CLASS P PREDICATES test who_am_i() ENDCLASS CLASS D : P PREDICATES who_am_i() ENDCLASS IMPLEMENT P CLAUSES test:-who_am_i(). who_am_i():write("I am of class P\n"). ENDCLASS IMPLEMENT D CLAUSES who_am_i():write("I am of class D\n"). ENDCLASS GOAL O = D::new, O:test, O:delete.

The output from the above program would be: I am of class D

Note, that if you define a predicate in a subclass with different domains or number of arguments, Prolog will treat this as a different declaration, and it will not work as a virtual predicate.

Chapter 167 Classes and objects

213

Static facts and predicates It is possible to declare predicates or facts as being static, which for facts means that the facts are not generated for each instance but there exists only one version for the class. This is useful for example to count the number of instances for a class. Preceding a predicate with the keyword static means that it will not carry the invisible extra argument, which is a pointer to the actual instance. Example: CLASS Count PREDICATES procedure new( ) ENDCLASS

IMPLEMENT Count STATIC FACTS single countInstance( INTEGER ) CLAUSES CountInstance( 0 ).

new( ):countInstance( Num ), NumNext = Num +1, assert( countInstance( NumNext ) ), writef( "Count = %d\n", NumNext ). ENDCLASS GOAL NewObject = count::new(), NewObject1 = count::new().

The output of this program will be: Count = 1 Count = 2 1 Solution

Class Scopes A predicates scope is defined as the area, in which you can access it. Predicate and fact names may be redefined in the class hierarchy. In order to refer to names

in the previous scope the class-name::name() notation can be used to do an explicit naming. CLASS parent PREDICATES p(INTEGER) ENDCLASS CLASS child : parent PREDICATES p(STRING, INTEGER) ENDCLASS % IMPLEMENTATION not shown for clarity GOAL O = child::new, O : parent:p(99)

% Access the definition in parent

Another usage of the explicit scoping is in using classes with static predicates and static facts as packages, like a module system: Example: ILIST = INTEGER* CLASS list static PREDICATES append(ILIST, ILIST, ILIST) - (i,i,o) ILIST gen(INTEGER) ENDCLASS

IMPLEMENT list CLAUSES append([],L,L). append([H|L1],L2,[H|L3]):append(L1,L2,L3). gen(0,[]):-!. gen(N,[N|L]):N1=N-1, L = gen(N1). ENDCLASS

Chapter 167 Classes and objects

215

GOAL L1 = list::gen(3), L2 = list::gen(5), list::append(L1,L2,L3).

Constructors and Destructors The Visual Prolog system will itself allocate and initialize the memory during creation of an object. However there might still be the desire to specify how an object is created, for instance initialize facts to special values, create a window on the screen or open a file. In the same way, a programmer may wish to control, how an object is deleted, for instance closing windows or files. User defined predicates for creating of deleting objects are called constructors and destructors. A constructor is made by giving a definition for the predicate new for an object, and a destructor is made by giving a definition for the predicate delete for an object. In the clauses for the new predicate, It is possible to refer to the constructors of the base class by baseclass::new. CLASS mywind : wind PREDICATES new(INFO,Color) ENDCLASS IMPLEMENT mywind CLAUSES new(Info,Color):wind::new(... , ...,Color), assert(info(INFO)). ENDCLASS GOAL O = mywind::new(“info”,blue), ..... O:delete.

Note! Any base class constructors must be called before doing any references to the object, the compiler will check this. The constructors and destructors are not allowed to fail, they are implicitly declared as procedures. If a constructor or destructor exits with a runtime error, the state of the object is undefined. new will automatically return a reference to the created instance.

Reference to the Object Itself (This) All the non-static predicates of an object have an invisible (to the programmer) extra parameter, which is a pointer to the object. In a clause like: IMPLEMENT x CLAUSES inc:count(X), X1=X+1, assert(count(X1)). ENDCLASS

The object is entirely invisible. If it is necessary to refer to the object itself for instance to access a predicate in an inherited class, it is possible to use the built-in predicate this. The predicate allows for an instance to get access to any member-predicate, which is defined, in corresponding class or in its parents. The syntax for making call to this predicate is: this (

name_of_variable

)

The usage of the predicate is allowed only in predicates, which were declared as not static. This predicate has the single output-parameter. For example: IMPLEMENT x CLAUSES inc:this(Object), Object:x::count(X), X1=X+1, assert(count(X1)). ENDCLASS

This piece of code is functionally identical with the piece of code just above, with the only difference, that you create a pointer to this object. This can be passed on as a parameter to other predicates.

Chapter 167 Classes and objects

217

Abstract Classes An abstract class is a class definition without an implementation. It is only inherited by subclasses. The purpose of an abstract class is to have a declaration of some predicates in order to have some other predicates working on a different specialization of the more general class. An abstract class is defined by the keyword ABSTRACT. In case an abstract class inherits some base classes, these must also be declared abstract. As example if you want to create a browser that can work on many different kinds of data, you will opening the browser by passing it in an object which it can make call to to get the data and move forward or backward. By using an abstract class, the browser knows which predicates it can make calls to. ABSTRACT CLASS browseinterface PREDICATES STRING get_Current() next() prev() ENDCLASS CLASS dbinterface : browseinterface PREDICATES new(DB_SELECTOR,CHAIN) STRING get_Current() next() prev() ENDCLASS CLASS fileinterface : browseinterface STRING get_Current() next() prev() ENDCLASS CLASS browser PREDICATES new(browseinterface) ENDCLASS

Protected facts and predicates It is possible to declare whether you can access the facts or predicates from outside of the class. By default all the facts and predicates are public which means that they can be called from all other predicates. The default access rights can be changed by preceding a fact or predicates declaration with the keywords: protected. Protected means that all classes derived from the class are allowed to access the fact or predicate, but the fact or predicate can not be accessed from outside the class. An example of the usage of protected predicates is call back event handlers, where subclasses might redefine some handling predicates, but it makes no sense to call these from the outside: CLASS window PROTECTED PREDICATES onUpdate(RCT) onCreate(LONG) ENDCLASS

Derived class access control An important issue in building the hierarchies of objects correct, so that you can reuse as much code as possible, is inheritance. You can define methods on one level, and these can then be reused on lower levels. If a class inherits from other classes, we say, that this class is a derived class. When you declare a derived class D, you list the parent classes P1, P2 .. in a comma-delimited parent class-list: CLASS D : P1, P2 ...

D inherits all the facts and predicates of the parent class. Redefined names can be accessed using scope overrides, if needed. In case the name of some member can be achieved by several ways with the directed acyclic graph (DAG), the longest way is accepted. If 2 or more predicates in a class hierarchy are named the same, we say, that this name is overloaded. We then have to consider the scope of the name, here defined as the area in which each predicate is valid. Every usage of name for any member of some class must be unambiguous (to an approximation of overloading). The access to the member of base class is ambiguous if the expression of access names more than one member. The test for unambiguous is performed before the access control. Chapter 167 Classes and objects

219

In case the synonymously defined name is the name of any overloaded member, then the scope of overloading is performed after the ambiguous control (but before the access control). The ambiguity can be scoped with explicit qualifying name of the member by the name of the corresponding class. All predicates from the class declaration, except new and delete, are virtual. Opposite, all predicates declared in the implementation become non-virtual.

Formal definition for classes <class_definition> ::= <definition_begin> <definition_body> [ <class_section_end> ] <definition_begin> ::= [ ABSTRACT ] CLASS <class_name> [ : <parent_class_list> ] <class_name> ::= identifier <parent_class_list> ::= <parent_class_list_item> [ , <parent_class_list_item> . . . ] <parent_class_list_item> ::= [ <parent_mode> ] [ <parent_access> ] <class_name> <definition_body> ::= { <definition_body_item> } <definition_body_item> ::= <domains_section> | [ <access_mode> ] [ STATIC ] <facts_section> | [ <access_mode> ] [ <predicate_mode> ] <predicates_section> <access_mode> ::= PROTECTED <domains_section> ::= DOMAINS { declaration_of_domain } <facts_section> ::= FACTS [ - <name_of_facts_section> ] { <fact_item> } <name_of_facts_section> ::= identifier <fact_item> ::= [ <fact_kind> ] declaration_of_fact <fact_kind> ::= SINGLE | DETERM | NODETERM <predicate_mode> ::= STATIC <predicates_section> ::= PREDICATES { declaration_of_predicate } <class_section_end> ::= ENDCLASS [ <class_name> ] <class_implementation> ::= <implement_begin> <implement_body> <class_section_end> <implement_begin> ::= IMPLEMENT <class_name> <implement_body> ::= { <implement_body_item> } <implement_body_item> ::= <domains_section> | <predicates_section> | <clauses_section> <clauses_section> ::= CLAUSES [ clause_section_body ]

Chapter 167 Classes and objects

221

CHAPTER

168

Advanced Topics 169This is an advanced chapter; we expect that you have been working with the various examples earlier in this book and are now beginning to be an experienced Visual Prolog user. In this chapter, we illustrate how you can control the flow analysis by using the standard predicates free and bound, reference domains, how to use them and how to separate them from the other domains. We also discuss more advanced topics about domains, including the binary domain, pointers to predicates and functions, and return values from functions. Finally, we look at error-handling, dynamic cutting, free type conversions, and a discussion of some programming style issues that will improve your programs' efficiency.

The Flow Analysis In a given predicate call, the known arguments are called input arguments (i), and the unknown arguments are called output arguments (o). The pattern of the input and output arguments in a given predicate call is called the flow pattern. For example, if a predicate is to be called with two arguments, there are four possibilities for its flow pattern: (i, i)

(i, o)

(o, i)

(o, o)

When compiling programs, Visual Prolog carries out a global flow analysis of the predicates. It starts with the main goal and then performs a pseudo-evaluation of the entire program, where it binds flow patterns to all the predicate calls in the program. The flow analysis is quite simple; you are actually carrying it out yourself when you write your program. Here are some examples: GOAL cursor(R, C), R1 = R+1, cursor(R1, C).

In the first call to the cursor, the two variables R and C are free; this means that the cursor predicate will be called with the flow pattern cursor(o,o). You know that the variables are free because this is the first time they've been encountered.

In the expression R1=R+1, the flow analyzer knows that the variable R is bound because it comes from the cursor predicate. If it were free, an error message would have been issued. R1 will be a known argument after this call. In the last call to cursor, both of the variables R1 and C have been encountered before, so they will be treated as input arguments; the call will have the flow pattern cursor(i,i). For each flow pattern that a user-defined predicate is called with, the flow analyzer goes through that predicate's clauses with the variables from the head set to either input or output (depending on the flow pattern being analyzed). Here's an example illustrating this: % This example will only run for DOS Textmode Target PREDICATES changeattrib(Integer, Integer) CLAUSES changeattrib(NewAttrib, OldAttrib) :attribute(OldAttrib), attribute(NewAttrib). GOAL changeattrib(112, Old), write("Hello"), attribute(Old), write(" there").

In the goal section, the first call to the predicate changeattrib is made with the flow pattern changeattrib(i, o) (because 112 is known, and Old is not). This implies that, in the clause for changeattrib, the variable NewAttrib will be an input argument, and OldAttrib will be an output argument. Therefore, when the flow analyzer encounters the first subgoal attribute(OldAttrib), the predicate attribute will be called with the flow pattern attribute(o), while the second call to attribute will have the flow pattern attribute(i). Finally, the call to attribute in the goal will have an input flow pattern, because Old came out of changeattrib. Compound Flow If a predicate argument is a compound term it's also possible to have a compound flow pattern, where the same argument has both input and output flow. Suppose for instance that you have a database of information about countries. To enable easy expansion with new data, it may well be desirable to contain each piece of information in its own domain alternative:

Chapter 170 Advanced topics

223

/* Program ch171e01.pro */ diagnostics DOMAINS cinfo = area(string,ulong); popn(string,ulong); apital(string,string) PREDICATES nondeterm country(cinfo) CLAUSES country(area("Denmark",16633)). country(popn("Denmark",5097000)). country(capital("Denmark","Copenhagen")). country(area("Singapore",224)). country(popn("Singapore",2584000)). country(capital("Singapore","Singapore")).

The following depicts some of the different flow patterns country can be called with: country(C) country(area(Name,Area))area country(popn("Denmark",Pop))popn country(capital("Singapore","Singapore"))

(o) (o,o) (i,o) (i)

Note that as all elements of the term are known in the last example, the flow pattern defaults to plain input. Load 1 and try the examples above. When you look at the diagnostics output, don't be confused by the presence of several flow variants as you progress. The compiler keeps previously compiled code between executions of external goals, unless the source code is changed. When the domains involved in a compound flow pattern are reference domains, the distinction between known and unknown arguments becomes blurred. We'll return to this example in the reference domain section later.

Specifying Flowpatterns for Predicates It is sometimes convenient to specify flowpatterns for your predicates. If you know, that your predicates will only be valid for special flow patterns, it is a good idea to specify flowpatterns for your predicates because the flow analyzer will then catch any wrong usage of these predicates. After specifying the domains, a dash and the possible flowpatterns can be given like in:

PREDICATES frame_text_mask(STRING,STRING,SLIST) - (i,o,o)(o,i,o)

Controlling the Flow Analysis When the flow analyzer recognizes that a standard predicate is called with a nonexistent flow pattern, it issues an error message. This can help you identify meaningless flow patterns when you're creating user-defined predicates that call standard predicates. For example, if you use: Z = X + Y

where the variable X or Y is not bound, the flow analyzer will give an error message saying that the flow pattern doesn't exist for that predicate. To control this situation, you can use the standard predicates free and bound. Suppose you want to create a predicate for addition, plus, which can be called with all possible flow patterns. Program 2 gives the code for such a predicate. /* Program ch172e02.pro */ PREDICATES nondeterm plus(integer, integer, integer) nondeterm num(integer) CLAUSES plus(X,Y,Z):bound(X),bound(Y),Z=X+Y. /* (i,i,o) */ plus(X,Y,Z):bound(Y),bound(Z),X=Z-Y. /* (o,i,i) */ plus(X,Y,Z):bound(X),bound(Z),Y=Z-X. /* (i,o,i) */ plus(X,Y,Z):free(X),free(Y),bound(Z),num(X),Y=Z-X. /* (o,o,i) */ plus(X,Y,Z):free(X),free(Z),bound(Y),num(X),Z=X+Y. /* (o,i,o) */ plus(X,Y,Z):free(Y),free(Z),bound(X),num(Y),Z=X+Y. /* (i,o,o) */ plus(X,Y,Z):free(X),free(Y),free(Z),num(X),num(Y),Z=X+Y. /* (o,o,o) */

Chapter 170 Advanced topics

225

% Generator of numbers starting from 0 num(0). num(X):num(A), X = A+1.

Reference Variables When the flow analyzer has been through a clause, it checks that all output variables in the clause head have been bound in the clause body. If a variable is not bound in a clause, it needs to be treated as a reference variable. Here's an example demonstrating this dilemma: PREDICATES p(integer) CLAUSES p(X):- !. GOAL p(V), V = 99, write(V).

In the goal, the predicate p is called with an output pattern but, in the clause for p, the argument X is not bound. When the flow analyzer recognizes this, it will take a look at the domain corresponding to the variable. If the domain is already declared as a reference domain, there's no problem; if it's not, Visual Prolog gives a warning. When a variable is not bound in a clause, the clause can't return a value. Instead, it will return a pointer to a reference record where the actual value can be inserted at a later time. This requires that the whole domain be treated equally; instead of just passing the values directly for some of the variables of that type, pointers to records will be passed through arguments belonging to the reference domain. When a compound domain becomes a reference domain, all of its subdomains must also become reference domains, because they must also be capable of containing free variables. If you just declare a compound domain to be a reference domain, the compiler will automatically know that all the subdomains are also reference domains.

Declaring Domains as Reference When the flow analyzer encounters an unbound variable, it will only give a warning if the variable is not bound on return from a clause. If you accept this, the domain will automatically be treated as a reference domain. However, you

should always explicitly declare the domains intended to be reference domains in the domains section. This is also required in projects (programs consisting of several modules); when global domains should handle unbound values, the compiler will not allow automatic conversion of the domains. Global domains and predicates are covered in the chapter 173. Notice that the following special predefined domains are not allowed to become reference domains: file, reg, db_selector, bt_selector, and place.

Reference Domains and the Trail Array Because coercion’s and some extra unification are needed, reference domains will in general give a reduction in execution speed. However, some problems can be solved far more elegant and efficiently when you use reference domains, and Visual Prolog has facilities to limit their effect. When you use reference domains, Visual Prolog uses the trail array. The trail array is used to remember when reference variables become instantiated. This is necessary because if you backtrack to a point between the creation and the instantiation of a reference variable, it must be uninstantiated. This problem doesn't exist with ordinary variables, as their points of creation and instantiation are the same. Each instantiation recorded in the trail uses 4 bytes (the size of a 32bit pointer). However, the trail usage is heavily optimized and no record will be placed there if there are no backtrack points between the variable's creation and instantiation. The trail is automatically increased in size when necessary. The maximum size is 64K in the 16-bit versions of Visual Prolog, and practically unbounded in the 32bit versions. Because the code is treated the same for the whole domain, it is usually not a good idea to treat the basic domains as reference domains. Instead, you should declare a domain as being a reference domain to the desired base domain. For instance, in the following code excerpt, the user-defined domain refinteger is declared to be a reference domain to the integer domain. All occurrences of refinteger types will be handled as reference domains, but any occurrence of other integers will still be treated as integers. DOMAINS refinteger = reference integer PREDICATES p(refinteger) CLAUSES p(_).

Chapter 170 Advanced topics

227

Using Reference Domains The right way to use reference domains is to use them only in the few places where they are needed and to use non-reference domains for all the rest. Visual Prolog allows you to convert reference domains to non-reference domains whenever needed. For example, you can create a predicate that converts a reference integer to a non-reference integer with a single fact: DOMAINS refint = reference integer PREDICATES conv(refint,integer) CLAUSES conv(X, X).

Visual Prolog does the conversion automatically when the same variable is used with both a reference domain and a non-reference domain, as it does in the clause when converting X from a refint to an integer. The above is only an explicit example, you don't need to write any special code to convert from reference to non-reference domains. Note that the reference variable needs to be instantiated to a value before it can be converted to the non-reference value. In the same way, if you try to convert a variable from one reference domain to another (such as from reference integers to reference characters), you should make sure the value is bound. Otherwise, Visual Prolog will issue an error message to the effect that free variables are not allowed in the context. Pay attention to the automatic type conversions when you're creating a new free reference variable through a call to free, like so: free(X), Y = X, bind_integer(X), ...

or creating a free variable with the = predicate (equal), like this: Y = X, bind_integer(X), ...

In these examples, free and the = predicate have difficulty finding the correct domain. The type-checker will try to find a suitable domain for the variable during backtracking by means of successive attempts to carry out the flow analysis. The type-checker starts with the character domain and (because char types can be converted to integer types) will choose the character domain instead of proceeding to the integer domain. With reference domains you can return variables that will receive values at a later point. You can also create structures where some places are left uninstantiated until later.

Example To get a feel for how reference domains work, you should try some goals with the well-known predicates member and append: /* Program ch174e03.pro */ DOMAINS refinteger = integer reflist = reference refinteger* PREDICATES nondeterm member(refinteger, reflist) append(reflist, reflist, reflist) CLAUSES member(X,[X|_]). member(X,[_|L]):member(X,L). append([],L,L). append([X|L1],L2,[X|L3]):append(L1, L2, L3).

Load and run this example program, and try the following goals: member(1,L). /* Give all lists where 1 is a member */ member(X,L), X=1. /* Same as before */ member(1,L), member(2,L). /* Lists where both 1 and 2 are members */ X=Y,member(X,L),member(Y,L), X=3.% Both X and Y are members in the list member(1,L), append(L,[2,3],L1). append(L,L,L1), member(1,L). /* All lists where 1 is A member twice */

You will discover that the answers are what you logically expect them to be.

Flow Patterns Revisited A reference variable may well be unbound and yet exist at the time it's used in a predicate call. In example 1, this will happen if for instance you want to find all countries having the same name as their capital, using e.g. samecaps:- country(capital(C,C)), write(C,'\n'), fail.

Here the variable C is used twice with output flow, but what the code really says is that the two variables in capital should share the same value once one of them becomes instantiated. Therefore, both variables are created and unified before the call. In order to do this their domain is converted to a reference domain, and both Chapter 170 Advanced topics

229

variables are in effect known at the time of call, giving a straight input flow pattern. Note that, as said before, it's bad practice to let the standard domains become reference domains. If you want to use the above call, you should declare a suitable reference domain. However, this would create an overhead in all accesses of the country predicate, and it would probably be more efficient to use backtracking to find the special case where name and capital are identical, by using. country(capital(Co,Ca)), Co = Ca, !, ...

Whether this is true or not depends on the size of the database, how many times you perform the call, how many other calls you have, how the arguments are used after the calls, etc.

Using Binary Trees with Reference Domains In chapter 175, you saw how binary trees could be used for fast and efficient sorting. However, sorting can actually be done in a more elegant fashion with reference domains. Because there is no way to change the leaves of a tree when they get new values, a lot of node copying occurs when the tree is created. When you are sorting large amounts of data, this copying can result in a memory overflow error. A reference domain can handle this by letting the leaves of the tree remain as free variables (where the subtrees will later be inserted). By using a reference domain this way, you don't need to copy the tree above the place where the new node is to be inserted. Consider the predicate insert in ch176e04.pro during the evaluation of the following goal: GOAL insert("tom", Tree), insert("dick", Tree), insert("harry", Tree).

In this program, the insert predicate creates a binary tree using the reference domain tree. /* Program ch177e04.pro */ DOMAINS tree = reference t(val, tree, tree) val = string

PREDICATES insert(val, tree) CLAUSES insert(ID,t(ID,_,_)):-!. insert(ID,t(ID1,Tree,_)):ID<ID1, !,insert(ID,Tree). insert(ID,t(_,_,Tree)):insert(ID,Tree). GOAL insert("tom",Tree), insert("dick",Tree), insert("harry",Tree), write("Tree=",Tree),nl

The first subgoal, insert("tom",Tree), will match with the first rule, and the compound object to which Tree is bound takes this form: t("tom", _, _)

Even though the last two arguments in t are not bound, t carried is forward to the next subgoal evaluation: insert("dick", Tree)

This, in turn, binds Tree to t("tom", t("dick", _, _), _)

Finally, the subgoal insert("harry", Tree)

binds Tree to t("tom", t("dick", _, t("harry", _, _)), _)

which is the result returned by the goal.

Sorting with Reference Domains In this section, we add onto the preceding binary tree example (ch178e04.pro) to show how you can isolate the use of reference domains and convert between reference and non-reference domains. The next example defines a predicate that is able to sort a list of values. Chapter 170 Advanced topics

231

/* Program ch179e05.pro */ DOMAINS tree = reference t(val, tree, tree) val = integer list = integer* PREDICATES insert(integer,tree) instree(list,tree) nondeterm treemembers(integer,tree) sort(list,list) CLAUSES insert(Val,t(Val,_,_)):-!. insert(Val,t(Val1,Tree,_)):Val<Val1,!, insert(Val,Tree). insert(Val,t(_,_,Tree)):insert(Val,Tree). instree([],_). instree([H|T],Tree):insert(H,Tree), instree(T,Tree). treemembers(_,T):free(T),!,fail. treemembers(X,t(_,L,_)):treemembers(X,L). treemembers(X,t(Refstr,_,_)):X = Refstr. treemembers(X,t(_,_,R)):treemembers(X,R). sort(L,L1):instree(L,Tree), findall(X,treemembers(X,Tree),L1). GOAL sort([3,6,1,4,5],L), write("L=",L),nl.

In this example, note that reference domains are only used in the tree. All the other arguments use non-reference domains.

Functions and Return Values Visual Prolog includes syntax for letting predicates be considered functions having a return value, rather than plain predicates using an output argument. The difference is a bit more than syntactic, however. Because return values are stored in registers, Prolog functions can return values to, and get return values from, foreign languages, but that's an issue covered in the chapter 180. A function declaration looks like an ordinary predicate declaration, except that the name is prefixed by the domain it's returning: PREDICATES unsigned triple(unsigned)

However, the clauses for a function should have an extra last argument, to be unified with the return value upon success: CLAUSES triple(N,Tpl):- Tpl = N*3. GOAL TVal = triple(6), write(TVal).

The return value need not be one of the standard domains, it can be any domain. If you declare a function that doesn't take any arguments, you must supply an empty pair of brackets when calling it, in order to distinguish it from a string symbol. Given for instance a function to return the hour of the day PREDICATES unsigned hour() CLAUSES hour(H):- time(H,_,_,_).

you must call it like this: ..., Hour = hour(), ...

and not like this ..., Hour = hour, ...

as this will simply consider hour to be the text string "hour", following which the compiler will complain about type errors once you try to use Hour.

Chapter 170 Advanced topics

233

It is also recommended to supply an empty pair of brackets in the declaration of functions and predicates having no arguments. If not, confusing syntax errors may result from misunderstandings between predicate names and domain names, if they clash. If for instance you have a domain named key and you also have a predicate named key, then the declaration: PREDICATES key mypred

can be interpreted in two ways: 1) a predicate named key and a predicate named mypred; 2) a predicate name mypred returning a key. If instead you write: PREDICATES key() mypred()

all ambiguity is resolved. Note that when a predicate is declared as a function, having a return value, it cannot be called as an ordinary Prolog predicate using the extra argument as an output argument; it must be called as a function. The reason for this is that functions store return values in registers, meaning that the code compiled before and in particular after a function call is different from the code around a call of an ordinary predicate. For the same reason, functions calling themselves are currently not tail recursive but this may change in future versions of Visual Prolog. For instance, if you write a function neg to negate each element in a list, like this: DOMAINS ilist = integer* PREDICATES ilist neg(ilist) CLAUSES neg([],[]). neg([Head|Tail],[NHead|NTail]):NHead = -Head, NTail = neg(Tail).

it is not tail-recursive, while neg as a predicate: DOMAINS ilist = integer*

PREDICATES neg(ilist,ilist) CLAUSES neg([],[]). neg([Head|Tail],[NHead|NTail]):NHead = -Head, neg(Tail,NTail).

is tail-recursive. Therefore, don't overdo the use of functions. Their primary aim is to enable you to get returned values from, and return values to, foreign language routines. As a final note, you should be aware that functions with arithmetic return values must be deterministic if they take part in arithmetic expressions.

Determinism Monitoring in Visual Prolog Most programming languages are deterministic in nature. That is, any set of input values leads to a single set of instructions used to produce output values. Furthermore in most languages, for example in C, a called function can produce only a single set of output values. On the contrary, Visual Prolog naturally supports non-deterministic inference based on non-deterministic predicates. The object behind the determinism monitoring is to save run-time storage space. In fact, when a deterministic clause succeeds, the corresponding run-time stack space can be dispensed with at once, thus freeing the storage it occupied. There are a number of reasons why determinism should also concern programmers, most of them involving programming optimization. Visual Prolog has a strongly typed determinism system. Visual Prolog's determinism checking system enforces the programmer to declare the following two behavior aspects of predicates: whether a call to a predicate can fail; number of solutions a predicate can produce. According to these aspects of determinism the following Types of Predicates (rules) are supported in Visual Prolog:

Chapter 170 Advanced topics

235

Predicate Types Number of Solutions, that can be produced 0

Can fail:|

failure

determ

nondeterm

Never fails:

erroneous

procedure multi

Using keywords from the above table in declarations of predicates and predicate domains the programmer can declare the six different types of predicates. Applied these aspects of determinism to declarations of facts we can obtain the following table: Facts Types Number of Solutions can be Produced 0 Can fail:| Never fails:

determ

nondeterm single

In this table Never fails means that as less one instance of a fact always exists, and therefore such a fact never fails if it is called with free arguments. Using keywords from the above table in declarations of facts the programmer can declare the three different types of database predicates (facts).

Visual Prologs determinism checking system Visual Prolog offers unique determinism monitoring facilities based on declarations of types of predicates and facts. All Visual Prolog's standard predicates are internally defined as nondeterm, multi, determ, procedure, failure or erroneous. For user-defined predicates declared with the keywords determ, procedure, failure or erroneous, the compiler always checks and gives warnings for each program clause that results in a non-deterministic predicate. There are two kinds of non-deterministic clauses: If a clause does not contain a cut, and there are one or more clauses that can match with the same input arguments for that flow pattern.

If a clause calls a non-deterministic predicate, and that predicate call is not followed by a cut. Because of the second reason above, non-determinism has a tendency to spread like wildfire throughout a program unless (literally) cut off by one or more cuts. By default, the compiler checks clauses and gives a warning (595 or 596) if it cannot guarantee that a predicate corresponds to the declared type. For example, if the compiler is unable to guarantee that a predicate declared with the keyword multi, procedure or erroneous never fails. Take into account that the compiler is able to verify only necessary conditions for fail (not necessary and sufficient). Therefore, the compiler can sometimes generate warnings 595 and 596 for predicates (declared with the keyword multi, procedure or erroneous) that, in fact, will never fail. For example, PREDICATES procedure str_chrlist(STRING,CHARLIST) - (i,o) CLAUSES str_chrlist("",[]):-!. str_chrlist(Str,[H|T]):frontchar(Str,H,Str1), str_chrlist(Str1,T).

The frontchar predicate can fail if the first parameter is an empty string. The compiler is not sophisticated enough to detect that Str in the second clause cannot be empty. For this example the compiler will generate warning 595 "Nonprocedure clause in the procedure predicate". Of course, non-procedure predicates checking can be switched off if the programmer specifies the compiler option -upro- (), but it is not good programming style. Instead you should reorder the clauses like: str_chrlist(Str,[H|T]):frontchar(Str,H,Str1),!, str_chrlist(Str1,T). str_chrlist(_,[]):-!.

The declaration of procedures catches many small mistakes, like forgetting a catchall clause. There are two rules that you must use when writing predicates declared with the keyword multi, procedure or erroneous: If anyone predicate clause can fail than the final catchall clause must be defined (see the str_chrlist example above). Chapter 170 Advanced topics

237

For any possible (according to declared domains) set of input arguments, a clause, having a head which matches this set, must exist. Otherwise, the compiler will generate the warning 596. For instance, in the following example the third clause p(_) can be redundant if the predicate is declared without procedure keyword, but is required to satisfy this rule if the predicate is declared as procedure. DOMAINS BOOLEAN = INTEGER % b_True = 1, b_False = 0 PREDICATES procedure p(BOOLEAN) CLAUSES p(b_False):- !, ... . p(b_True): - !, ... . p(_): - dlg_error("An illegal argument value").

Notice that the compiler handles erroneous predicates in a special way providing possibility to use them in the final catchall clauses (for handling error situations) in predicates of other types. For instance, the catchall clause in the previous example can be rewritten as the following: p(_): - errorexit(error_vpi_package_bad_data).

Predicates as Arguments So far we have only seen predicate calls of a static nature. That is, the predicates being called as subgoals are specified satirically in the source code. However, in many cases it may be desirable to call different predicates, depending on previous events and evaluations, from the same place, to avoid large-scale duplication of code. To this end, you can declare a predicate domain, and pass pointers to predicates of that domain as variables. The main usage of this feature in Visual Prolog is to pass eventhandler predicates to the VPI layer.

Predicate Domains The declaration for a predicate domain is of the form

pdom = { determ | nondeterm | multi | procedure | failure | erroneous } [ domain ] arglist [ - flowpattern ] [ language ]

(curly braces indicate "choose one", square brackets indicate optional items) where domain is the return domain, if you're declaring a function arglist is of the form ( [ domain [ , domain ]* ] )

flowpattern is of the form ( flow [ , flow ]* ) where flow is { i | o | functor flowpattern | listflow } where listflow is '[' flow [ , flow ]* [ '|' { i | o | listflow } ] ']'

language is of the form language { c | asm | pascal | stdcall | syscall }

The language specification tells the compiler which calling convention to use, and is only required when declaring domains for routines written in other languages (see the chapter 181). The flowpattern specifies how each argument is to be used. It should be the letter i for an argument with input flow, the letter o for one with output flow, a functor and flowpattern for a compound term (e.g. (i,o,myfunc(i,i),o)), or a listflow (eg [i,myfunc(i,o),o], [o,o|i]). You can have no more than one flowpattern declaration for a predicate pointer domain, and it must be given unless the argument list is empty. Hence, the declaration for a group of deterministic predicates taking an integer as argument and returning an integer, would be DOMAINS list_process = determ integer (integer) - (i)

This group, or class of predicates, is now known as list_process. To declare a predicate square as belonging to this group, the syntax is: PREDICATES square: list_process

The clause for square is just like an ordinary clause, but as it's declared as a function it needs a return argument: Chapter 170 Advanced topics

239

CLAUSES square(E,ES):- ES = E*E.

Elaborating on the above, a domain declaration for a group of deterministic predicates, to be called ilist_p, taking an integer list and a pointer to a list_process predicate as input arguments, and an integer list as output argument, would hence be DOMAINS ilist = integer* list_process = determ integer (integer) - (i) ilist_p = determ (ilist,list_process,ilist) - (i,i,o)

Examples Now look at the following program: /* Program ch182e06.pro */ DOMAINS ilist = integer* list_process = determ integer (integer) - (i) ilist_p = determ (ilist,list_process,ilist) - (i,i,o) PREDICATES list_square: list_process list_cube: list_process il_process: ilist_p CLAUSES list_square(E,ES):- ES = E*E. list_cube(E,EC):- EC = E*E*E. il_process([],_,[]). il_process([Head|Tail],L_Process,[P_Head|P_Tail]):P_Head = L_Process(Head), il_process(Tail,L_Process,P_Tail). GOAL List = [-12,6,24,14,-3], il_process(List,list_square,P_List1), write("P_List1 = ",P_List1,'\n'), il_process(List,list_cube,P_List2), write("P_List2 = ",P_List2,'\n').

This declares two functions, list_square and list_cube, belonging to the list_process group, and a predicate il_process creating a new integer list by

applying the listelement-processing predicate L_Process to each element of the original list. Note that the domain declaration ilist_p is only included for illustration; il_process could equally well have been declared using: PREDICATES il_process(ilist,list_process,ilist)

since it isn't referred to as a variable. With the goal shown, il_process is called twice, first creating a list of squares by applying the list_square function, and then a list of cubes by applying the list_cube function. Compile and run this program, and you will get: P_List1 = [144,36,576,196,9] P_List2 = [-1728,216,13824,2744,-27]

Make sure you understand the complexities of this, and, when you do, make sure you don't abuse it. It's all too easy to create totally unreadable programs. Program ch183e07, which is a somewhat elaborated version of ch184e06, illustrates the concept taken to a resonable limit: /* Program ch185e07.pro */ DOMAINS ilist = integer* list_process = determ integer (integer) - (i) list_p_list = list_process* elem_process = determ (integer,integer,integer) - (i,i,o) elem_p_list = elem_process* PREDICATES list_same: list_process list_square: list_process list_cube: list_process elem_add: elem_process elem_max: elem_process elem_min: elem_process il_process(ilist,list_process,ilist) il_post_process(ilist,elem_process,integer) apply_elemprocess(ilist,elem_p_list) apply_listprocess(ilist,list_p_list,elem_p_list) string lpname(list_process) string epname(elem_process)

Chapter 170 Advanced topics

241

CLAUSES lpname(list_same,list_same). lpname(list_square,list_square). lpname(list_cube,list_cube).

% Map predicate pointer to its name

epname(elem_add,elem_add). epname(elem_min,elem_min). epname(elem_max,elem_max). elem_add(E1,E2,E3):- E3 = E1+E2. elem_max(E1,E2,E1):- E1 >= E2, !. elem_max(_,E2,E2). elem_min(E1,E2,E1):- E1 <= E2, !. elem_min(_,E2,E2). list_same(E,E). list_square(E,ES):- ES = E*E. list_cube(E,EC):- EC = E*E*E. il_process([],_,[]). il_process([Head|Tail],E_Process,[P_Head|P_Tail]):P_Head = E_Process(Head), il_process(Tail,E_Process,P_Tail). il_post_process([E],_,E):-!. il_post_process([H|T],L_Process,Result):il_post_process(T,L_Process,R1), L_Process(H,R1,Result). apply_elemprocess(_,[]). apply_elemprocess(P_List,[E_Process|E_Tail]):il_post_process(P_List,E_Process,PostProcess), NE_Process = epname(E_Process), write(NE_Process,": Result = ",PostProcess,'\n'), apply_elemprocess(P_List,E_Tail). apply_listprocess(_,[],_). apply_listprocess(I_List,[L_Process|L_Tail],E_List):il_process(I_List,L_Process,P_List), NL_Process = lpname(L_Process), write('\n',NL_Process,":\nProcessed list = ",P_List, "\nPost-processing with:\n"), apply_elemprocess(P_List,E_List), apply_listprocess(I_List,L_Tail,E_List).

GOAL List = [-12,6,24,14,-3], write("Processing ",List," using:\n"),nl, apply_listprocess(List,[list_same,list_square,list_cube], [elem_add,elem_max,elem_min]).

Among other things, this program illustrates the use of lists of predicate pointers. If you run it, you'll get the following output: Processing [-12,6,24,14,-3] using: list_same: Processed list = [-12,6,24,14,-3] Post-processing with: elem_add: Result = 29 elem_max: Result = 24 elem_min: Result = -12 list_square: Processed list = [144,36,576,196,9] Post-processing with: elem_add: Result = 961 elem_max: Result = 576 elem_min: Result = 9 list_cube: Processed list = [-1728,216,13824,2744,-27] Post-processing with: elem_add: Result = 15029 elem_max: Result = 13824 elem_min: Result = -1728

Predicate pointers may be used like almost any other object in a program. In particular, they can appear as parts of compound terms, creating object oriented possibilities where each object carries with it a series of routines for its own management. You should take note, though, that predicate pointers is a fairly lowlevel mechanism. The actual value of such a pointer is simply a code-address, and it's therefore only valid in the particular program where it was created. Hence, although you can store and retrieve predicate pointers via the databases, highly unexpected and quite possibly disastrous results will occur if you try to use a predicate pointer not originating in the current program.

Chapter 170 Advanced topics

243

The Binary Domain Visual Prolog has a special binary domain for holding binary data, as well as special predicates for accessing individual elements of binary terms. The main use for binary terms is to hold data that has no reasonable representation otherwise, such as screen bitmaps and other arbitrary memory blocks. There are separate predicates for reading binary terms from, and writing them to, files. These will be discussed in chapter 186. With the help of the built-in conversion predicate term_bin, conversion from things such as binary file-headers to Prolog terms is a snap, and binary items going into or coming out of foreign language routines are easily handled. Finally arrays may also be implemented easily and efficiently. Binary terms is a low-level mechanism, whose primary aim is to allow easy and efficient interfacing to other, non-logical, objects, and foreign languages. To this end, binary terms do not behave like other Prolog terms with respect to backtracking. Binary terms will be released if you backtrack to a point previous to their creation, but if you don't backtrack that far any changes done to the term will not be undone. We will illustrate this in the example program at the end of this section.

Implementation of binary terms Pointer

Size

bytes ^ |

A binary term is simply a sequence of bytes, preceded by a word (16bit platforms) or dword (32bit platforms), holding its size. When interfacing to other languages, you should be aware that a term of binary type (the variable passed in the foreign language function call) points to the actual contents, not the size. The Size field includes the space taken up by the field itself. Binary terms are subject to the usual 64K size restriction on 16bit platforms.

Text syntax of Binary Terms Binary terms can be read and written in text format, and also specified in source form in Visual Prolog source code. The syntax is: $[b1,b2,...,bn]

where b1, b2, etc. are the individual bytes of the term. When a binary term is specified in source form in a program, the bytes may be written using any suitable unsigned integral format: decimal, hexadecimal, octal, or as a character. However, the text-representation of binary terms created and converted at runtime is fixed hexadecimal, with no leading "0x" on the individual bytes. Program 8 illustrates this: /* Program ch187e08.pro */ GOAL write("Text form of binary term: ",$['B',105,0o154,0x73,'e',0],'\n').

Load and run this program, and Visual Prolog will respond Text form of binary term: $[42,69,6C,73,65,00]

You should hence be careful if you use e.g. readterm to read a binary term at runtime.

Creating Binary Terms Below we discuss the standard predicates Visual Prolog includes, for creation of binary terms. makebinary/1 makebinary creates and returns a binary term with the number of bytes specified, and sets its contents to binary zero. ..., Bin = makebinary(10), ...

The number of bytes should be the net size, excluding the size of the size field. makebinary/2 makebinary is also available in a two-arity version, allowing specification of an element size. ..., USize = sizeof(unsigned), Bin = makebinary(10,USize), ...

Chapter 170 Advanced topics

245

This creates a binary term with a size given by the number of elements (10 in the above example) multiplied by the elementsize (sizeof(unsigned) in the above), and sets its contents to zero. composebinary/2 composebinary creates a binary term from an existing pointer and a length. It's useful in converting pointers to arbitrary blocks of memory returned by foreign language functions. composebinary takes two arguments, and returns a binary. ..., Bin = composebinary(StringVar,Size), ...

composebinary takes a copy of the StringVar given as input, so changes to the binary term will not affect StringVar, and vice versa. getbinarysize/1 getbinarysize returns the net size (in bytes) of the binary term, excluding the size field in front of the data. ..., Size = getbinarysize(Bin), ...

Accessing Binary Terms There are eight predicates for accessing binary terms, four for setting entries and four for getting entries. Both groups perform range checking based on the size of the binary term, the index specified, and the size of the desired item (byte, word, dword, or real). It's an error to try to get or set entries outside the range of the binary term. Take special note that indices (element numbers) are 0-relative; the first element of a binary term has index 0, and the last element of an N-element binary term has index N-1. getentry/2 getentry is either getbyteentry, getwordentry, getdwordentry, or getrealentry, accessing and returning the specified entry as a byte, word, dword, or real, respectively. ..., SomeByte = getbyteentry(Bin,3), ...

setentry/3 setentry is the counterpart to getentry, setting the specified byte, word, dword, or real entry. ..., setbyteentry(Bin,3,SomeByte), ...

Unifying Binary Terms Binary terms may be unified just like any other term, in clause heads or using the = predicate: ..., Bin1 = Bin2, ...

If either of the terms is free at the time of unification, they will be unified and point to the same binary object. If both are bound at the time of unification, they will be compared for equality. Comparing Binary Terms The result of comparing two binary terms is as follows: If they are of different sizes, the bigger is considered larger; otherwise, they're compared byte by byte, as unsigned values; comparison stops when two differing bytes are found, and the result of their comparison is also the result of the comparison of the binary terms. For instance, $[1,2] is bigger than $[100], and smaller than $[1,3].

Example Program 9 demonstrates a number of aspects of binary terms. /* Program ch188e09.pro */ PREDICATES comp_unify_bin comp_unify(binary,binary) access(binary) CLAUSES comp_unify_bin:Bin = makebinary(5), comp_unify(Bin,_), comp_unify($[1,2],$[100]), comp_unify($[0],Bin), comp_unify($[1,2,3],$[1,2,4]).

Chapter 170 Advanced topics

247

comp_unify(B,B):-!, write(B," = ",B,'\n'). comp_unify(B1,B2):B1 > B2,!, write(B1," > ",B2,'\n'). comp_unify(B1,B2):write(B1," < ",B2,'\n'). access(Bin):setwordentry(Bin,3,255), fail. % Changes are not undone when backtracking! access(Bin):Size = getbinarysize(Bin), X = getwordentry(Bin,3), write("\nSize=",Size," X=",X," Bin=",Bin,'\n'). GOAL % Illustrate comparison and unification of binary terms comp_unify_bin, % Allocate a binary chunk of 4 words WordSize = sizeof(word), Bin = makebinary(4,WordSize), access(Bin), % Illustrate range checking; element numbers are 0-relative write("Run-time error due to wrong index:\n"), Index = 4, trap(setwordentry(Bin,Index,0),E, write("Error ",E," setting word index ",Index," of ",Bin,'\n')).

This example uses the trap predicate, which will be discussed in the section about error handling below.

Converting Terms to Binary Terms A compound term may have its arguments scattered all over memory, depending on what domains they belong to. Simple types are stored directly in the term record itself, while complex types (those accessed via a pointer, and allocated separately on the global stack) will not necessarily be anywhere near the term they appear in. This is a problem if a term has to be sent out of a program, so to speak, as there is no way make an explicit copy of its contents. Unifying a term variable with another variable will only take a copy of the pointer to the term.

Using term_str (discussed in chapter 189), it is possible to convert the term to a string and back again, but this is rather inefficient when all that's needed is a copy of the term's contents. term_bin solves this problem. term_bin/3 term_bin will convert between a term of any domain and a block of binary data, holding the term's contents as well as pointer fixup information. The pointer fixup information will be applied to the binary data when converted back to a term, allowing recreation of any pointers to complex terms the term contains. term_bin looks like this: term_bin(domain,Term,Bin)

/* (i,i,o) (i,_,i) */

The domain is the domain the Term belongs, or should belong, to, and Bin is a binary term holding the Term's contents. Example Program 11 demonstrates conversion between a term and its binary representation. The domains and alignment have been explicitly chosen to ease description, as they would otherwise differ between 16- and 32-bit platforms. Alignment of terms is usually only relevant when interfacing to foreign languages, and is fully described in the chapter 190. /* Program ch191e11.pro */ DOMAINS dom = align byte cmp(string,short) GOAL T = cmp("Bilse",31), term_bin(dom,T,B), write("Binary form of ",T,":\n",B), term_bin(dom,T1,B), write("\nConverted back: ",T1,'\n').

If you run this, you'll get: Binary form of cmp("Bilse",31): $[01,07,00,00,00,1F,00,42,69,6C,73,65,00,01,00,00,00,01,00,00,00] Converted back: cmp("Bilse",31)

Chapter 170 Advanced topics

249

You shouldn't be too concerned about the actual format of this, in particular as we're dealing with implementation details, which may change. Nevertheless, we'll briefly describe the contents of the binary information: $[01,07,00,00,00,1F,00,42,69,6C,73,65,00,01,00,00,00,01,00,00,00] | |_________| |___| |_______________| |_________| |_________| | | | | | | functor | 31 "Bilse"\0 offset of # of ptrs | ptr to fix in fixup 0-relative (array, but ptr to string only one element here)

The offset of ptr to fix array will be 16-bit quantities on 16-bit platforms, as will the # of ptrs in fixup. If the term contains elements from the symbol domain, the binary term will contain additional information to insert the symbols in the symboltable when the term is recreated. Visual Prolog uses term_bin itself when storing things in the database system and when sending terms over a message pipe to another program. If several programs share external databases or communicate over pipes, it's hence crucial that the domains involved use the same alignment.

Errors and Exception Handling As software quality improves, error handling becomes increasingly important in providing safe and trustworthy programs that users feel they can rely on. In this section we look at the standard predicates Visual Prolog provides, giving you control over the errors and exceptions that may occur when your application is running. This includes trapping run-time errors and controlling user interruption. If you look in Visual Prolog's error-message file (PROLOG.ERR on the DOSrelated platforms, "PDCProlog.err" in UNIX), you'll see all the error numbers applicable to both compile-time and run-time problems. All numbers above and including 10000 are reserved for user program exit codes, and you may modify and distribute the error message file if required. Additionally, in the include directory you'll find the ERROR.CON include file, containing constant declarations for all error codes. To guard against future changes, use this file for error codes rather than hard-coding numbers into your application.

Exception Handling and Error Trapping The cornerstone of error and exception handling is the trap predicate, which can catch run-time errors as well as exceptions activated by the exit predicate. You can also use this mechanism to catch signals, such as that generated by CtrlBreak in the textmode platforms, as well as a kind of "block exit" mechanism. exit/0 and exit/1 A call to exit has an effect identical to a run-time error. exit exit(ExitCode)

/* (no arguments) */ /* (i) */

exit without an argument is equivalent to exit(0). If the call to exit is executed in a direct or indirect subgoal of a trap, the ExitCode will be passed to the trap. The behavior of an untrapped exit depends on the platform. The VPI event handlers do their own trapping, and an exit will be caught here resulting in an error message. An untrapped call of the exit predicate on the textmode platforms results in program termination, and the OS return code ('ErrorLevel' in the DOS-related operating systems, '$?' in UNIX sh) will be set to the value used in the call. The maximum value a process can exit with is 254; 255 is reserved for Visual Prolog's system call, but no checks are performed. trap/3 trap, which takes three arguments, carries out error trapping and exception handling. The first and the last arguments to trap are predicate calls, and the second argument is a variable; it takes this format: trap(PredicateCall, ExitCode, PredicateToCallOnError)

For example, consider the call: trap(menuact(P1, P2, P3), ExitCode, error(ExitCode, P1)), ...

If an error occurs during execution of menuact--including all further called subgoals--an error code will be returned in the variable ExitCode, and the errorhandling predicate error will be called. trap will then fail on return from error. If menuact returns successfully, evaluation will continue after the trap, which will no longer be effective. Before calling the error predicate, the system resets the stack, global stack, and trail to the values they had before the goal specified in the trap (menuact in the Chapter 170 Advanced topics

251

example above) was called. This means that you can use a trap to catch memory overflows, but you shouldn't rely on big memory consuming operations such as database updates to be in either a complete or unaltered state - a heap-full error may occur anytime. If Break is enabled on the textmode platforms, and a Break occurs (because the user pressed Ctrl-Break during execution of a predicate with a surrounding trap), the trap will catch the Break and return 0 in the ExitCode variable. Example: catching file-not-open /* Program ch192e12.pro */ include "error.con" DOMAINS file = inpfile PREDICATES ioehand(integer,file) getline(file,string) CLAUSES ioehand(err_notopen,File):-!, write(File," isn't open\n"), exit(1). ioehand(Err,File):write("Error ",Err," on ",File,'\n'), exit(1). getline(File,Line):readdevice(Old), readdevice(File), readln(Line), readdevice(Old). GOAL trap(getline(inpfile,First),Err,ioehand(Err,inpfile)), write(First).

errormsg/4 You can use the errormsg predicate to access files that are structured the same as Visual Prolog's error-message file. errormsg(File name, ErrorNo, ErrorMsg, ExtraHelpMsg) /* (i,i,o,o) */

A typical use of errormsg is in error-trapping predicates to obtain an explanation of an error code, as illustrated below. PREDICATES error(integer) main /*....*/ CLAUSES error(0) :- !. % discard break. error(E) :errormsg("prolog.err", E, ErrorMsg, _), write("\nSorry; the error\n", E, " : ", ErrorMsg), write("\nhas occurred in your program."), write("\nYour database will be saved in the file error.sav"), save("error.sav"). GOAL trap(main, ExitCode, error(Exitcode)).

Error reporting Visual Prolog includes several compiler directives that you can use to control runtime error reporting in your programs. These directives allow you to select the following: whether code should be generated to check for integer overflows. the level of detail in reporting run-time errors. whether code should be generated for stack overflow checking. You can place these compiler directives at the top of your program, or choose them from the Compiler Options dialog. errorlevel Visual Prolog has a mechanism to locate the source position where a run-time error occurs. To do this, it generates code before predicate calls to save the source code position where executions are actually performed. The level of error reporting and the storing of source positions are selected by the errorlevel compiler directive. The syntax is: errorlevel = d

where d is one of 0, 1, or 2, representing the following levels: Chapter 170 Advanced topics

253

This level generates the smallest and most efficient code. No source positions are saved. When an error occurs, just the error number is reported.

This is the default level. When an error occurs, the Visual Prolog system displays its origin (module name and include file, if applicable). The place where the error was detected within the relevant source file is also displayed, expressed in terms of the number of bytes from the beginning of the file.

At this level, certain errors not reported at level 1, including stack overflow, heap overflow, trail overflow, etc., are also reported. Before each predicate call code is generated to store the source position.

When a source position are reported, the source program can be loaded into the editor, and you can activate the Edit | Go To Line Number menu item, where you can enter the position number and the cursor will move to the place where the error occurred. In a project, the errorlevel directive in each module controls that module's detail of error reporting. However, if the errorlevel directive in the main module is higher than that of the other modules, the system might generate misleading error information. If, for example, an error occurs in a module compiled with errorlevel = 0, which is linked with a main module compiled with errorlevel set to 1 or 2, the system will be unable to show the correct location of the error--instead, it will indicate the position of some previously executed code. For more information about projects, refer to "Modular Programming" in the chapter 193. lasterror/4 Hand in hand with trap and errormsg goes lasterror. It returns all relevant information about the most recent error, and looks like this: lasterror(ErrNo,Module,IncFile,Pos)

/* (i,i,i,i) */

where ErrNo is the error number, Module is the source file name, IncFile is the include file name, and Pos is the position in the source code where the error occurred. However, the program must be compiled with an errorlevel greater than 1 in order for the information to be relevant in case of memory overflow. For ordinary errors, an errorlevel of 1 is sufficient.

The primary aim of lasterror is to ease debugging when subgoals are trapped, but it may equally well form the basis of a cause-of-death indicator in commercially distributed software. Using lasterror, your code can provide the user with a quite sober error message in addition to exact information about what happened.

Handling Errors from the Term Reader When you call consult or readterm and a syntax error occurs in the line read, the predicates will exit with an error. The syntax error could be any one of the following: A string is not terminated. A symbol is placed where an integer is expected. Upper-case letters are used for the predicate name. A symbol is not surrounded by double quotes. Etc. When consult was originally introduced in Visual Prolog, it was not meant to be used for reading user-edited files: It was designed to read back files that were saved by the save predicate. In order to make it easier to consult user-created files, we have introduced the two predicates readtermerror and consulterror. You can call these to obtain information about what went wrong in readterm or consult, respectively. If the errors from consult and readterm are caught by the trap predicate, consulterror and readtermerror allow you to inspect and possibly edit the cause of the syntax error. consulterror/3 consulterror returns information about the line containing a syntax error. consulterror(Line, LinePos, Filepos),

/* (o,o,o) */

Line is bound to the line that has the syntax error, LinePos is bound to the position in the line where the syntax error was found, and FilePos is bound to the position in the file where the line was read. /* Program ch194e13.pro */ CONSTANTS helpfile = "prolog.hlp" errorfile = "prolog.err"

Chapter 170 Advanced topics

255

DOMAINS dom = f(integer) list = integer* DATABASE - mydba p1(integer, string, char, real, dom, list) PREDICATES handleconsulterr(string, integer) CLAUSES handleconsulterr(File, Err):Err>1400, Err<1410, !, retractall(_, mydba), consulterror(Line, LinePos, _), errormsg(errorfile, Err, Msg, _), str_len(Blanks,LinePos), write("Syntax error in ",File,'\n',Line,'\n',Blanks,"^\n",Msg,'\n'), exit(1). handleconsulterr(File,Err):errormsg(errorfile,Err,Msg,_), write("Error while trying to consult ",File,":\n",Msg,'\n'), exit(2). GOAL File="faulty.dba", trap(consult(File, mydba), Err, handleconsulterr(File,Err)), write("\nSUCCESS\n").

readtermerror/2 readtermerror returns information about the readterm-read line containing a syntax error. readtermerror(Line, LinePos),

/* (o,o) */

Line is bound to the line that has the syntax error, and LinePos is bound to the position in the line where the syntax error was found.

Break Control (Textmode Only) It is important to understand how the break/signal mechanism is implemented in Visual Prolog. Generally, a break does not immediately abort the current execution. Rather, Visual Prolog has an exception handler installed, which sets a

flag when activated by the signal. Visual Prolog checks this flag in two different cases: If the code is compiled with break-checking enabled, the status of the break-flag is examined each time a predicate is entered. Break-checking may be disabled by using the nobreak directive in the source code, through the Options/Compiler Directives/Run-time check menu item, or from the commandline. Several of the library routines check the break-flag. If the break-flag is set, the outcome depends on the break-status, set by the predicate break: If break-status is Off, the signal is ignored for the time being, otherwise the code will exit with an appropriate exitcode (discussed below). This exit will of course be caught by a trap, if any is set. break/1 break enables and disables the sensing of the break-flag during execution. break takes one of the following forms: break(on) break(off) break(BreakStatus)

/* (i); enables the BREAK key */ /* (i); disables the BREAK key */ /* (o); returns the current BREAK status */

You can read the current break status by calling break with an output variable. This means that, during critical operations, you can disable break and then return to the original break state afterwards. For example: update :break(OldBreak), break(off), /* .... do the updating, */ break(OldBreak).

For the DOS-related versions, the exitcode resulting from a break will always be 0, as the only signal recognized is the user interrupt. For the UNIX version, SIGINT also results in an exit value of 0, for backwards compatibility with the large base of installed DOS programs written in Visual Prolog. For other signals which the process has been set up to catch, the exitcode is the signal number plus the constant err_signaloffset, defined in the include file ERROR.CON.

Chapter 170 Advanced topics

257

breakpressed/0 breakpressed succeeds if the break-flag is set, even when the break-state has been turned off by break(off) or the program compiled with the nobreak option. If successful, breakpressed returns the exitcode generated by the most recently caught signal, and clears the break-flag. For the DOS-related versions of Visual Prolog, this will always be 0; for UNIX, it will be the same value as would otherwise be passed through the exit/trap mechanism, as described above. This too will be 0 if SIGINT is received.

Manual Break and Signal Checking in UNIX This section, down to page 261, only applies to UNIX and may be skipped by users of the DOS-related versions. A Visual Prolog program may be configured to catch any of the many different signals a UNIX process can receive (see signal(S)). However, as signals may arrive at any time, quite asynchronously from the running process, it's important that they don't interrupt the process while in the middle of something critical, such as memory allocation. The reason for this is that, due to Prolog's modularity, the only means of communication between different predicates is through arguments or through databases. Obviously, an asynchronously executed signalhandler predicate can't communicate to the rest of the program through arguments, leaving only the database. And since databases rely on memory allocation, which invariably is in use by the rest of the program, an asynchronously executed signal-handling predicate could create havoc, if trying to e.g. assert something to indicate that a signal was received, while the interrupted process was in the middle of allocating memory. It really all boils down to Prolog not having global variables, leaving asynchronously executed predicates with no means of communication with the rest of the program. Therefore, rather than invoking a signal-handling predicate the instant the signal is received, signals are routed through the exit/trap mechanism. signal/2 Signal-handling in Visual Prolog programs is controlled by the signal predicate, defined in the include file ERROR.PRE: GLOBAL DOMAINS sighand = determ (integer) - (i) language C GLOBAL PREDICATES sighand signal(integer,integer) - (i,i) language C as "_BRK_Signal" sighand signal(integer,sighand) - (i,i) language C as "_BRK_Signal"

CONSTANTS sig_default = 0 sig_ignore = 1 sig_catch = 2

To modify the handling of a specific signal, call signal with the signal exitcode you want to catch, such as err_sigalrm, defined in ERROR.PRE, specifying in the second argument what to do: sig_default to reset the handling of the signal to the default for the process sig_ignore to ignore the signal completely sig_catch to have the signal routed through the exit/trap mechanism anything else is taken to be the address of a function to be invoked when the signal occurs The return value of signal is the previous handling of the signal in question, which will be one of the values outlined above. The only cases where you may use the fourth alternative (address of function) is when this value was returned by a previous call to signal, or when the function is one you have written yourself in C, exercising the usual precautions when writing signal handlers. In particular, SIGINT is masked out during the execution of the signal handler, so if you intend to do a longjump from a signal handler you're written in C, SIGINT must be enabled first (see sigprocmask(S)). The validity of the function address is not verified at the time signal is called and results may be highly erratic if it's an invalid address; see signal(S). Although the name and argument profile of signal matches that of signal(S), it is implemented using sigaction(S) and SIGINT is ignored during execution of installed signal handlers. By default, Visual Prolog catches the following signals: SIGINT (user interrupt); results in exit of 0 when detected. SIGFPE (floating point exception); results in an exit of err_realoverflow immediately after the erroneous calculation. SIGBUS and SIGSEGV (memory fault); these signals result from attempting to access memory not belonging to the process, typically due to a faulty pointer. A short message, indicating where in the program the error happened, will be printed if possible (see the errorlevel compiler directive), and the process is terminated, leaving a core dump. Unless you have made a mistake in modules you have written yourself in C, this invariably indicates an internal error. Chapter 170 Advanced topics

259

SIGILL (illegal instruction); the processor encountered an unrecognized or illegal instruction. Other details as for SIGBUS and SIGSEGV. Any signals caught will be routed through the same function as SIGINT. Note that once catching has been enabled for a signal, it remains in effect until explicitly reset by another call to signal. Receiving and catching a signal will not reset the signal handling. Needless to say, signal catching should be used very carefully, and the break-state should always be turned off if you intend to receive and test for signals without interrupting the program. In particular, a number of operating system calls will be terminated prematurely if a signal is caught while they're executing. When the break-state is off, the reception of the signal will be noted in the break-flag and the interrupted system call called again, meaning the program should work as expected. However, while every care has been taken to ensure the integrity of this scheme, no guarantees can be given. For instance, some versions of SCO UNIX and SCO Open Desktop will allow interrupts of certain terminal I/O functions, without giving any indication that such an interrupt occurred. Below are two examples, using the alarm clock signal. Both use the breakpressed predicate, which will be described later. The first example will print the message "Do something!" every three seconds, until the user enters a character. It doesn't turn the break-state off during the central parts of the program, as the whole purpose is to interrupt a system call. /* Program ch195e14.pro */ /* For UNIX platform only */ include error.con" GLOBAL PREDICATES alarm(integer) - (i) language C

% See alarm(S)

PREDICATES brkclear nondeterm repeat ehand(integer) getchar(char) CLAUSES brkclear:-breakpressed,!. brkclear. repeat. repeat:-repeat.

% Clear break-flag, if set

ehand(2214):-!, write("Do something!\n"). ehand(E):write("\nUnknown exit ",E,'\n'), exit(2). getchar(C):write("Enter char: "), alarm(3), % Alarm to go off in 3 seconds readchar(C), % This will exit with err_sigalrm when receiving SIGALRM alarm(0), % Cancel pending alarm signal break(off), brkclear, % Clear break-flag, in case alarm went off break(on). % just before cancellation above. GOAL Old=signal(err_sigalrm,sig_catch), repeat, trap(getchar(C),Err,ehand(Err)), !, signal(err_sigalrm,Old), write("\nYou entered '",C,"'\n").

% Declared in error.con

The next example, which has been deliberately written to be somewhat inefficient, displays program progress during lengthy processing. Break-status is turned off in this program, and the detection of any signals is handled manually, using the breakpressed predicate. /* Program ch196e15.pro */ /* For UNIX platform only */ include "error.con" GLOBAL PREDICATES alarm(integer) - (i) language C% See alarm(S) DATABASE rcount(unsigned) dba(real,real,real) PREDICATES nondeterm repeat process_dba bcheck bcheck1(integer)

Chapter 170 Advanced topics

261

CLAUSES repeat. repeat:- repeat. rcount(0). dba(1,1,1). process_dba:retract(dba(F1,F2,F3)), !, F = F1 * F2 * F3, assert(dba(F,F,F)), retract(rcount(N)), !, NN = N+1, assert(rcount(NN)), NN = 25000. % fail back to repeat in goal bcheck:Break = breakpressed(),!, bcheck1(Break). bcheck. bcheck1(err_sigalrm):-!, rcount(N),!, time(H,M,S,_), writef("\r%:%:% % records ",H,M,S,N), alarm(1). % Next alarm in 1 second bcheck1(0):-!, write("\nInterrupt\n"), exit(1). bcheck1(Exit):write("\nUnknown exit ",Exit,"; runtime error?\n"), exit(2). GOAL break(off), Old = signal(err_sigalrm,sig_catch), alarm(1), repeat, bcheck, process_dba, !, alarm(0), signal(err_sigalrm,Old), dba(F1,F2,F3), !, write('\n',F1,' ',F2,' ',F3,'\n').

The writef predicate is covered in chapter 197.

% Declared in error.pre % First alarm in 1 second

% Cancel pending alarm

Critical Error Handling under DOS Textmode This section applies only to the DOS textmode platform, and are not relevant for VPI programs. The DOS-version of Visual Prolog's library contains some default routines for handling error situations, but you can actually substitute the default code with your own clauses. In this section, we describe two routines -- criticalerror and fileerror. DOS will call criticalerror when a DOS error occurs. The Visual Prolog system calls fileerror when it gets a file error in the run-time editor. If you define these predicates as global and supply your own clauses for them, the linker will take your code instead of the code from the library. The result is that you gain better control over error situations. Your .EXE program's size might also decrease (because the code for the default routines draw in window routines). Global declarations for criticalerror and fileerror are given in the include file ERROR.PRE shipped with the Visual Prolog system in the include directory. criticalerror/4 Visual Prolog defines this routine for handling DOS critical errorscritical errors (DOS interrupt 24H)DOS, interrupt 24H, but not for OS/2. If you want to use your own version of criticalerror, you should include ERROR.PRE which gives a global declaration as follows: GLOBAL PREDICATES criticalerror(ErrNo, ErrType, DiskNo, Action) - (i, i, i, o) language c as "_CriticalError_0"

Refer to the chapter 198 for information on how to use global declarations. The criticalerror predicate must never fail, and it works only from an .EXE file application. The criticalerror predicate replaces the DOS critical error interrupt handler and has the same restriction as the original interrupt handler. (Refer to the DOS Technical Reference for details.) You can only use DOS function calls 01h to 0Ch and 59h ("Get extended error")--that means console I/O and nothing else. If your application uses any other DOS function calls, the operating system is left in an unpredictable state.

Chapter 170 Advanced topics

263

Table 199.1: Argument Values for the criticalerror Predicate200 Argument

Value

Meaning

ErrNo

= 0 = 1 = 2 = 3 = 4 = 5 = 6 = 7 = 8 = 9 = 10 = 11

Attempt to write on write-protected disk Unknown unit Drive not ready Unknown command CRC error in data Bad drive request structure length Seek error Unknown media type Sector not found = 12 Printer out of paper Write fault Read fault General failure

ErrType

= 0 = 1 = 2

Character device error Disk read error Disk write error

DiskNo

= 0-25

Means device A to Z

Action

= 0 = 1 = 2

Abort current operation Retry current operation Ignore current operation (this could be very dangerous and is not recommended)

fileerror/2 Visual Prolog will activate the predicate fileerror when a file in the textmode editor action fails. If you define your own fileerror predicate, it is not allowed to fail, and it works only from an .EXE file application. The declaration for fileerror in the ERROR.PRE file is:

GLOBAL PREDICATES fileerror(integer, string) - (i, i) language c as "_MNU_FileError"

Note that this declaration is correct -- you must specify language c even though the source code will be in Prolog.

Dynamic Cutting The traditional cut in Prolog is static. One problem with this is that the effect of the cut happens when execution passes the ! symbol, and it affects only those clauses in which it was placed (in the source text). There is no way of passing the effect of a cut in an argument to another predicate, where the cut might only be evaluated if some conditions were fulfilled. Another problem with the traditional cut is that it is impossible to cut away further solutions to a subgoal in a clause, without also cutting away the backtrack point to the following clauses in the predicate. Visual Prolog has a dynamic cutting mechanism, which is implemented by the two standard predicates getbacktrack and cutbacktrack. This mechanism allows you to handle both of these problems. The predicate getbacktrack returns the current pointer to the top of the stack of backtrack points. You can remove all backtrack points above this place, at some later time, by giving the pointer thus found to the cutbacktrack predicate. Examples Here are some examples that illustrate the use of these two predicates. Suppose you have a database of people and their incomes, and you have registered who their friends are. DATABASE person(symbol, income) friends(symbol, symbol)

If you define a happy person as one who either has some friends or pays little tax, the clauses that return happy persons could be as follows: happy_person(has_friends(P)) :- person(P, _), friends(P, _). happy_person(is_rich(P)) :- person(P, Income), not(rich(Income)).

If a person has more than one friend, the first clause will return a multiple number of solutions for the same person. You could, of course, add another predicate have_friends(P,P) that has a cut, or you could use the dynamic cut instead. Chapter 170 Advanced topics

265

happy_person(has_friends(P)) :person(P, _), getbacktrack(BTOP), friends(P, _), cutbacktrack(BTOP).

Although the friends predicate might return many solutions if backtracked into, that possibility is cut away with the call to cutbacktrack. A subsequent failure would backtrack into the person predicate. The more important use of a dynamic cut is when you pass the backtrack pointer to another predicate and execute the cut conditionally. The pointer is of unsigned type and can be passed in arguments of unsigned type. As an illustration of this, let's say you want a predicate to return numbers until the user presses a key. PREDICATES number(integer) return_numbers(integer) checkuser(unsigned) CLAUSES number(0). number(N) :- number(N1), N = N1+1. return_numbers(N) :- getbacktrack(BTOP), number(N), checkuser(BTOP). checkuser(BTOP) :- keypressed, cutbacktrack(BTOP). checkuser(_).

The compiler does not recognize the cutbacktrack predicate in the pass that analyzes the clauses for determinism. This means you could get the warning Non-deterministic clause when using the check_determ directive, even if you called cutbacktrack. You should use dynamic cutting very judiciously. It's all too easy to destroy program structure with dynamic cutting, and careless use will invariably lead to problems that are very hard to track down.

Free Type Conversions In most cases there is little need to start mixing wildly differing types. However, from time to time, in particular when dealing with system level programming or when interfacing to foreign languages, rather reckless conversions have to be dealt with. To this end the cast function will convert from anything to anything.

No checks are performed on the supplied values, and quite disastrous results will occur if you try to use incorrectly cast variables. The format of cast is Result = cast(returndomain,Expr)

where Expr is evaluated (if it's a numerical expression), converted to returndomain type, and unified with Result. For instance, a null string pointer (a character pointer with a value of 0; not an empty string, which is a pointer to a byte with a value of 0) can be created using: NullPtr = cast(string,0)

Don't try to write the resulting string, you'd most probably get a protection violation, a hung system, or at best garbage characters. If you don't see any obvious use for cast, don't worry. It plays no part in ordinary Prolog programs.

Programming Style In this section, we provide some comprehensive guidelines for writing good Visual Prolog programs. After summarizing a few rules of thumb about programming style, we give you some tips about when and how to use the fail predicate and the cut.

Rules for Efficient Programming Rule 1. Use more variables rather than more predicates. This rule is often in direct conflict with program readability. To achieve programs that are efficient (both in their demands upon relatively cheap machines and upon relatively expensive human resources) requires a careful matching of objectives. Often, the purely declarative style of Prolog leads to code that is significantly less efficient than other (non-declarative) approaches. For instance, if you're writing a predicate to reverse the elements of a list, this code fragment: reverse(X, Y) :- reverse1([], X, Y). /* More efficient */ reverse1(Y, [], Y). reverse1(X1, [U|X2], Y) :- reverse1([U|X1], X2, Y).

Chapter 170 Advanced topics

267

makes less demands upon the stack than the next one, which uses the extra predicate append: reverse([], []). /* Less efficient */ reverse([U|X], Y) :- reverse(X, Y1), append(Y1, [U], Y). append([], Y, Y). append([U|X], Y, [U|Z]) :- append(X, Y, Z).

Rule 2. Try to ensure that execution fails efficiently when no solutions exist. Suppose you want to write a predicate singlepeak that checks the integers in a list to see if, in the order given, they ascend to a single maximum and then descend again. With this predicate, the call: singlepeak([1, 2, 5, 7, 11, 8, 6, 4]).

would succeed, while the call: singlepeak([1, 2, 3, 9, 6, 8, 5, 4, 3]).

would fail. The following definition for singlepeak breaks Rule 3, since the failure of a list to have a single peak is only recognized when append has split the list into every possible decomposition: /* Definition 1 - Breaks Rule 2 */ singlepeak(X) :- append(X1, X2, X), up(X1), down(X2). up[_]. up([U, V|Y]) :- U<V, down([]). down([U]). down([U, V|Y]) :- U>V,

up([V|Y]).

down([V|Y]).

append([], Y, Y). append([U|X], Y, [U|Z]) :- append(X, Y, Z).

On the other hand, the next definition recognizes failure at the earliest possible moment: /* Definition 2 - Follows Rule 2 */

The third and final definition shortens singlepeak even further by observing Rule 1. /* Definition 3 - Follows Rule 1 */ singlepeak([], _). singlepeak([U, V|W], up) :- U<V, singlepeak([V|W], up). singlepeak([U, V|W], _) :- U>V, singlepeak([V|W], down).

Using Definition 3, this call to singlepeak singlepeak(Y, up)

succeeds if Y is bound to a single peaked list appended to an ascending list. This call singlepeak(Y, down)

succeeds if Y is bound to a descending list.

Rule 3. Let Visual Prolog's unification mechanism do as much of the work as possible. At first thought, you might define a predicate equal to test two lists from the same domain for equality as follows: equal([], []). equal([U|X], [U|Y]) :- equal(X, Y).

This is unnecessary. Using the definition equal(X, X).

or, even simpler, unification by means of =, Visual Prolog's unification mechanism does all the work! Chapter 170 Advanced topics

269

Rule 4. Use backtracking--instead of recursion--for repetition. Backtracking decreases stack requirements. The idea is to use the repeat ... fail combination repeat--fail combination instead of recursion. This is so important that the next section is dedicated to the technique.

Using the fail Predicate To evaluate a particular sequence of subgoals repeatedly, it is often necessary to define a predicate like run with a clause of the form subgoals, evaluating repeatedly run :readln(X), process(X, Y), write(Y), run.

This kind of definition incurs unnecessary recursion overheads that can't be automatically eliminated by the system if process(X,Y) is non-deterministic. In this case, the repeat ... fail combination avoids the need for the final recursive call. Given repeat. repeat :- repeat.

you can redefine run without recursion as follows: run :repeat, readln(X), process(X, Y), write(Y), fail.

fail causes Visual Prolog to backtrack to process and eventually to repeat, which always succeeds. But how do you break out of a repeat ... fail combination? Well, in the cases where you want infinite execution (the run:- ..., ..., run variety, you will usually only want to quit if some exceptional condition arises. To this end, you can use the exit predicate in non-interactive programs, or just press break in interactive ones. In other cases, where you have a clear condition of completion, replace the fail with a test for completion:

run:repeat, getstuff(X), process(X,Y), putstuff(Y), test_for_completion(Y), !.

Determinism vs. Non-determinism: Setting the Cut The compiler directive check_determ is useful when you need to decide where to place the cut, since it marks those clauses that give rise to non-deterministic predicates. If you want to make these predicates deterministic, you must insert the cut to stop the backtracking (which causes the non-determinism). As a general rule, in such cases, the cut should always be inserted as far to the left as possible (close to the head of a rule) without destroying the underlying logic of the program. Keep in mind these two rules used by the compiler to decide that a clause is nondeterministic: There is no cut in the clause, and there is another clause that can match the input arguments in the clause head. There is a call to another non-deterministic predicate in the clause body, and this non-deterministic call is not followed by a cut.

Chapter 170 Advanced topics

271

PA RT

Tutorial Chapters 201 -- 202: Using Visual Prolog.203204Visual Prolog

CHAPTER

205

Writing, Reading, and Files 206In this chapter, we first cover the basic set of built-in predicates for writing and reading. Next we describe how the file system works in Visual Prolog and show how you can redirect both input and output to files. We also discuss the file domain and some predefined files.

Writing and Reading In these tutorials, most of the Input/Output has been interactive via screen and keyboard. In this section, we provide formal covererage of the standard predicates you use for I/O, including predicates for file operations.

Writing Visual Prolog includes three standard predicates for writing. These predicates are write, nl and writef. write/* and nl The predicate write can be called with an arbitrary number of arguments:

write(Param1, Param2, Param3, ..., ParamN) /* (i, i, i, ..., i) */

These arguments can either be constants from standard domains or they can be variables. If they're variables, they must be input parameters. The standard predicate nl (for new line) is often used in conjunction with write; it generates a newline on the display screen. For example, the following subgoals: pupil(PUPIL, CL), write(PUPIL," is in the ",CL," class"), nl, write("-----------------------------------").

could result in this display: Helen Smith is in the fourth class ----------------------------------

while this goal: ...., write("List1= ", L1, ", List2= ", L2 ).

could give: List1= [cow,pig,rooster], List2= [1,2,3]

Also, if My_sentence is bound to sentence(subject(john),sentence_verb(sleeps))

in the following program DOMAINS sentence = sentence(subject, sentence_verb) subject = subject(symbol) ; ...... sentence_verb = sentence_verb(verb) ; ...... verb = symbol CLAUSES .... write( " SENTENCE= ", My_sentence ).

you would obtain this display: SENTENCE= sentence(subject(john),sentence_verb(sleeps))

Chapter 170 Advanced topics

273

Note that with respect to strings, the backslash (\) is an escape character. To print the backslash character verbatim, you must type two backslashes. For example, to designate the DOS directory path name A:\PROLOG\MYPROJS\MYFILE.PRO in a Visual Prolog program, you would type a:\\prolog\\myprojs\\myfile.pro. If a backslash is followed by one of a few specially recognized characters, it will be converted to a print control character. These are 'n' 't' 'r'

newline and carriage return tab carriage return

Alternatively, the backslash may be followed by up to three decimal digits, specifying a particular ASCII code. However, avoid putting \0 into a string unless you know what you're doing. Visual Prolog uses the C convention with 0terminated strings. Be very careful with the '\r' option. It sets the current write position back to the start of the current line, but if you accidentally do that in between writing different things, it may happen so quickly that the first thing you write becomes overwritten before you even notice it's there. Also, if you write something which is too long for a single line, causing the output to wrap, the '\r' will set the cursor back to the beginning of the last line, not the beginning of the line where the writing started. Often write does not, by itself, give you as much control as you'd like over the printing of compound objects such as lists, but it's easy to write programs that give better control. The following four small examples illustrate the possibilities. Examples Demonstrating the write Predicate These examples show how you can use write to customize your own predicates for writing such things as lists and compound data structures. 1. Program ch11e01.pro prints out lists without the opening bracket ([) and closing bracket (]). /* Program ch207e01.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS integerlist = integer* namelist = symbol* PREDICATES writelist(integerlist) writelist(namelist).

CLAUSES writelist([]). writelist([H|T]):write(H, " "), writelist(T).

Notice how this program uses recursion to process a list. Load the program and try this goal: writelist([1, 2, 3, 4]).

2. The next example, Program ch11e02.pro, writes out the elements in a list with no more than five elements per line. /* Program ch208e02.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS integerlist = integer* PREDICATES writelist(integerlist) write5(integerlist,integer) CLAUSES writelist(NL):nl, write5(NL,0),nl. write5(TL,5):-!, nl, write5(TL, 0). write5([H|T],N):-!, write(H," "), N1=N+1, write5(T,N1). write5([],_).

If you give the program this goal: writelist([2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22]).

Visual Prolog responds with: 2 4 6 8 10 12 14 16 18 20 22

Chapter 170 Advanced topics

275

3. Frequently, you may want a predicate that displays compound data structures in a more readable form. Program ch11e03.pro displays a compound object like: plus(mult(x, number(99)), mult(number(3), x))

in the form: x*99+3*x

(This is known as infix notation.) /* Program ch209e03.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS expr = number(integer); x; log(expr); plus(expr, expr); mult(expr, expr) PREDICATES writeExp(expr) CLAUSES writeExp(x):-write('x'). writeExp(number(No)):-write(No). writeExp(log(Expr)):write("log("), writeExp(Expr), write(')'). writeExp(plus(U1, U2)):writeExp(U1),write('+'),writeExp(U2). writeExp(mult(U1,U2)):writeExp(U1),write('*'),writeExp(U2).

4. Program ch11e04.pro is similar to Program ch11e03.pro. /* Program ch210e04.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS sentence = sentence(nounphrase, verbphrase) nounphrase = nounp(article, noun); name(name) verbphrase = verb(verb); verbphrase(verb, nounphrase) article, noun, name, verb = symbol PREDICATES write_sentence(sentence) write_nounphrase(nounphrase) write_verbphrase(verbphrase)

CLAUSES write_sentence(sentence(S,V)):write_nounphrase(S),write_verbphrase(V). write_nounphrase(nounp(A,N)):write(A,' ',N,' '). write_nounphrase(name(N)):-write(N,' '). write_verbphrase(verb(V)):-write(V,' '). write_verbphrase(verbphrase(V,N)):write(V,' '),write_nounphrase(N).

Try Program ch11e04.pro with this goal: write_sentence(sentence(name(bill), verb(jumps))).

Exercise Write a Visual Prolog program that, when given a list of addresses contained in the program as clauses of the form: address("Sylvia Dickson", "14 Railway Boulevard","Any Town", 27240).

displays the addresses in a form suitable for mailing labels, such as: Sylvia Dickson 14 Railway Boulevard Any Town 27240

writef/* The writef predicate allows you to produce formatted output; it uses this format: writef(FormatString, Arg1, Arg2, Arg3, ...,ArgN) /* (i, i, i, i, ..., i) */

Arg1 to ArgN must be constants or bound variables belonging to standard domains; it is not possible to format compound domains. The format string contains ordinary characters and format specifiers; ordinary characters are printed without modification, and format specifiers take the following form: %-m.pf

The characters in the format specifiers following the % sign are optional and have these meanings:

Chapter 170 Advanced topics

277

hyphen (-)

Indicates that the field is to be left-justified (right-justified is the default).

m field

Decimal number specifying a minimum field length.

.p field

Decimal number specifying either the precision of a floating-point number or the maximum number of characters to be printed from a string.

f field

Specifies formats other than the default format for the given object. For example, the f field can specify that integers should be printed as unsigned or hexadecimal numbers.

Visual Prolog recognizes the following format specifiers in the f field: f

reals in fixed-decimal notation (such as 123.4 or 0.004321)

reals in exponential notation (such as 1.234e2 or 4.321e-3)

reals in the shorter format of f or e (this is the default for reals)

d,D

integral domains as a signed decimal number

u,U

integral domains as an unsigned decimal integer

o,O

integral domains as an octal number

x,X

integral domains as a hexadecimal number

integral domains as a char

as a string (symbols and strings)

as a database reference number (ref domain only)

as a binary (binary domain only)

as a function pointer

The ref domain will be described in chapter 211, and the binary and function pointer domains in chapter 212. For the integral domain specifiers, an uppercase format letter denotes that the associated object is a long type. If no format letter is given, Visual Prolog will automatically select a suitable format.

Examples of Formatted Output 1. The following example program illustrates the effect of different format specifiers on output formatted with writef. /* Program ch213e05.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ % Note that the format strings in this example specify 16bit integers GOAL A = one, B = 330.12, C = 4.3333375, D = "one two three", writef("A = '%-7' \nB = '%8.1e'\n",A,B), writef("A = '%' \nB = '%8.4e'\n",A,B),nl, writef("C = '%-7.7g' \nD = '%7.7'\n",C,D), writef("C = '%-7.0f' \nD = '%0'\n",C,D), writef("char: %c, decimal: %d, octal: %o, hex: %x",'a','a','a','a').

When run, this program will produce the following output: A = 'one ' B = ' 3.3E+02' A = 'one' B = '3.3012E+02' C = '4.3333375' D = 'one two' C = '4 ' D = 'one two three' char: a, decimal: 97, octal: 141, hex: 61

2. Here's another example, showing how you can use writef to format your output. /* Program ch214e06.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DATABASE person(string,integer,real) CLAUSES person("Pete Ashton",20,11111.111). person("Marc Spiers",32,33333.333). person("Kim Clark",28,66666.666).

Chapter 170 Advanced topics

279

GOAL % Name % Age % Income %

is left-justified, at least 15 characters wide is right-justified, 2 characters wide is right-justified, 9 characters wide, with 2 decimal places, printed in fixed-decimal notation

person(N, A, I), writef("Name= %-15, Age= %2, Income= $%9.2f \n",N,A,I), fail ; true.

This produces the following: Name= Pete Ashton Name= Marc Spiers Name= Kim Clark

, Age= 20, Income= $ , Age= 32, Income= $ , Age= 28, Income= $

11111.11 33333.33 66666.67

Reading Visual Prolog includes several standard predicates for reading. The four basic ones are readln for reading whole lines of characters, readchar for reading single characters/keystrokes, readint for reading integers, and readreal for reading floating point numbers. Additionally, readterm will read any term, including compound objects. These predicates can all be redirected to read from files. Another, more specialized, predicate that belong in the reading category is file_str for reading a whole text file into a string. readln/1 readln reads a line of text; it uses this format: readln(Line)

/* (o) */

The domain for the variable Line will be a string. Before you call readln, the variable Line must be free. readln reads up to 254 characters (plus a carriage return) from the keyboard, up to 64K from other devices. If Esc is pressed during input from the keyboard, readln will fail. readint/1, readreal/1, and readchar/1 readint reads an integer value, using this format: readint(X)

/* (o) */

The domain for the variable X must be of integer type, and X must be free prior to the call. readint will read an integer value from the current input device (probably the keyboard) until the Enter key is pressed. If the line read does not correspond to the usual syntax for integers, readint fails and Visual Prolog invokes its backtracking mechanism. If Esc is pressed during input from the keyboard, readint will fail. readreal does as its name conveys: it reads a real number (as opposed to readint, which reads an integer). readreal uses the following format: readreal(X)

/* (o) */

The domain for the variable X must be real, and X must be free prior to the call. readreal will read a real value from the current input device until the Enter key is pressed. If the input does not correspond to the usual syntax for a real number, readreal fails. If Esc is pressed during input from the keyboard, readreal will fail. readchar reads a single character from the current input device, using this format: readchar(CharParam)

/* (o) */

CharParam must be a free variable before you call readchar and must belong to a domain of char type. If the current input stream is the keyboard, readchar will wait for a single character to be typed before it returns. If Esc is pressed during input from the keyboard, readchar will fail. readterm/2 readterm reads a compound term and converts it to an object; it takes this format: readterm(DomainName, Term)

/* (i, i) */

You call readterm with two arguments: a domain name and a term. readterm reads a line and converts it to an object of the given domain. If the line does not look like write would have formatted the object, readterm gives an error. The standard predicate readtermerror may be used in connection with a trap to produce customized error handling for readterm. See chapter 215. readterm is useful for handling terms in text files. For example, you can implement you own version of consult. file_str/2 file_str reads characters from a file and transfers them to a variable, or creates a file and writes the string into the file. It uses this format: file_str(Filename, Text)

Chapter 170 Advanced topics

/* (i, o), (i, i) */

281

If, before your program calls file_str, the variable Text is free, file_str reads the entire contents of the file Filename into Text. In the DOS-related versions of Visual Prolog, an eof character (Ctrl+Z) will terminate reading when encountered and will not be included in the string. For example, the call file_str("t.dat", My_text)

binds My_text to the contents of the file t.dat. The file size can't exceed the maximum length of a string, which is 64 Kbytes on the 16-bit platforms. If the file exceeds the maximum size, file_str will return an error message. With My_text bound to the text in "t.dat", file_str("t.bak", My_text) will create a file called t.bak that contains the text from "t.dat". If t.bak already exists it will be overwritten. Examples These examples demonstrate how you can use the standard reading predicates to handle compound data structures and lists as input. 1. Program ch11e07.pro illustrates assembling of compound objects from individually read pieces of information. /* Program ch216e07.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS person = p(name, age, telno, job) age = integer telno, name, job = string PREDICATES readperson(person) run CLAUSES readperson(p(Name,Age,Telno,Job)):write("Which name ? "), readln(Name), write("Job ? "), readln(Job), write("Age ? "), readint(Age), write("Telephone no ? "), readln(Telno). run :readperson(P),nl,write(P),nl,nl, write("Is this compound object OK (y/n)"), readchar(Ch),Ch='y', !.

run :nl,write("Alright, try again"),nl,nl,run. GOAL run.

2. This next example reads one integer per line until you type a non-integer (such as the X key); then readint will fail and Visual Prolog displays the list. /* Program ch217e08.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS list = integer* PREDICATES readlist(list) CLAUSES readlist([H|T]):write("> "), readint(H),!, readlist(T). readlist([]). GOAL write("*************** Integer List *****************"),nl, write(" Type in a column of integers, like this:"),nl, write(" integer (press ENTER)\n integer (press ENTER)\n"), write(" etc.\n\n Type X (and press ENTER) to end the list.\n\n"), readlist(TheList),nl, write("The list is: ",TheList).

Load Program ch12e08.pro and run it. At the prompt, enter a column of integers (such as 1234 Enter 567 Enter 89 Enter 0 Enter), then press X Enter to end the list. Exercise Write and test clauses for a predicate readbnumb, which, in the call: readbnumb(IntVar)

converts a user-input, 16-bit binary number like "1111 0110 0011 0001" to a corresponding integer value to which IntVar is bound. Check your work by writing a program that contains readbnumb. Chapter 170 Advanced topics

283

Binary Block Transfer Three standard predicates allow reading and writing of binary blocks, or byte sequences of a given length. They all use the binary standard domain. This emulates an array, with a word (dword on the 32-bit versions of Visual Prolog) in front, holding the size. For a complete description of the binary domain, see chapter 218. readblock/2 readblock has the following format: readblock(Length,BTerm)

/* (i, o) */

where Length is the number of bytes to read and BTerm is the returned binary term. As described in chapter 11, there are no restrictions on the contents of a binary term, and it will contain whatever was read from the file including zeros and DOS eof-markers (Ctrl+Z). The current input device must be assigned to a file (see readdevice). If Length = 0 is specified, the readblock attempts to read maximum possible number of bytes from an input device. (Remember that BinBlock < 64 K on 16-bit platforms). If Length is larger than the actual remaining number of bytes in the file - then the readblock generates the run-time error 1111: "Wrong number of bytes read from file". writeblock/2 writeblock complements readblock: writeblock(Length,BTerm)

/* (i, i) */

As with readblock, there are no restrictions on the contents of BTerm. The Length specifies how many bytes to write; a length of 0 will write the whole term. For compatibility with previous versions of Visual Prolog, where binary blocks were disguised as strings, writeblock may be called with a string argument instead of a binary term. In this case, it is imperative that Length contains the correct number of bytes to write. file_bin/2 file_bin will read a whole file into a binary term, and vice versa. It takes two arguments, the filename and the binary term:

file_bin(FileName,BinTerm)

/* (i, o) (i, i) */

Visual Prolog's File System In this section, we give you a look at Visual Prolog's file system and the standard predicates relevant to files. We also introduce I/O redirection, an efficient method for routing input and output to various devices. With a few exceptions, the file system works identically in the different versions of Visual Prolog. Visual Prolog uses a current_read_device, from which it reads input, and a current_write_device, to which it sends output. Normally, the keyboard is the current read device, and the screen is the current write device. However, you can specify other devices. For instance, input could be read from an externally stored file (on disk perhaps). Not only can you specify other devices, you can even reassign the current input and output devices while a program is running. Regardless of what read and write devices you use, reading and writing are handled identically within a Visual Prolog program. To access a file, you must first open it. A file can be opened in one of four ways: for reading for writing for appending for modification A file opened for any activity other than reading must be closed when that activity is finished, or the changes to the file might be lost. You can open several files simultaneously, and input and output can be quickly redirected between open files. In contrast, it takes much longer to open and close a file than to redirect data between files. When Visual Prolog opens a file, it connects a symbolic name to the actual file name. Visual Prolog uses this symbolic name when directing input and output. Symbolic file names must start with a lower-case letter and must be declared in the file domain declaration like this: file = file1; source; auxiliary; somethingelse

Only one file domain is allowed in any program. Visual Prolog recognizes five predefined file alternatives: keyboard Chapter 170 Advanced topics

reading from the keyboard (default) 285

screen

writing to the screen

stdin

reading from standard input

stdout

writing to standard output

stderr

writing to standard error

These predefined alternatives must not appear in the file domain declaration; they don't need to be opened and they should not be closed. Note, that when using the VPI strategy, only the screen alternative can be used.

Opening and Closing Files The following sections describe the standard predicates for opening and closing files. Note: In the DOS-related versions of Visual Prolog, remember that the backslash character, used to separate subdirectories, is an escape character. You must always use two backslash characters when you give a path in the program; for example, the string "c:\\prolog\\include\\iodecl.con"

represents the path name c:\prolog\include\iodecl.con

openread/2 openread opens the file OSFileName for reading, using this format: openread(SymbolicFileName, OSFileName)

/* (i, i) */

Visual Prolog refers to the opened file by the symbolic name SymbolicFileName declared in the file domain. If the file can't be opened, Visual Prolog returns an error message. openwrite/2 openwrite opens the file OSFileName for writing; it takes this format: openwrite(SymbolicFileName, OSFileName)

/* (i,i) */

If the file already exists, it is erased. Otherwise, Visual Prolog creates a new file and makes an entry in the appropriate directory. If the file can't be created, the predicate exits with an error message.

openappend/2 openappend opens the file OSFileName for writing at the end, using this format: openappend(SymbolicFileName, OSFileName)

/* (i, i) */

If the file can't be opened for write access, Visual Prolog issues an error message. openmodify/2 openmodify opens the file OSFileName for both reading and writing; if the file already exists, it won't be overwritten. openmodify takes this format: openmodify(SymbolicFileName, OSFileName)

/* (i, i) */

If the system can't open OSFileName, it issues an error message. openmodify can be used in conjunction with the filepos standard predicate to update a randomaccess file. filemode/2 When a file has been opened, filemode sets the specified file to text mode or binary mode, using this format: filemode(SymbolicFileName, FileMode)

/* (i, i) */

If FileMode = 0, the file specified by SymbolicFileName is set to binary mode; if FileMode = 1, it's set to text mode. In text mode, newlines are expanded to carriage- return/line-feed pairs during writes, and carriage-return/line-feed pairs are converted to newlines during reads. Carriage return Line feed

= ASCII 13 = ASCII 10

In binary mode, no expansions or conversions occur. To read a binary file, you can only use readchar or the binary file-access predicates discussed in chapter 219. filemode is only relevant in the DOS-related versions of Visual Prolog. In the UNIX version it has no effect. Don't confuse filemode with the peculiarly named DosQFileMode and DosSetFileMode OS/2 primitives - they get or change the attributes of physical disk files. closefile/1 closefile closes the indicated file; it takes this format: Chapter 170 Advanced topics

287

closefile(SymbolicFileName)

/* (i) */

This predicate always succeeds, even if the file has not been opened. readdevice/1 readdevice either reassigns the current_read_device or gets its name; the predicate takes this format: readdevice(SymbolicFileName)

/* (i), (o) */

readdevice reassigns the current read device if SymbolicFileName is bound and has been opened for reading. If SymbolicFileName is free, readdevice binds it to the name of the current active read device. writedevice/1 writedevice either reassigns or gets the name of the current_write_device; it takes this format: writedevice(SymbolicFileName)

/* (i), (o) */

writedevice reassigns the current write device if the indicated file has been opened for either writing or appending. If SymbolicFileName is free, writedevice binds it to the name of the current active write device. Examples 1. The following sequence opens the file mydata.fil for writing, then directs all output produced by clauses between the two occurrences of writedevice to that file. The file is associated with the symbolic file name destination appearing in the declaration of the file domain. DOMAINS file = destination GOAL openwrite(destination, "mydata.fil"), writedevice(OldOut), /* gets current output device */ writedevice(destination), /* redirects output to the file */ : : writedevice(OldOut), /* resets output device */

2. Program ch11e09.pro uses some standard read and write predicates to construct a program that stores characters typed at the keyboard in the file

tryfile.one. Characters typed are not echoed to the display; it would be a good exercise for you to change the program so that characters are echoed. The file is closed when you press the # key. /* Program ch220e09.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS file = myfile PREDICATES readloop CLAUSES readloop:readchar(X), X<>'#',!, write(X), readloop. readloop. GOAL write("This program reads your input and writes it to"),nl, write("tryfile.one. For stop press #"),nl, openwrite(myfile,"tryfile.one"), writedevice(myfile), readloop, closefile(myfile), writedevice(screen), write("Your input has been transferred to the file tryfile.one"),nl.

Redirecting Standard I/O The file domain has three additional options: stdin, stdout, and stderr. The advantage of these file streams is that you can redirect I/O at the command line. stdin

Standard input is a read-only file; the keyboard, by default. readdevice(stdin) directs the input device to stdin.

stdout

Standard output is a write-only file that defaults to the screen. writedevice(stdout) directs the output device to stdout.

stderr

Standard error is a write-only file that defaults to the screen. writedevice(stderr) directs the output device to stderr.

Chapter 170 Advanced topics

289

Working with Files In this section, we describe several other predicates used for working with files; these are filepos, eof, flush, existfile, searchfile, deletefile, renamefile, disk, and copyfile. filepos/3 filepos can control the position where reading or writing takes place; it takes the form filepos(SymbolicFileName, FilePosition, Mode)

% (i, i, i), (i, o, i)

With FilePosition bound, this predicate can change the read and write position for the file identified by SymbolicFileName. It can return the current file position if called with FilePosition free. FilePosition is a long value. Mode is an integer and specifies how the value of FilePosition is to be interpreted, as shown in Table 221.1. Table 222.2: Mode and FilePosition223 Mode

FilePosition

Relative to the beginning of the file.

Relative to current position.

Relative to the end of the file. (The end of the file is position 0.)

When returning FilePosition, filepos will return the position relative to the beginning of the file irrespective of the value of Mode. Note: In the DOS-related versions of Visual Prolog, filepos does not consider files in text mode to be different from files in binary mode. No translation of DOS newline conventions takes place, and a newline in a file following DOS newline conventions consists of two characters. Example 1. The following sequence writes the value of Text into the file somefile.pro (referred to by Prolog as myfile), starting at position 100 (relative to the beginning of the file).

Text = "A text to be written in the file", openmodify(myfile, "somefile.pro"), writedevice(myfile), filepos(myfile, 100, 0), write(Text), closefile(myfile).

2. Using filepos, you can inspect the contents of a file on a byte-by-byte basis, as outlined in Program ch11e10.pro. This program requests a file name, then displays the contents of positions in the file as their position numbers are entered at the keyboard. /* Program ch224e10.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS file = input PREDICATES inspect_positions(file) CLAUSES inspect_positions(UserInput):readdevice(UserInput), nl,write("Position No? "), readln(X), term_str(ulong,Posn,X), readdevice(input), filepos(input,Posn,0), readchar(Y),nl, write("Char is: ",Y), inspect_positions(UserInput). GOAL write("Which file do you want to work with ?"),nl, readln(FileName), openread(input, FileName), readdevice(UserInput), inspect_positions(UserInput).

eof/1 eof checks whether the file position is at the end of the file, in which case eof succeeds; otherwise, it fails. eof has the form eof(SymbolicFileName)

Chapter 170 Advanced topics

/* (i) */

291

eof gives a run-time error if the file has been opened with write-only access. Note that it doesn't consider a DOS eof character (Ctrl+Z) to have any particular meaning. Example eof can be used to define a predicate repfile that's handy when operating with files. repfile generates backtrack points as long as the end of the file has not been reached. PREDICATES repfile(FILE) CLAUSES repfile(_). repfile(F):- not(eof(F)), repfile(F).

The following program converts one file to another where all the characters are upper-case. /* Program ch225e11.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS file = input; output PREDICATES convert_file nondeterm repfile(FILE) CLAUSES convert_file :repfile(input), readln(Ln), upper_lower(LnInUpper,Ln), write(LnInUpper),nl, fail. convert_file. repfile(_). repfile(F):not(eof(F)), repfile(F).

/* converts the string to uppercase */

GOAL write("Which file do you want convert ?"), readln(InputFileName),nl, write("What is the name of the output file ?"), readln(OutputFileName),nl, openread(input, InputFileName), readdevice(input), openwrite(output, OutputFileName), writedevice(output), convert_file, closefile(input), closefile(output).

flush/1 flush forces the contents of the internal buffer to be written to the named file. It takes this format: flush(SymbolicFileName)

/* (i) */

flush also requests the operating system to flush its buffers. For versions of DOS previous to 3.30, this entails closing and re-opening the file. For newer versions of DOS, as well as the other platforms, the appropriate operating system function is called. existfile/1 existfile succeeds if OSFileName exists. It takes this format: existfile(OSFileName)

/* (i) */

where OSFileName may contain a directory path and the name itself may contain wildcards, e.g. c:\psys\*.cfg. existfile fails if the name does not appear in the directory. However, note that although existfile finds all files, including those with the 'system' and 'hidden' attribute set, it doesn't find directories. This may be accomplished using the directory search predicates described later on. You can use the following sequence to verify that a file exists before attempting to open it. open(File, Name) :existfile(Name), !, openread(File, Name). open(_, Name) :write("Error: the file ", Name," is not found").

Chapter 170 Advanced topics

293

existfile/2 In UNIX, existfile is also available in a two- arity version: existfile(OSFileName,AccessMode)

/* (i, i) */

with AccessMode specifying the type of access desired. This should be one of the following constants: f_ok to test for existence x_ok to test for execute permission w_ok to test for write permission r_ok to test for read permission These constants are declared in the include file iodecl.con. existfile with only one argument tests for file-existence only. searchfile/3 searchfile is used to locate a file along a path list, and is a kind of automated existfile. It takes three arguments, as follows: searchfile(PathList,Name,FoundName)

/* (i,i,o) */

The PathList is a string containing one or more paths, separated by semicolons (or colons, for UNIX), and Name is the name of the file to locate. If found, FoundName will be bound to the fully qualified name, otherwise searchfile will fail. For instance, for DOS SearchFile(".;..;C:\\","autoexec.bat",FoundName),

will - provided autoexec.bat is located in the root of drive C - set FoundName to C:\AUTOEXEC.BAT. The file name may contain wildcards. In that case, FoundName is bound to the fully qualified wildcard name, which may subsequently be used as input to the directory matching predicates described later on. For instance, if the name is specified as *.bat instead of autoexec.bat in the above example, FoundName will be bound to C:\*.BAT. deletefile/1 deletefile removes the file specified by its argument: deletefile(OSFileName)

/* (i) */

deletefile gives an error if it can't remove the file. The OSFileName can not contain wildcards. renamefile/1 renamefile renames the file OldOSFileName to NewOSFileName. It takes this format: renamefile(OldOSFileName, NewOSFileName)

/* (i, i) */

renamefile succeeds if a file called NewOSFileName doesn't already exist and both names are valid file names; otherwise, it gives an error. disk/1 disk is used to change the current disk and/or directory; it takes this format: disk(Path)

/* (i) (o) */

Called with a free variable, disk will return the current directory. In the DOSrelated versions, to change to another disk without changing the existing current directory on that disk, use D:. where D is the drive letter. copyfile/2 copyfile is used to copy a file. It takes two file names as follows: copyfile(SourceName,DestinationName)

/* (i,i)*/

The names may be partly or fully qualified file names, including disks and directories. However, no wildcards are allowed. The copied file will have the same attributes (and permissions) as those of the source.

File Attributes Although the standard file open predicates described previously cover all general cases, there may be a need to open or create files with specialized attributes and non-obvious sharing modes. To this end Visual Prolog incorporates a general purpose open predicate, but before discussing that we need to look at file attributes and sharing modes. The attributes and access modes used by Visual Prolog use the same values as your operating system, with the exception of the default ('normal') attribute in the NonUNIX-related versions of Visual Prolog. However, for easy porting to other environments, you should avoid coding inherently non-portable constructs such as file attributes (and even the fact that files have attributes) all over an Chapter 170 Advanced topics 295

application. Rather, wrap things up nicely and write your own intermediate level of predicates, getting and setting information in transparent ways. The attributes and sharing modes are found in the include file IODECL.CON. Opening and creating files When opening or creating a file, the OS needs to know the file's attributes (e.g. 'hidden'), the type of use or access (e.g. 'read'), and how the file may be shared with other programs while open (e.g. 'deny write'). Don't confuse these - they are three different pieces of information, only partly related: Attributes The file attributes are the permanent attributes relating to the physical file on disk, whether currently in use by a program or not. In DOS and OS/2 there's only a few attributes, such as 'read only' and 'hidden'. These attributes inform the operating system about how it may handle the file. Network and multiuser operating systems, such as UNIX, typically have a much wider range of attributes. These may include access allowed by other users (e.g. 'execute-only', no read or write, giving copy-protection) and direct instructions to the OS ('this is an executable program'). The attributes have no effect when opening an existing file, as files are unique based on names only. They only apply when creating a new file. The standard predicates described in the previous section all reference 'normal' files. However, when a file has been modified the archive bit will automatically be set by the operating system when the file is closed. Access Modes The access modes indicate how the file will be used. The OS will combine this information with the files physical attributes, to determine if the access requested is acceptable. For instance, opening a file having the read-only physical attribute set, with either fm_access_wo or fm_access_rw will not be accepted. Sharing Modes The sharing modes indicate how this process views sharing with other processes. The OS will combine the sharing and access modes with the sharing and access modes specified by other processes, if the file is already in use, to determine if the open call should succeed. If successful, the modes will restrict future open attempts.

Note that conceptually the sharing and access modes work both ways to form a combined set of restrictions on the file: they specify both what the process wants from a file and what it will allow from other processes. For instance, if a file has been opened with 'deny write' and 'read only', an open attempt with 'deny none' and 'write only' will fail because the first process has specified 'deny write' - in this case it is the existing restriction on the file that rejects the open attempt. On the other hand, an open attempt with 'deny read' and 'read only' will fail because the file is already open with read access - in this case it is the current requirement that rejects the open attempt. Note that the fm_sh_denyrw denies all modes from other processes; it doesn't mean 'deny readwrite, but allow read-only or write-only'. All the standard predicates described in the previous section specify the sharing mode as 'deny write'. Special File Modes for OS/2, DOS >= 4.0 and UNIX OS/2 and DOS versions greater than or equal to 4.0, have a special fm_returnerr mode: The fm_returnerr specify that "media" errors occurring after the file has been opened should return an error to the program, rather than be reported through a pop-up window. Media errors are those indicating a malfunction of the device, e.g. if writing to a floppy and the drive door is opened - this generates the wellknown 'Drive not ready' error. The standard predicates described in the previous section do not specify fm_returnerr, so media errors will generate a pop-up through OS/2's critical error handler. UNIX, OS/2 and DOS >= 4.0 also have a write- through mode: The fm_writethru specifies that no write buffering should take place. In this case, every single byte written to the file cause both the physical disk image and the directory entry to be updated, giving a secure file. However, disk writes performed with write-through are excessively slow compared to ordinary writes. openfile/5 With the general-purpose openfile predicate, files may be created or opened in nonstandard ways. openfile looks like this: openfile(SymbolicName,OSName,OpenMode,Attributes,Creation) /* (i,i,i,i,i) */

Chapter 170 Advanced topics

297

The SymbolicName and OSName are the same as for the previously described standard predicates. The rest of the arguments are as follows (please refer to the section on File Attributes a few pages back): OpenMode is the access and sharing modes for the file. It is formed by adding together one of the fm_access_XX values, one of the fm_sh_XXXXXX and optionally fm_returnerr and fm_writethru. If no access mode is specified, it will be set to 'read only'. If no sharing mode is specified, it will be set to 'deny write'. Attributes are the attributes for the physical disk file. Valid attributes on DOS and OS/2 are fa_rdonly, fa_hidden, fa_system, fa_arch and fa_normal. If nothing (0) is specified, the attributes will be set to fa_normal. The system and the hidden attributes both have the same effect, namely to hide the file when a 'dir' command is executed. Note that DOS and OS/2 automatically set the archive attribute when an updated file is closed. For UNIX, the attributes correspond to the file's permissions. Creation specifies how the presence or absence of an existing file with the same name is to be handled. It is formed by adding at most one from the cr_ex_XX group and at most one from the cr_noex_XX group. Pay attention to Creation defaults - if nothing (0) is specified. Note that this is the equivalent of specifying cr_ex_fail and cr_noex_fail, i.e. fail if it exists and fail if it doesn't exist. But remember that the actual default Creation action will be set according to the access mode as follows: fm_access_ro fm_access_wo fm_access_rw

-> -> ->

cr_ex_open + cr_noex_fail cr_ex_replace + cr_noex_create cr_ex_open + cr_noex_create

A sensible Creation default for read-write access is a bit tricky: If readwrite is specified because the file is opened for 'modification', an existing file of the same name should be opened, not replaced. This is therefore the default. However, if read-write is specified because one wants bidirectional access to a new file, an existing file of the same name should be deleted. This is possible with a call to openfile as follows: : FMode = fm_access_rw + fm_sh_denywr + fm_returnerr, FCrea = cr_ex_replace + cr_noex_create, openfile(dbfile,"salient.dba",FMode,fa_normal,FCrea), :

File and Path Names A set of standard predicates ease file name handling and enable searching for files on a disk. filenamepath/3 filenamepath is used to compose and decompose a fully qualified name around its path and file name. It takes three arguments, as follows: filenamepath(QualName,Path,Name)

/* (i,o,o)

(o,i,i)*/

filenamepath converts between QualName on one side, and Path and Name on the other. The programs ch11e12.pro and ch11e13.pro contain examples for DOS and UNIX respectively; both examples do essentially the same thing: /* Program ch226e12.pro */ GOAL QualName="c:\\vip\\bin\\prolog.err", FileNamePath(QualName,Path,Name), write("\nQualName=",QualName), write("\nPath=",Path), write("\nName=",Name), FileNamePath(NewName,Path,Name), write("\nConverted back: ",NewName),nl. /* Program ch227e13.pro */ GOAL QualName="/usr/bin/prolog.err", FileNamePath(QualName,Path,Name), write("\nQualName=",QualName), write("\nPath=",Path), write("\nName=",Name), FileNamePath(NewName,Path,Name), write("\nConverted back: ",NewName),nl.

This will set Path to C:\VIP\BIN and name to PROLOG.ERR; finally, NewName will be set to C:\VIP\BIN\PROLOG.ERR. Note that under DOS, all Visual Prolog file name handling converts the name to upper case. This is because there has in the past been confusion with respect to upper and lower case versions of some foreign characters. Chapter 170 Advanced topics

299

Please, using the (o,i,i) flow pattern of this predicate, take into account some special cases described in the filenamepath topic in VDE help. filenameext/3 filenameext is used to compose and decompose a (fully qualified) file name around its extension, defined by a dot. It takes three arguments, as follows: filenameext(Name,BaseName,Ext)

/* (i,o,o)

(o,i,i)*/

Here is the DOS example: /* Program ch228e14.pro */ GOAL Name="c:\\vip\\bin\\win\\16\\vip.exe", FileNameExt(Name,BaseName,Ext), write("\nName=",Name), write("\nBaseName=",BaseName), write("\nExt=",Ext), FileNameExt(NewName,BaseName,Ext), write("\nConverted back: ",NewName), % Override the old extension FileNameExt(NewName1,"VIP.EXE",".HLP"), write("\nNewName1=",NewName1),nl.

This will set BaseName to C:\VIP\BIN\WIN\16\VIP and Ext to .EXE; then NewName is set to C:\VIP\BIN\WIN\16\VIP.EXE and finally NewName1 demonstrates a direct extension override - it isn't necessary to explicitly remove the old extension first. Note that the dot is considered a part of the extension and that - as with filenamepath - in the DOS version, everything is converted to upper case.

Directory Searching Visual Prolog includes directory search predicates, enabling file name matching with wildcards. In addition, the predicates return all relevant information about the directory entries found. Directory searching is very file system dependent and you should therefore guard yourself against future changes by isolating these predicates when they're used. Don't spread them all over your application, and don't rely on their arguments and functionality remaining unchanged. In fact, don't even rely on them being the same for different versions of OS/2: the installable file system concept means that

future versions of OS/2 may behave very differently, although great effort will be spent in attempts to provide portability. Below, all references to OS/2 refer to the FAT file system. Basically, to find matching files the directory has to be opened; this is done with the diropen predicate, specifying the file name mask and the attributes to look for. Then, by calling the dirmatch predicate, the matching files are found one by one. Finally, the directory is closed by a call to the dirclose predicate. Generally, the predicates behave identically irrespective of platform: a file name optionally containing wildcards - is used to specify the names to match, and a set of search attributes refine the match (for a list of attributes, see the section on File Attributes earlier in this chapter). However, unlike the DOS and OS/2 directory search mechanisms, the search attributes don't increase the search beyond 'normal' files. Visual Prolog considers all attributes as strictly informational, and they may all be used for file selection. When using the directory search predicates, you should therefore specify the attributes you are interested in: if you for instance want everything with the archive bit set, specify fa_arch; if you want everything with the system bit set, specify fa_system; if you want 'normal' files, specify fa_normal, etc. You should be aware, though, that the attributes specified are inclusive of each other: if several attributes are combined, the directory search will find everything matching at least one of the attributes, but the entry found won't necessarily match all the attributes. In other words, using set terminology, it is the union of the matching files, not the intersection, which is returned. Exactly what is found may be determined by bitwise testing of the returned attribute. UNIX users should be aware that only one kind of directory entry (such as a normal file, a pipe, a directory, etc.) may be searched for at a time. No permissions of the entries are considered, and none should be specified. diropen/3 diropen is used to gain access to a directory. It takes the following format: diropen(Wild,Attrib,Block)

/* (i,i,o) */

where Wild is a file name, optionally containing wildcards, Attrib are the required search attributes, and Block is an information block used by subsequent calls to dirmatch. To the compiler this block looks like a string, but it contains more information than meets the eye. Therefore, it cannot be asserted in a database and then retracted for use at a later stage - as long as the directory is open, it must be held in a local variable (or an argument, which is the same thing). diropen will fail if there are no files matching the specification; however, if the file name includes a directory which doesn't exist, diropen will exit with an error. Chapter 170 Advanced topics

301

Several diropens may be active simultaneously; in other words, they may be nested and used recursively. dirmatch/10 dirmatch will, after diropen has returned an information block, return the name and other information for each matching file, one at each call. It looks as follows: dirmatch(Block,Name,Attr,Hour,Min,TwoSec,Year,Month,Day,Size) /* (i,o,o,o,o,o,o,o,o,o) */

The Block is the information block returned by diropen, Name is the matching name, and Attr are the attributes for the entry found. The rest of the arguments should be self-explanatory - they're all unsigned integers, apart from Size, which is an unsigned long. Note that DOS and OS/2 use only 5 bits to encode the seconds part of the time stamp, giving at most 32 different values - hence the TwoSec name. Upon each call, dirmatch returns the next matching directory entry. When there are no more matches, dirmatch fails; if this happens, dirmatch will automatically close the directory. You should be aware that if searching for subdirectories with a name specification of e.g. "*.*", dirmatch will always return the entries "." and ".." if these are returned by the operating system. Therefore, dirmatch is likely to find directories in all directories except perhaps the root. dirclose/1 dirclose will close a previously opened directory. It takes the form: dirclose(Block)

/* (i) */

where Block is the information block returned by diropen. Note that if dirmatch is repeatedly called until it fails (because there are no more matching files), dirclose should not be called, as dirmatch will have closed the directory. Example The following demonstrates the use of the directory matching predicates, to make an existdir predicate to complement the existfile standard predicate described previously.

/* Program ch229e16.pro */ include "iodecl.con" PREDICATES existdir(string) exd1(string) exd2(string,string) CLAUSES existdir(Wild):diropen(Wild,fa_subdir,Block), exd1(Block), dirclose(Block). exd1(Block):dirmatch(Block,Name,_,_,_,_,_,_,_,_), exd2(Block,Name). exd2(_,Name):not(frontchar(Name,'.',_)),!. exd2(Block,_):exd1(Block).

Given for instance the goal existdir("c:\\*.*") in DOS, it will - unless you have a rather extraordinary disk organization - say 'yes'. However, it will only find subdirectories in existing paths - if you ask for e.g. existdir(c:\\jnk\\*.*") without having a directory called 'JNK' in the root of drive c, it will exit with an error. You should also be aware that in DOS the root itself can't be matched: there is no directory called '\', and existdir("c:\\") will fail. This is an operating system defined restriction of DOS, and is not relevant in UNIX where '/' does exist. Note, by the way, how the current and parent directory entries ("." and "..") are filtered out in the example. dirfiles/11 Having presented the hard way of finding files, here's the easy way. dirfiles is a non- deterministic standard predicate which, upon backtracking, returns all matching files one by one. It looks as follows: dirfiles(Wild,Attrib,Fnam,RetAttr,Hour,Min,Sec, Year,Month,Day,Size) /* (i,i,o,o,o,o,o,o,o,o,o) */

The use of dirfiles obviates the need to open and close the directory as this is handled automatically, but there is a condition attached: in order to use it Chapter 170 Advanced topics

303

correctly, it must be backtracked into until it fails. It is the final failure of the predicate which automatically closes the directory. You should be aware that neither the compiler nor the code supporting a running program has any way of detecting if this won't happen - it is entirely the programmers responsibility. Having said that, no serious harm will come from leaving a couple of directories open, but eventually the system will run out of handles. As with diropen, calls to dirfiles may be nested and used recursively. DOS Example Below is a sample program which will traverse all directories on drive C, searching for entries having the 'system' or 'hidden' attribute set. The OS will typically have a couple of hidden files in the root directory. However, if there are hidden files elsewhere on the disk, be suspicious! They're probably harmless copy- protection or configuration files for commercial software you have installed, but why hide any files? /* Program ch230e17.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ CONSTANTS fa_hidden = $02 fa_system = $04 fa_subdir = $10 fa_hidsys = $06

/* Hidden file /* System file /* Subdirectory /* hidden + system */

PREDICATES findhidden(string,string) wrattr(integer) CLAUSES wrattr(A):bitand(A,fa_hidden,AA), AA<>0,write('H'),fail. wrattr(A):-bitand(A,fa_system,AA), AA<>0,write('S'),fail. wrattr(A):bitand(A,fa_subdir,AA), AA<>0,write('D'),fail. wrattr(_).

*/ */ */

findhidden(CurrPath,Wild):write(CurrPath,":\n"), filenamepath(FileSpec,CurrPath,Wild), dirfiles(FileSpec,fa_hidsys,FileName,RetAttr,_,_,_,_,_,_,_), wrattr(RetAttr), write('\t',FileName,'\n'), fail. findhidden(CurrPath,Wild):filenamepath(DirSpec,CurrPath,"*.*"), dirfiles(DirSpec,fa_subdir,Name,_,_,_,_,_,_,_,_), not(frontchar(Name,'.',_)), filenamepath(DirName,CurrPath,Name), findhidden(DirName,Wild), fail. findhidden(_,_). GOAL findhidden("C:\\","*.*").

This example also demonstrates decoding the returned attribute (in the wrattr predicate), by means of bitwise testing.

Manipulating File Attributes A standard predicate enables getting and setting the (informational) attributes of files. Although documentation for DOS, OS/2 and MS Windows frequently talks about a "directory attribute", a file cannot be changed to a directory just by clearing this attribute. fileattrib/2 Depending on dataflow fileattrib will get or set the attributes for a file. In UNIX this corresponds to the file mode, meaning permissions, sticky bits, etc; see chmod(S). fileattrib(Name,Attrib)

/* (i,o) (i,i) */

The Name must specify an existing file, otherwise fileattrib exits with an error. Note that the definition of getting or setting attributes is entirely operating system defined; in particular, you cannot set the file attributes for a directory. The attributes for the file appear in Attrib as an unsigned short. This may then be decomposed and/or changed using bitwise manipulation. For instance, the following will clear the system attribute for the DOS file "JNK": Chapter 170 Advanced topics

305

CONSTANTS fa_system = $04 fa_notsys = $FFFB

/* System file /* ~system file.

*/ */

GOAL fileattrib("jnk",FA), bitand(FA,fa_notsys,Plain), fileattrib("jnk",Plain).

The constant fa_notsys is the bitwise negation of fa_system. If you don't know how to find the negation, use the bitxor (see chapter 16) standard predicate: CONSTANTS fa_system = $04

/* System file

GOAL bitxor(fa_system,$FFFF,NotSys), fileattrib("jnk",FA), bitand(FA,NotSys,Plain), fileattrib("jnk",Plain).

Handling terms in text files The readterm predicate makes it possible to access facts in a file. readterm can read any object written by the write predicate and takes the form readterm(<name>,TermParam).

where <name> is the name of a domain. The following code excerpt shows how readterm might be used. DOMAINS name,addr = string one_data_record = p(name, addr) file = file_of_data_records PREDICATES person(name, addr) moredata(file)

CLAUSES person(Name,Addr) :openread(file_of_data_records, "dd.dat"), readdevice(file_of_data_records), moredata(file_of_data_records), readterm(one_data_record, p(Name, Addr)). moredata(_). moredata(File) :not(eof(File)), moredata(File).

If the file DD.DAT contains facts belonging to the one_data_record domain, such as p("Peter","28th Street") p("Curt","Wall Street")

the following is an example of a dialog that retrieves information from that file: Goal person("Peter",Address). Address="28th Street" 1 Solution Goal person("Peter","Not an address"). no

Manipulating Facts Like Terms Facts that describe database predicates can also be manipulated as though they were terms. This is made possible by the dbasedom domain (which is automatically declared by the Visual Prolog system) or any user-defined database domain name. Visual Prolog generates one alternative for each predicate in the database. It describes each database predicate by a functor and by the domains of the arguments in that predicate. For example, given this database declaration: DATABASE person(name, telno) city(cno, cname)

Visual Prolog generates the corresponding dbasedom domain: DOMAINS dbasedom = person(name, telno) ; city(cno, cname)

Chapter 170 Advanced topics

307

A named database similarly generates a domain corresponding to the database name, as in the following example: DATABASE - test person(name, telno) city(cno, cname)

generates the domain DOMAINS test = person(name, telno) ; city(cno, cname)

These domains can be used like any other predefined domain. Example The following example shows how you could construct a predicate my_consult, similar to the standard predicate consult. /* User-defined Predicate my_consult using readterm */ DOMAINS file = dbase DATABASE - dba1 /* ... Declare database predicates to be read from file */ PREDICATES my_consult(string) repfile(file) CLAUSES my_consult(FileName) :openread(dbase, FileName), readdevice(dbase), repfile(dbase), readterm(dba1, Term), assertz(Term), fail. my_consult(_) :- eof(dbase). repfile(_). repfile(F):-not(eof(F)),repfile(F).

If, for example, the database program section contains the declaration p(string, string)

and a file called DD.DAT exists (with contents as described on page 329), you could obtain the following dialog: Goal my_consult("dd.dat"). yes Goal p(X,Y). X=Peter, Y=28th Street X=Curt", Y=Wall Street 2 Solutions

Summary These are the important points covered in this chapter: 1. Visual Prolog provides three standard predicates for basic writing: a. write (for plain output) b. writef (for output formatted according to format specifiers) c. nl (for generating new lines) 2. Visual Prolog basic standard predicates for reading are a. readln (for reading whole lines of characters) b. readint, readreal, and readchar (for reading integers, reals, and characters, respectively) c. readterm (for reading compound objects) d. file_str (for reading a whole text file into a string) 3. Additionally, binary blocks may be transferred using a. readblock (reads a binary block from a file) b. writeblock (writes a binary block to a file) c. file_bin transfers between whole files and binary blocks 4. Visual Prolog uses a current_read_device (normally the keyboard) for reading input, and a current_write_device (normally the screen) for sending output. You can specify other devices, and can reassign the current input and output devices at run time. (This reassigning is known as redirecting I/O.) 5. Visual Prolog's basic file-access predicates are: a. openread (open a file for reading) Chapter 170 Advanced topics

309

b. openwrite (open a file for writing) c. openappend (open a file for appending) d. openmodify (open a file for modification) e. openfile (general purpose file opener) f. filemode (set a file to text mode or binary mode) g. closefile (close a file) h. readdevice (reassign the current_read_device or get its name) i. writedevice (reassign the current_write_device or get its name) 6. To access files you use the FILE domain, which has five predefined alternatives: a. keyboard

(for reading from the keyboard)

b. screen

(for writing to the screen)

c. stdin

(for reading from standard input)

d. stdout

(for writing to standard output)

e. stderr

(for writing to standard error)

7. To work within and manipulate files, you use the following standard predicates: a. filepos (controls the position where reading or writing takes place) b. eof (checks whether the file position during a read operation is at the end of the file) c. flush (forces the contents of the internal buffer to be written to a file) d. existfile (verifies that a file exists) e. searchfile (locates a file among several directories) f. deletefile (deletes a file) g. renamefile (renames a file) h. copyfile (copies a file) i. fileattrib (gets or sets file attributes) j. disk (changes the current disk and directory/subdirectory) 8. To search directories, the following standard predicates are available: a. diropen (opens a directory for searching)

b. dirmatch (finds matching files in a directory) c. dirclose (closes a directory) d. dirfiles (non-deterministically matches files in a directory) 9. To manipulate file names, the following are available: a. filenamepath (joins or splits qualified file names) b. filenameext (joins or splits file names and extensions) 10.The standard predicate readterm allows your program to access facts in a file at run time. readterm can read any object written by write, plus facts describing database predicates.

CHAPTER

231

String-Handling in Visual Prolog 232Visual PrologVisual PrologVisual PrologVisual Prolog provides several standard predicates for powerful and efficient string manipulations. In this chapter, we've divided these into two groups: the family of basic string-handling predicates, and a set of predicates used for converting strings to other types and vice versa. Strings may also be compared to each other, but this is covered in chapter 233.

String Processing A few formalities apply to strings and string processing, in that the backslash acts as an escape character, allowing you to put non-keyboardable characters into strings. Please see the description on page 102.

Basic String-Handling Predicates The predicates described in this section are the backbone of string-handling in Visual Prolog; as such, they serve several purposes: dividing a string into component strings or tokens Chapter 170 Advanced topics

311

building a string from specified strings or tokens verifying that a string is composed of specified strings or tokens returning a string, token, or list of these from a given string verifying or returning the length of a string creating a blank string of specified length verifying that a string is a valid Visual Prolog name formatting a variable number of arguments into a string variable frontchar/3 frontchar operates as if it were defined by the equation String1 = the concatenation of Char and String2

It takes this format: frontchar(String1,Char,String2) /* (i,o,o) (i,i,o) (i,o,i) (i,i,i) (o,i,i) */

frontchar takes three arguments; the first is a string, the second is a char (the first character of the first string), and the third is the rest of the first string. frontchar can be used to split a string up into a series of characters, or to create a string from a series of characters, and to test the characters within a string. If the argument String1 is bound to a zero-length string, the predicate fails. Example In Program 1, frontchar is used to define a predicate that changes a string to a list of characters. Try the goal string_chlist("ABC", Z)

This goal will return Z bound to ['A','B','C']. /* Program ch234e01.pro */ DOMAINS charlist = char* PREDICATES string_chlist(string, charlist)

CLAUSES string_chlist("", []):-!. string_chlist(S, [H|T]):frontchar(S,H,S1), string_chlist(S1,T).

fronttoken/3 fronttoken performs three related functions, depending on the type of flow pattern you use when calling it. fronttoken(String1, Token, Rest) /* (i,o,o) (i,i,o) (i,o,i) (i,i,i) (o,i,i) */

In the (i,o,o) flow variant, fronttoken finds the first token of String1, binds it to Token, and binds the remainder of String1 to Rest. The (i,i,o), (i,o,i), and (i,i,i) flow variants are tests; if the bound arguments are actually bound to the corresponding parts of String1 (the first token, everything after the first token, or both, respectively), fronttoken succeeds; otherwise, it fails. The last flow variant (o,i,i) constructs a string by concatenating Token and Rest, then binds String1 to the result. A sequence of characters is grouped as one token when it constitutes one of the following: a name according to normal Visual Prolog syntax a number (a preceding sign is returned as a separate token) a non-space character fronttoken is perfectly suited for decomposing a string into lexical tokens. Example Program 2 illustrates how you can use fronttoken to divide a sentence into a list of names. If 2 is given the goal: string_namelist("bill fred tom dick harry", X).

X will be bound to: [bill, fred, tom, dick, harry]

Chapter 170 Advanced topics

313

/* Program ch235e02.pro */ DOMAINS namelist = name* name = symbol PREDICATES string_namelist(string, namelist) CLAUSES string_namelist(S,[H|T]):fronttoken(S,H,S1),!, string_namelist(S1,T). string_namelist(_,[]).

frontstr/4 frontstr splits String1 into two parts. It takes this format: frontstr(NumberOfChars, String1, StartStr, EndStr) /* (i,i,o,o) */

StartStr contains the first NumberOfChars characters in String1, and EndStr contains the rest. When frontstr is called, the first two parameters must be bound, and the last two must be free. concat/3 concat states that String3 is the string obtained by concatenating String1 and String2. It takes this format: concat(String1, String2, String3) /* (i,i,o), (i,o,i), (o,i,i), (i,i,i) */

At least two of the parameters must be bound before you invoke concat, which means that concat always gives only one solution (in other words, it's deterministic). For example, the call concat("croco", "dile", In_a_while)

binds In_a_while to crocodile. In the same vein, if See_ya_later is bound, the call concat("alli", "gator", See_ya_later)

succeeds only if See_ya_later is bound to alligator.

str_len/2 str_len can perform three tasks: It either returns or verifies the length of a string, or it returns a string of blank spaces of a given length. It takes this format: str_len(StringArg, Length)

/* (i,o), (i,i), (o,i) */

str_len binds Length to the length of StringArg or tests whether StringArg has the given Length. The Length is an unsigned integer. In the third flow version, str_len returns a string of spaces with a given length; this can be used to allocate buffers, etc. allocating buffers with str_len, but makebinary is preferable especially for binary data. isname/1 isname verifies that its argument is a valid name in accordance with Visual Prolog's syntax; it takes this format: isname(String)

/* (i) */

A name is a letter of the alphabet or an underscore character, followed by any number of letters, digits, and underscore characters. Preceding and succeeding spaces are ignored. format/* format performs the same formatting as writef (see page 277), but format delivers the result in a string variable. format(OutputString,FormatString,Arg1,Arg2,Arg3,...,ArgN) /* (o,i,i,i,...,i) */

subchar/3 subchar returns the character at a given position in a string; it takes the form: subchar(String,Position,Char)

/* (i,i,o) */

The first character has position 1. For example, subchar("ABC",2,Char)

will bind Char to B. If the position specifies a character beyond the end of the string, subchar exits with an error.

Chapter 170 Advanced topics

315

substring/4 substring returns a part of another string; it takes the form: substring(Str_in,Pos,Len,Str_out)

/* (i,i,i,o) */

Str_out will be bound to a copy of the string starting with the Pos’th character, Len characters long, in Str_in. For example substring("GOLORP",2,3,SubStr)]

binds SubStr to OLO. If Pos and Len specify a string partly or wholly outside of Str_in, substring exits with an error. However, it is not an error to ask for 0 bytes at the extreme end of the string: substring("ABC",4,0,SubStr)]

will bind SubStr to an empty string (""), while substring("ABC",4,1,SubStr)/* WRONG */]

is an error. By the way, so is substring("ABC",5,-1,SubStr)/* WRONG */]

searchchar/3 searchchar returns the position of the first occurrence of a specified character in a string; it takes the form: searchchar(String,Char,Position)

/* (i,i,o) */]

For example, searchchar("ABEKAT",'A',Pos)]

will bind Pos to 1. If the character isn't found, searchchar will fail. Note that searchchar is not re-satisfiable (i.e. if there are more occurrences of the specified character in the string, backtracking won't find them), but you can easily make your own: /* Program ch236e03.pro */ PREDICATES nondeterm nd_searchchar(string,char,integer) nondeterm nd_searchchar1(string,char,integer,integer) nondeterm nd_sc(string,char,integer,integer,integer) run

CLAUSES nd_searchchar(Str,Ch,Pos):nd_searchchar1(Str,Ch,Pos,0). nd_searchchar1(Str,Ch,Pos,Old):searchchar(Str,Ch,Pos1), nd_sc(Str,Ch,Pos,Pos1,Old). nd_sc(_,_,Pos,Pos1,Old):- Pos = Pos1+Old. nd_sc(Str,Ch,Pos,Pos1,Old):frontstr(Pos1,Str,_,Rest), Old1 = Old + Pos1, nd_searchchar1(Rest,Ch,Pos,Old1). GOAL nd_searchchar("abbalblablabbala",'a',P), write(P,'\n'), fail.

This implements a non-deterministic predicate (nd_searchchar) which is plugcompatible with searchchar; if you don't mind typing the extra argument (Old) to nd_searchchar1 yourself, you can of course discard a level of calls. searchstring/3 searchstring returns the position of the first occurrence of a string in another string; it takes the form: searchstring(SourceStr,SearchStr,Pos)

/* (i,i,o) */]

For example, searchstring("ABEKAT","BE",Pos)]

will bind Pos to 2. If the search string isn't found in, or is longer than, the source string, searchstring will fail. As with searchchar, searchstring isn't resatisfiable, but you can easily make your own. As a matter of fact, all that's necessary is to take 3 and do a global substitution with 'string' replacing 'char', and change the 'a' in the goal to a suitable search string, e.g. "ab": GOAL nd_searchstring("abbalblablabbala","ab",P), write(P,'\n'), fail.

Chapter 170 Advanced topics

317

Type Conversion In this section, we summarize the standard predicates available for type conversion. The predicates are char_int, str_char, str_int, str_real, upper_lower, and finally term_str which converts between terms of any kind and strings. char_int/2 char_int converts a character into an integer or an integer into a character; it takes this format: char_int(Char, Integer)

/* (i,o), (o,i), (i,i) */

With both its arguments bound, char_int tests that the arguments correspond. With one argument bound and the other free, char_int performs the conversion and binds the output argument to the converted form of the input one. Note: This predicate is really not needed in newer versions of Visual Prolog because there is automatic conversion between characters and integers. We've left char_int in to be compatible with older versions. str_char/2 str_char converts a string containing one and only one character into a character, or converts a single character into a string of one character; it takes this format: str_char(String, Char)

/* (i,o), (o,i), (i,i) */]

In the (i,i) flow variant, str_char succeeds if String is bound to the singlecharacter string equivalent of Char. If the length of the string is not 1, str_char fails. str_int/2 str_int converts a string containing an integer into an integer, or converts an integer into its textual representation; it takes this format: str_int(String, Integer)

/* (i,o), (o,i), (i,i) */]

In the (i,i) flow variant, str_int succeeds if Integer is bound to the integer equivalent of the integer represented by String.

str_real/2 str_real converts a string containing a real number into a real number, or converts a real number into a string; it takes this format: str_real(String, Real)

/* (i,o), (o,i), (i,i) */]

In the (i,i) flow variant, str_real succeeds if Real is bound to the real equivalent of the real number represented by String. upper_lower/2 upper_lower converts an upper-case (or mixed) string or character to all lowercase, or a lower-case (or mixed) string or character to all upper-case; it takes this format: upper_lower(Upper, Lower)

/* (i,o), (o,i), (i,i) */]

With both its arguments bound, upper_lower succeeds if Upper and Lower are bound to strings that--except for the case of the letters--are identical; for instance, the goal: Str1=samPLEstrING, Str2=sAMpleSTRing, upper_lower(Str1, Str2)} succeeds. Otherwise, it fails.

term_str/3 term_str is a general-purpose conversion predicate and will convert between terms of a specified domain and their string representations. It looks like this: term_str(Domain,Term,String)

/* (i,i,o),(i,_,i) */]

where Domain specifies which domain the term belongs to. term_str could replace the various str_* predicates above, for instance, str_real could be implemented as str_real(S,R):- term_str(real,R,S). However, term_str is a somewhat heavier mechanism. The Domain need not be one of the standard domains, it can be any user-defined domain: /* Program ch237e04.pro */ DOMAINS intlist = integer*

Chapter 170 Advanced topics

319

GOAL write("Input list (example [66,73,76,83]): "), readln(L),nl, str_len(L,Len), write("The stringlength of ",L), write(" is ",Len,'\n').

Examples This example defines the predicate scanner, which transforms a string into a list of tokens. Tokens are classified by associating a functor with each token. This example uses the predicates isname, str_int, and str_len to determine the nature of the tokens returned by fronttoken. /* Program ch238e05.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS tok = numb(integer); name(string); char(char) toklist = tok* PREDICATES nondeterm scanner(string, toklist) nondeterm maketok(string, tok) CLAUSES scanner("",[]). scanner(Str,[Tok|Rest]):fronttoken(Str, Sym, Str1), maketok(Sym, Tok), scanner(Str1, Rest). maketok(S,name(S)):-isname(S). maketok(S,numb(N)):-str_int(S,N). maketok(S,char(C)):-str_char(S, C). GOAL write("Enter some text:"),nl, readln(Text),nl, scanner(Text,T_List), write(T_List).

Conversions between the domain types symbol and string, and between char, integer, and real, are handled automatically when using standard predicates and during evaluation of arithmetic expressions. Reals will be rounded during automatic conversions. Visual Prolog performs this automatic conversion as necessary when a predicate is called, as in the following example:

PREDICATES p(integer) CLAUSES p(X):- write("The integer value is ",X,'\n').

With this example, the following goals have the same effect: X=97.234, p(X). X=97, p(X). X='a', p(X).

The following very simple English parser is a practical example of string parsing. This example directly parses strings; if the parser were to be extended, the string should be tokenized using a scanner similar to the one used in Program 4. Whether you're parsing tokens or strings, the algorithm in this program is a good example of how to start. If you are interested in English-language parsing, we recommend that you take a look at the Sentence Analyzer and Geobase programs in the VPI\PROGRAMS subdirectory. /* Program ch239e06.pro */ DOMAINS sentence

= s(noun_phrase,verb_phrase)

noun_phrase = noun(noun) ; noun_phrase(detrm,noun) noun = string verb_phrase = verb(verb) ; verb_phrase(verb,noun_phrase) verb = string detrm

= string

PREDICATES nondeterm s_sentence(string,sentence) nondeterm s_noun_phrase(string,string,noun_phrase) nondeterm s_verb_phrase(string,verb_phrase) d(string) n(string) v(string) CLAUSES s_sentence(Str,s(N_Phrase,V_Phrase)):s_noun_phrase(Str, Rest, N_Phrase), s_verb_phrase(Rest, V_Phrase).

Chapter 170 Advanced topics

321

s_noun_phrase(Str,Rest,noun_phrase(Detr,Noun)):fronttoken(Str,Detr,Rest1), d(Detr), fronttoken(Rest1,Noun,Rest), n(Noun). s_noun_phrase(Str,Rest,noun(Noun)):fronttoken(STR,Noun,Rest), (Noun). s_verb_phrase(Str, verb_phrase(Verb,N_Phrase)):fronttoken(Str,Verb,Rest1), v(Verb), s_noun_phrase(Rest1,"",N_Phrase). s_verb_phrase(Str,verb(Verb)):fronttoken(STR,Verb,""), v(Verb). /* determiner */ d("the"). d("a"). /* nouns */ n("bill"). n("dog"). n("cat"). /* verbs */ v("is").

Load and run this program, and enter the following goal: Goal s_sentence("bill is a cat", Result).

The program will return: Result = s(noun("bill"),verb_phrase("is", noun_phrase("a","cat"))) 1 Solution

Summary These are the important points covered in this chapter:

1. Visual Prolog's string-handling predicates are divided into two groups: basic string manipulation and string type conversions. 2. The predicates for basic string manipulation are summarized here: a. frontchar, fronttoken, and concat for dividing a string into components, building a string from specified components, and testing if a string is composed of specified components; these components can be characters, tokens, or strings b. subchar and substring for returning a single character from, or a part of, another string c. searchchar and searchstring for finding the first occurrence of a character, or a string, in a string d. str_len for verifying or returning the length of a string, or creating a blank string of specified length e. frontstr for splitting a string into two separate strings f. isname for verifying that a string is a valid Visual Prolog name g. format for formatting a variable number of arguments into a string variable Several of the basic string manipulation predicates have different flow variants. The variants with only input parameters perform tests that succeed when the string in question is made up of the specified components (or is of the specified length). 3. The predicates for type conversion are listed here: a. char_int for converting from a character to an integer, or vice versa b. str_char for converting a single character into a string of one character, or vice versa c. str_int for converting from an integer to its textual representation, or vice versa d. str_real for converting from a real number to a string, or vice versa e. upper_lower for converting a string to all upper-case or all lower-case characters, or testing case-insensitive string equality f. term_str for conversion between arbitrary domains and strings The type conversion predicates each have three flow variants; the (i,o) and variants perform the appropriate conversions, and the (i,i) variants are tests that succeed only if the two arguments are bound to the converted representations of one another (o,i)

Chapter 170 Advanced topics

323

CHAPTER

240

The External Database System 241Visual PrologIn this chapter, we cover Visual Prolog's external database system. An external database is composed of an external collection of chained terms; these chains give you direct access to data that is not a part of your Prolog program. The external database can be stored in any one of three locations: in a file, in memory, or in EMS-type expanded memory under DOS. The external database supports B+ trees, which provide fast data retrieval and the ability to sort quickly, and it supports multi-user access by a mechanism for serializing the file accesses inside transactions.

External Databases in Visual Prolog Visual Prolog's internal fact’s database, which uses asserta, assertz, retract, and retractall, is very simple to use and suitable for many applications. However, the RAM requirements of a database can easily exceed the capacity of your computer; the external database system has been designed partly with this problem in mind. For example, you might want to implement one or more of the following: a stock control system with an large number of records an expert system with many relations but only a few records with complicated structures a filing system in which you store large text files in the database your own database product--which maybe has nothing to do with a relational database system--in which data is linked together in other, nonrelational ways a system including several of these possibilities Visual Prolog's external database system supports these different types of applications, while meeting the requirement that some database systems must not lose data during update operations--even in the event of power failure. Visual Prolog's external database predicates provide the following facilities:

efficient handling of very large amounts of data on disk the ability to place the database in a file, in memory, or in EMS-type expanded memory cards under DOS multi-user access greater data-handling flexibility than provided by the sequential nature of Visual Prolog's automatic backtracking mechanism the ability to save and load external databases in binary form

An Overview: What's in an External Database? A Visual Prolog external database consists of two components: the data items-actually Prolog terms--stored in chains, and corresponding B+ trees, which you can use to access the data items very quickly. The external database stores data items in chains (rather than individually) so that related items stay together. For example, one chain might contain part numbers to a stock list, while another might contain customer names. Simple database operations, such as adding new items or replacing and deleting old items, do not require B+ trees. These come into play when you want to sort data items or search the database for a given item; they are covered in detail later in this chapter. Naming Convention The names of all the standard predicates concerned with database management follow a certain convention. The first part of the name (db_, chain_, term_, and so on) is a reminder of what you must specify as input. The second part of the name (flush, btrees, delete, and so on) is a reminder of what action occurs or what is returned or affected. For example, db_delete deletes a whole database, chain_delete deletes a whole chain, and term_delete deletes a single term.

Chapter 170 Advanced topics

325

Chain1:

term

Chain2 :

term

ChainN :

term

B+Tree1

B+Tree 2

. . . B+TreeN

Figure 242.1: Structure of a Visual Prolog External Database243Visual Prolog

External Database Selectors It is possible to have several external databases simultaneously in memory, on disk, and in an EMS-type memory expansion card under DOS. With this flexibility, you can place external databases where they give the best speed and space compromise.

In order to distinguish between several open databases, you use a selector in every call to an external database standard predicate. You must declare these selectors in a domain called db_selector. This works like the file domain in the file system. For example, the following domains, declarations, external databases domain declaration declares customers and parts to be external database selectors: DOMAINS db_selector = customers; parts

Chains An external database is a collection of Prolog terms. Some examples of terms are integers, reals, strings, symbol values, and compound objects; for instance, 32, -194, 3.1417, "Wally", wages, and book(dickens, "Wally goes to the zoo"). Inside an external database, the terms are stored in chains. A chain can contain any number of terms, and an external database can contain any number of chains. Each chain is selected by a name, which is simply a string. The following figure illustrates the structure of a chain called MY_CHAIN.

term

Figure 244.2: Structure of a Chain245 Database relations and database tables are modeled by chains of terms. For example, suppose you have a customer, supplier, and parts database, and you want to put all the data into a single database with three relations: one for customers, one for suppliers, and one for parts. You do this by putting the customers in one chain called customers, the suppliers in another chain called suppliers, and the parts in a chain called parts. To insert a term in an external database, you must insert the term into a named chain. On the other hand, you can retrieve terms without explicitly naming the containing chain. In both cases, you must specify the domain to which the term belongs. In practice, it is best if all terms in the chain belong to the same domain, but there is actually no restriction on how terms are mixed in a database. It's up to you to ensure that a term you retrieve belongs to the same domain as it did when you inserted it. Chapter 170 Advanced topics

327

The following is a simple example of setting up two chained databases, dba1 and dba2. In this example, all the customer data is in dba1 and all the parts data in dba2. For now, just look over this example; we need to introduce a lot more information before we can explain what's happening here. /* Program ch246e01.pro */ DOMAINS db_selector = dba1 ; dba2 customers = customer(customer_name, address) parts = part(part_name, ID, customer_name) customer_name, part_name = symbol ID = integer address = string PREDICATES access CLAUSES access:chain_terms(dba1,chain1,customers,customer(Name, ADDR),_), chain_terms(dba2,chain2,parts,part(Part, Id, Name),_), write("send ",Part," part num ",Id," to ",Addr), nl, fail. access. GOAL % create the databases dba1 and dba2 db_create(dba1, "dd1", in_memory), db_create(dba2, "dd1.bin", in_file), % insert customer facts into chain1 in dba1 chain_insertz(dba1, chain1, customers, customer("Joe Fraser","123 West Side"), _), chain_insertz(dba1, chain1, customers, customer("John Smith","31 East Side"), _), chain_insertz(dba1, chain1, customers, customer("Diver Dan","1 Water Way"), _), chain_insertz(dba1, chain1, customers, customer("Dave Devine","123 Heaven Street"), _),

% insert parts facts into chain2 in dba2 chain_insertz(dba2, chain2, parts, part("wrench", 231, "John Smith"), _), chain_insertz(dba2, chain2, parts, part("knife", 331, "Diver Dan"), _), access, db_close(dba1), db_close(dba2), db_delete("dd1", in_memory), db_delete("dd1.bin", in_file).

This program first creates the databases dba1 (in memory) and dba2 (in a disk file). It then inserts facts into two chains: chain1 and chain2. After inserting the facts, it looks in these chains for a customer and the part ordered by that customer; finding these, it returns the address to which the shipper should ship the part. Finally, it closes and deletes the two databases.

External Database Domains The external database uses six standard domains, summarized here: Domain

What It's Used For

db_selector

Domain for declaring database selectors

bt_selector

Domain for declaring B+ tree selectors

place

Location of the database: in RAM, in a file, or in an extended memory system (EMS card under DOS)

accessmode

Decides how the file will be used.

denymode

Determines how other users can open the file.

ref

Reference to the location of a term in a chain

Database Reference Numbers Every time you insert a new term into an external database, Visual Prolog assigns it a database reference number. You can use the term's database reference number to retrieve, remove, or replace that term, or to get the next or previous term in the chain. You can also insert a database reference number in a B+ tree (as described later in this chapter), and then use the B+ tree to sort some terms or to carry out a fast search for a term. Database reference numbers are independent of the database location and any possible packing operations. Once a reference number has been associated with a Chapter 170 Advanced topics

329

term, you can use that number to access that term--no matter which database management operations are subsequently carried out--until the term is deleted. The ref Domain Database reference numbers are special because you can insert them in facts sections and write them out with write or writef, but you can't type them in from the keyboard. You must declare the arguments to predicates handling database reference numbers as belonging to the standard domain ref. When you delete a term with term_delete, the system will reuse that term's reference number when it inserts the next term into the external database. This happens automatically; however, if reference numbers have been stored in the facts section or in a B+ tree for some reason, it is your responsibility to ensure that a given reference is associated with the correct term. To assist you in this, there is an error-checking option, enabled with the db_reuserefs standard predicate: db_reuserefs/2 db_reuserefs has the following form: db_reuserefs(DBase,ReUse)

/* (i,i)*/

where DBase is a db_selector and ReUse is an unsigned integer. This should be set to 0 to enable checking for use of released terms, or 1 do disable this. The overhead of having the check enabled is very small (4 bytes per term, virtually no CPU overhead), but those 4 bytes will never be released. If you constantly create and release terms, your database will therefore grow at a steady rate. db_reuserefs's primary purpose is to assist you in tracking down bugs during development of programs.

Manipulating Whole External Databases When you create a new external database, or open an existing one, you can place it in a file, in memory, or in EMS-type expanded memory under DOS, depending on the value of the Place argument in your call to db_create or db_open. After you've finished working with the external database, you close it with a call to db_close. When you place an external database in main or expanded memory, closing the database with db_close does not delete the database from memory. You must do this explicitly with a call to db_delete, to free the memory the database occupies.

If you close such an external database but don't delete it, you can later reopen it with the db_open predicate. Since the external database system relies on the DOS buffer system, it will be very slow if no buffers have been allocated. To allocate 40 buffers (which isn't an excessive number), include the following line in your CONFIG.SYS file (a part of the DOS environment): buffers = 40

In this section, we discuss the predicates db_create, db_open, db_copy, db_loadems, db_saveems, db_close, db_delete, db_openinvalid, db_flush, db_garbagecollect, db_btrees, db_chains, and db_statistics. db_create/3 db_create creates a new database. db_create(Dbase, Name, Place)

/* (i,i,i) */

If the database is placed in a disk file, the name of the file will be Name; if it's placed in memory or EMS under DOS, you'll need Name if you close the database and want to open it later. Dbase and Name correspond to the internal and external names for files. Where you place an external database is determined by the Place argument. Place can take one of the following values: in_file

The external database is placed in a disk file, and there will be only a minimum of main memory overhead.

in_memory

The external database is placed in the main memory-usually this will be done to achieve maximum performance.

in_ems

The database is placed in EMS-type expanded memory, if a suitable card is installed in the computer. in_ems is only relevant for DOS. On other platforms it has the same effect as in_memory

These values, in_file, in_memory, and in_ems, are elements of the pre-declared domain place, which corresponds to the following declaration: DOMAINS place = in_file; in_memory; in_ems

For example, here are two different calls to db_create: Chapter 170 Advanced topics

331

db_create(db_sel1,"MYFILE.DBA",in_file) /* Creates disk file MYFILE.DBA */ db_create(db_sel2,"SymName2",in_memory) /* Creates memory database SymName2 */

db_open/3 db_open opens a previously created database, identified by Name and Place. db_open(Dbase, Name, Place)

/* (i,i,i) */

If Place is in_memory or in_ems, Name is the database's symbolic file name; if Place is in_file, Name is the actual DOS-style file name. db_copy/3 Irrespective of where you initially place an external database, you can later move it to another location with the db_copy predicate. db_copy(Dbase, Name, Place)

/* (i,i,i) */

For example, in this call to db_copy db_copy(my_base, "new_EMSbase", in_ems)

Visual Prolog copies the database identified by the database selector my_base into the new database file new_EMSbase, which is placed in EMS under DOS. When you copy a database, the original still exists; you will have two copies until you explicitly delete the original. Once you've moved a database, all processing can continue as if nothing happened, since all reference numbers to the external database terms will still be valid. In this way, if you're maintaining an external database in main memory, and free storage is running short, you can copy the database to a file and continue execution with the database in the file. An index set up to the external database in internal memory is still valid, even after you've copied the database to a file. db_copy has several uses; you can use it to do the following: Load a database from disk to memory and later save it again in binary form, instead of using save and consult with text files. Copy a medium-sized database from disk to memory for faster access. Pack a database containing too much free space; when the database is copied to another file, all free space will be eliminated.

db_loadems/2 and db_saveems/2 db_copy performs a full-scale record-by-record copy of the database in question. This has the advantage that the resulting database will be compacted and without unused space, but for large databases the process can be time consuming. For DOS only, db_loadems and db_saveems will transfer complete images of databases between disk and EMS: db_loadems(FileName,EmsName db_saveems(EmsName,FileName)

/* (i,i) */ /* (i,i) */

The only restriction on their use is that there can be no more than one database in EMS. db_openinvalid/3 db_openinvalid allows you to open a database that's been flagged as invalid. db_openinvalid(Dbase, Name, Place)

/* (i,i,i) */

If the power to the computer fails while a database is being updated, all the data in the database may be lost because part of some buffer has not been written to disk. A flag in the database indicates if it's in an invalid state after an update. A database is recorded as being invalid after a call to any of the predicates that change the content in the database. These include chain_inserta, chain_insertz, chain_insertafter, term_replace, term_delete, chain_delete, bt_create, key_insert, and key_delete. The database is recorded as being valid once again when it is closed with db_close, or when db_flush is called to flush out the buffers. By using db_openinvalid, it is sometimes possible to continue execution when a database is marked as invalid. This might make it possible to recover some data if all your backups have disappeared. However, all attempts to use an invalid database after opening it with db_openinvalid might yield unexpected results. db_flush/1 db_flush flushes the buffers and writes their contents to the appropriate destination in your database. db_flush(Dbase)

/* (i) */

When a database is updated it will be marked as invalid, and it remains flagged as invalid until it is either flushed with db_flush, or closed. Chapter 170 Advanced topics

333

The level of security you employ for a given database will, of course, depend on how important its data is. The most basic level of data security is to keep backups on disk. At the intermediate level, you could call db_flush after each important database update. However, flushing the buffers is a relatively slow operation; if it's done too often, your database system will grind to a halt. Finally, if the contents of an external database are especially valuable, you could record all changes in a special log file or maintain two identical databases--perhaps on different disks. db_close/1 A call to db_close closes an open database. db_close(Dbase)

/* (i) */

If the database Dbase is placed in a disk file, the file will be closed. The database won't be deleted, even if it is placed in memory or in an EMS-type memory expansion card, and you can reopen it later through a call to db_open. You can use db_delete to remove a closed database from memory. db_delete/1 When the database is situated in memory or in an EMS-type memory expansion card, db_delete releases all the occupied space. db_delete(Name, Place)

/* (i,i) */

When the database is situated in a file, db_delete erases the file. db_delete gives an error if the database Name does not exist in the given Place. db_garbagecollect/1 db_garbagecollect scans through the free lists in the database garbage collect and tries to merge some of the free space together into larger pieces. db_garbagecollect(Dbase)

/* (i) */

This scanning and merging is done automatically when the database is placed in memory or in an EMS card. Under normal circumstances, there should be no need to call this predicate. However, if there seems to be too much free space in the database that is not being reused when new terms are inserted, db_garbagecollect can regain some extra space.

db_btrees/2 During backtracking, db_btrees successively binds BtreeName to the name of each B+ tree in the Dbase database. nondeterm db_btrees(Dbase, BtreeName)

/* (i,o) */

The names are returned in sorted order. B+ trees are described later in this chapter. db_chains/2 During backtracking, db_chains successively binds ChainName to the name of each chain in the Dbase database. nondeterm db_chains(Dbase, ChainName)

/* (i,o) */

The names are returned in sorted order. db_statistics/5 db_statistics returns statistical information for the database Dbase. db_statistics(Dbase, NoOfTerms, MemSize, DbaSize, FreeSize) /* (i,o,o,o,o) */

The arguments to db_statistics represent the following: NoOfTerms is bound to the total number of terms in the database. MemSize

is bound to the size--in bytes--of the internal tables stored in memory for the database.

DbaSize

is bound to the total number of bytes that the terms and descriptors in the Dbase database occupy. If Dbase is stored in a disk file, and DbaSize gets a value much smaller than the size of that file, the file can be compressed by using db_copy.

FreeSize

becomes bound to a value representing the free memory space; the value depends on where the database Dbase is currently placed. When Dbase is placed in memory, FreeSize is bound to the number of unused bytes between the top of the global stack and the top of the heap. (Note: There might be some additional free bytes that are not included in this count.)

Chapter 170 Advanced topics

335

If Dbase is placed in EMS-type expanded memory, FreeSize is bound to the number of unoccupied bytes in that expansion memory. When Dbase is placed in a file, FreeSize is bound to the number of unused bytes on the disk containing the file.

Manipulating Chains To insert terms into an external database chain, you use the predicates chain_inserta, chain_insertz, or chain_insertafter. You can successively bind the terms in a chain, and their reference numbers, to the arguments of chain_terms, while chain_delete allows you to delete a whole chain of terms from the external database. Four standard predicates return database reference numbers. These are chain_first, chain_last, chain_next, and chain_prev. chain_inserta/5 and chain_insertz/5 The predicates chain_inserta and chain_insertz correspond to asserta and assertz, respectively. These take the following form: chain_inserta(Dbase, Chain, Domain, Term, Ref) chain_insertz(Dbase, Chain, Domain, Term, Ref)

/* (i,i,i,i,o) */ /* (i,i,i,i,o) */

chain_inserta inserts the term Term at the beginning of the chain Chain, while chain_insertz inserts Term at the chain's end. Dbase is the db_selector of the database, Domain is the domain of Term, and Ref is the database reference number corresponding to Term. For example, if my_dba is declared to be in the domain db_selector, like this: DOMAINS db_selector = my_dba; ....

then in this call to chain_inserta chain_inserta(my_dba, customer, person, p(john, "1 The Avenue", 32), NewRef)

customer is the name of the chain, and all customers are stored in one chain. It would be perfectly all right to store the suppliers as terms from the domain person but in a different chain, perhaps called supplier. person is the name of the

domain to which declaration:

p(john, "1 The Avenue", 32)

belongs, as shown in this domain

DOMAINS person = p(name, address, age)

If Chain doesn't already exist, these predicates will automatically create it. chain_insertafter/5 chain_insertafter inserts a term after a specified term, returning the inserted term's new reference number. It takes this format: chain_insertafter(Dbase, Domain, Ref, Term, NewRef) /* (i,i,i,i,o) */

chain_insertafter inserts the term Term after the chain element specified by Ref, while NewRef is bound to the database reference number corresponding to Term after it's been inserted. chain_terms/5 During backtracking, chain_terms successively binds Term and Ref to each term and its associated database reference number in the specified Chain. chain_terms takes the form: chain_terms(Dbase, Chain, Domain, Term, Ref)

/* (i,i,i,o,o) */

chain_delete/2 chain_delete deletes a specified chain from a given external database; this predicate takes the form: chain_delete(Dbase, Chain)

/* (i,i) */

chain_first/3 and chain_last/3 chain_first and chain_last return the database reference number for the first and last terms in a given chain, respectively. chain_first(Dbase, Chain, FirstRef) chain_last(Dbase, Chain, LastRef)

Chapter 170 Advanced topics

/* (i,i,o) */ /* (i,i,o) */

337

chain_next/3 and chain_prev/3 chain_next returns the reference number of the term following the given one, while chain_prev returns the reference number of the term preceding the given one. chain_next(Dbase, Ref, NextRef) chain_prev(Dbase, Ref, PrevRef)

/* (i,i,o) */ /* (i,i,o) */

Manipulating Terms Three standard predicates for external database management are all concerned with terms; these are term_replace, term_delete, and ref_term. Whenever you call any of the term-handling external database standard predicates, you must give the domain of the term as one of the arguments. Because of this, it's usually a good idea to declare all terms in a given database as alternatives in one domain, as in this declaration: DOMAINS terms_for_my_stock_control_database = customer(Customer, Name, ZipCode, Address); supplier(SupplierNo, Name, Address); parts(PartNo, Description, Price, SupplierNo)

Note that there are no restrictions on mixing types (domains) in an external database. One chain can contain text strings, another integers, a third some kind of compound structures, and so on. However, external database data items are not stored with type descriptors; for example, integers don't necessarily occupy just two bytes. It's your responsibility to retrieve a term into the same domain as that from which it was inserted. A run-time error will usually result if you attempt to mix domains. term_replace/4 term_replace replaces an old term (referenced by Ref, a database reference number) with a new term, Term. term_replace(Dbase, Domain, Ref, Term)

/* (i,i,i,i) */

term_delete/3 term_delete erases the term stored under Ref, a given database reference number. term_delete(Dbase, Chain, Ref)

/* (i,i,i) */

The storage occupied by the term will be released, and there must be no further references to Ref. ref_term/4 ref_term binds Term to the term stored under a given reference number, Ref. ref_term(Dbase, Domain, Ref, Term)

/* (i,i,i,o) */

A Complete Program Example The following example program 2, uses nearly all the external database predicates introduced so far. Working first in memory, this program goes through the following sequence of operations: 1. Writes 100 terms in a database. 2. Reads them back. 3. Replaces every second term. 4. Doubles the number of terms. 5. Erases every second term. 6. Examines every term with ref_term. 7. Calculates the size of the database. This program then copies the database to a disk file and carries out the same sequence of activities twice with the database held on disk. Finally, it calculates-in hundredths of a second--the total time taken to carry out these activities. Note, however, that for illustration the program generates large amounts of output, which slows it down considerably. The true speed is only revealed if you remove the output statements. The program 3 is for UNIX, as time-calculation is done differently in UNIX, and terminal output is significantly slower in UNIX than in DOS. Run the program to see what happens, and then try to alter the number of terms and study your system's performance. The DOS program appears below. /* Program ch247e02.pro */ /* Copyright (c) 1986, '95 by Prolog Development Center */ DOMAINS my_dom = f(string) db_selector = my_dba

Chapter 170 Advanced topics

339

PREDICATES write_dba(integer) read_dba rd(Ref) count_dba(integer) count(Ref, integer, integer) replace_dba replace(Ref) double_dba double(Ref) half_dba half(Ref) mixture CLAUSES write_dba(0):-!. write_dba(N):chain_inserta(my_dba,my_chain,my_dom,f("Prolog system"),_), chain_insertz(my_dba, my_chain, my_dom, f("Prolog Compiler"), _), N1=N-1, write_dba(N1). read_dba:db_chains(my_dba, Chain), chain_terms(my_dba, Chain, my_dom, Term, Ref),nl, write("Ref=", Ref), write(", Term=", Term), fail. read_dba:db_chains(my_dba, Chain), chain_first(my_dba, Chain, Ref), rd(Ref), fail. read_dba. rd(Ref):ref_term(my_dba, my_dom, Ref, Term), nl, write(Term), fail. rd(Ref):chain_next(my_dba,Ref,Next),!,rd(Next). rd(_). replace_dba:chain_first(my_dba, my_chain, Ref), replace(Ref).

replace(Ref):term_replace(my_dba, my_dom, Ref, f("Prolog Toolbox")), chain_next(my_dba, Ref, NN), chain_next(my_dba, NN, Next),!, replace(Next). replace(_). half_dba:chain_last(my_dba, my_chain, Ref), half(Ref). half(Ref):chain_prev(my_dba, Ref, PP), chain_prev(my_dba, PP, Prev), !, term_delete(my_dba, my_chain, Ref), half(Prev). half(_). double_dba:chain_first(my_dba, my_chain, Ref), double(Ref). double(Ref):chain_next(my_dba, Ref, Next),!, chain_insertafter(my_dba, my_chain, my_dom, Ref,f("Programmers Guide"), _), double(Next). double(_). count_dba(N):chain_first(my_dba, my_chain, Ref), count(Ref, 1, N). count(Ref, N, N2):chain_next(my_dba, Ref, Next),!, N1=N+1, count(Next, N1, N2). count(_, N, N).

Chapter 170 Advanced topics

341

mixture :-nl, write("Replace every second term:"), replace_dba,nl, write("Double the number of terms:"), double_dba,nl, write("Erase every second term:"), half_dba,nl, write("Use ref_term for all terms:"), read_dba, count_dba(N),nl, write("There are now ", N, " terms in the database"), db_statistics(my_dba, NoOfTerms, MemSize, DbaSize, FreSize),nl, writef("NoOfTerms=%, MemSize=%, DbaSize=%, FreeSize=%", NoOfTerms, MemSize,DbaSize,FreSize). GOAL nl,nl,nl, write("\tTEST OF DATABASE SYSTEM\n\t***********************\n\n"), time(H1, M1, S1, D1), db_create(my_dba, "dd.dat", in_memory),nl,nl, write("Write some terms in the database:"), write_dba(50), read_dba, mixture,nl,nl, write("Copy to file"), db_copy(my_dba, "dd.dat", in_file), db_close(my_dba), db_delete("dd.dat", in_memory), db_open(my_dba, "dd.dat", in_file), mixture, db_close(my_dba),nl,nl,nl, write("Open the database on file"), db_open(my_dba, "dd.dat", in_file), mixture, db_close(my_dba), time(H2, M2, S2, D2), Time = (D2-D1)+100.0*((S2-S1)+60.0*((M2-M1)+ 60.0*(H2-H1))),nl,nl, write("Time = ", Time, "/100 Sec" ), nl.

B+ Trees A B+ tree is a data structure you can use to implement a very sorting, large amounts of data efficient method for sorting large amounts of data; B+ trees enable a correspondingly efficient searching algorithm. You can think of a B+ tree as providing an index to a database, which is why B+ trees are sometimes referred to as indices. In Visual Prolog, a B+ tree resides in an external database. Each entry in a B+ tree is a pair of values: a key string key string and an associated database reference number. When building your database, you first insert a record in the database and establish a key for that record. The Visual Prolog btree predicates may then be used to insert this key and the database reference number corresponding to this record into a B+ tree. When searching a database for a record, all you have to do is to obtain a key for that record, and the B+ tree will give you the corresponding reference number. Using this reference number, you can retrieve the record from the database. As a B+ tree evolves, its entries are kept in key order. This means that you can easily obtain a sorted listing of the records. A B+ tree is analogous to a binary tree, with the exception that in a B+ tree, more than one key string is stored at each node. B+ trees are also balanced; this means that the search paths to each key in the leaves of the tree have the same length. Because of this feature, a search for a given key among more than a million keys can be guaranteed, even in the worst case, to require accessing the disk only a few times--depending on how many keys are stored at each node. Although B+ trees are placed in an external database, they don't need to point to terms in the same database. It is possible to have a database containing a number of chains, and another database with a B+ tree pointing to terms in those chains.

Pages, Order, and Keylength In a B+ tree, keys are grouped together in pages; each page has the same size, and all pages can contain the same number of keys, which means that all the stored keys for that B+ tree must be the same size. The size of the keys is determined by the KeyLen argument, which you must supply when creating a B+ tree. If you attempt to insert strings longer than KeyLen into a B+ tree, Visual Prolog will truncate them. In general, you should choose the smallest possible value for KeyLen in order to save space and maximize speed. When you create a B+ tree, you must also give an argument called its Order. This argument determines how many keys should be stored in each tree node; usually, Chapter 170 Advanced topics 343

you must determine the best choice by trial and error. A good first try for Order is 4, which stores between 4 and 8 keys at each node. You must choose the value of Order by experimentation because the B+ tree's search speed depends on the values KeyLen and Order, the number of keys in the B+ tree, and your computer's hardware configuration.

Duplicate Keys When setting up a B+ tree, you must allow for all repeat occurrences of your key. For example, if you're setting up a B+ tree for a database of customers in which the key is the customer's last name, you need to allow for all those customers called Smith. For this reason, it is possible to have duplicate keys in a B+ tree. When you delete a term in the database, you must delete the corresponding entry in a B+ tree with duplicate keys by giving both the key and the database reference number.

Multiple Scans In order multiple, scans of B+ trees to have more than one internal pointer to the same B+ tree, you can open the tree more than once. Note, however, that if you update one copy of a B+ tree, for which you have other copies currently open, the pointers for the other copies will be repositioned to the top of the tree.

The B+ Tree Standard Predicates Visual Prolog provides several predicates for handling B+ trees; these predicates work in a manner that parallels the corresponding db_... predicates. bt_create/5 and bt_create/6 You create new B+ trees by calling the bt_create predicate. bt_create(Dbase, BtreeName, Btree_Sel, KeyLen, Order) /* (i,i,o,i,i) */ bt_create(Dbase, BtreeName, Btree_Sel, KeyLen, Order, Duplicates) /* (i,i,o,i,i,i) */

The BtreeName argument specifies the name for the new tree. You later use this name as an argument for bt_open. The arguments KeyLen and Order for the B+ Tree are given when the tree is created and can't be changed afterwards. If you are calling bt_create/5 or bt_create/6 with the Duplicates argument set to 1, duplicates will be allowed in the B+Tree. If you call bt_create/6 with the

Duplicates argument set to 0 you will not be allowed to insert duplicates in the B+Tree. bt_open/3 bt_open opens an already created B+ tree in a database, which is identified by the name given in bt_create. bt_open(Dbase, BtreeName, Btree_Sel)

/* (i,i,o) */

When you open or create a B+ tree, the call returns a selector (Btree_Sel) for that B+ tree. A B+ tree selector belongs to the predefined domain bt_selector and refers to the B+ tree whenever the system carries out search or positioning operations. The relationship between a B+ tree's name and its selector is exactly the same as the relationship between an actual file name and the corresponding symbolic file name. You can open a given B+ tree more than once in order to handle several simultaneous scans. Each time a B+ tree is opened, a descriptor is allocated, and each descriptor maintains its own internal B+ tree pointer. bt_close/2 and bt_delete/2 You can close an open B+ tree with a call to bt_close or delete an entire B+ tree with bt_delete. bt_close(Dbase, Btree_Sel) bt_delete(Dbase, BtreeName)

/* (i,i) */ /* (i,i) */

Calling bt_close releases the internal buffers allocated for the open B+ tree with BtreeName. bt_copyselector bt_copyselector gives you a new pointer for an already open B+ tree selector (a new scan). bt_copyselector(Dbase,OldBtree_sel,NewBtree_sel)

/* (i,i,o) */

The new selector will point to the same place in the B+ tree as the old selector. After the creation the two B+ tree selectors can freely be repositioned without affecting each other. bt_statistics/8 bt_statistics returns statistical information for the B+ tree identified by Btree_Sel. Chapter 170 Advanced topics

345

bt_statistics(Dbase,Btree_Sel,NumKeys,NumPages, Depth,KeyLen,Order,PgSize)

/* (i,i,o,o, */ /* o,o,o,o) */

The arguments to bt_statistics represent the following: Dbase

is the db_selector identifying the database.

Btree_Sel

is the bt_selector identifying the B+ tree.

NumKeys

is bound to the total number of keys in the B+ tree Btree_Sel.

NumPages

is bound to the total number of pages in the B+ tree.

Depth

is bound to the depth of the B+ tree.

KeyLen

is bound to the key length.

Order

is bound to the order of the B+ tree.

PgSize

is bound to the page size (in bytes).

key_insert/4 and key_delete/4 You use the standard predicates key_insert and key_delete to update the B+ tree. key_insert(Dbase, Btree_Sel, Key, Ref key_delete(Dbase, Btree_Sel, Key, Ref)

/* (i,i,i,i) */ /* (i,i,i,i) */

By giving both Key and Ref to key_delete, you can delete a specific entry in a B+ tree with duplicate keys. key_first/3, key_last/3, and key_search/4 Each B+ tree maintains an internal pointer to its nodes. key_first and key_last allow you to position the pointer at the first or last key in a B+ tree, respectively. key_search positions the pointer on a given key. key_first(Dbase, Btree_Sel, Ref) key_last(Dbase, Btree_Sel, Ref) key_search(Dbase, Btree_Sel, Key, Ref)

/* (i,i,o) */ /* (i,i,o) */ /* (i,i,i,o)(i,i,i,i) */

If the key is found, key_search will succeed; if it's not found, key_search will fail, but the internal B+ tree pointer will be positioned at the key immediately after where Key would have been located. You can then use key_current to return the key and database reference number for this key. If you want to position on an exact position in a B+ tree with duplicates you can also provide the Ref as an input argument.

key_next/3 and key_prev/3 You can use the predicates key_next and key_prev to move the B+ tree's pointer forward or backward in the sorted tree. key_next(Dbase, Btree_Sel, NextRef) key_prev(Dbase, Btree_Sel, PrevRef)

/* (i,i,o) */ /* (i,i,o) */

If the B+ tree is at one of the ends, trying to move the pointer further will cause a fail, but the B+ tree pointer will act as if it were placed one position outside the tree. key_current/4 key_current returns the key and database reference number for the current pointer in the B+ tree. key_current(Dbase, Btree_Sel, Key, Ref)

/* (i,i,o,o) */

key_current fails after a call to the predicates bt_open, bt_create, key_insert, or key_delete, or when the pointer is positioned before the first key (using key_prev) or after the last (with key_next).

Example: Accessing a Database via B+ Trees The following example program handles several text files in a single database file at once. You can select and edit the texts as though they were in different files. A corresponding B+ tree is set up for fast access to the texts and to produce a sorted list of the file names. /* Program ch248e04.pro */ DOMAINS db_selector = dba PREDICATES % List all keys in an index list_keys(db_selector,bt_selector)

Chapter 170 Advanced topics

347

CLAUSES list_keys(dba,Bt_selector):key_current(dba,Bt_selector,Key,_), write(Key,' '), fail. list_keys(dba,Bt_selector):key_next(dba,Bt_selector,_),!, list_keys(dba,Bt_selector). ist_keys(_,_). PREDICATES open_dbase(bt_selector) main(db_selector,bt_selector) ed(db_selector,bt_selector,string) ed1(db_selector,bt_selector,string) CLAUSES % Loop until escape is pressed main(dba,Bt_select):write("File Name: "), readln(Name), ed(dba,Bt_select,Name),!, main(dba,Bt_select). main(_,_). % The ed predicates ensure that the edition will never fail. ed(dba,Bt_select,Name):ed1(dba,Bt_select,Name),!. ed(_,_,_). %* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * % There are three choices: %% a) The name is an empty string - list all the names % b) The name already exists - modify the contents of the file % c) The name is a new name - create a new file %* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */

ed1(dba,Bt_select,""):-!, key_first(dba,Bt_select,_), list_keys(dba,Bt_select), nl. ed1(dba,Bt_select,Name):key_search(dba,Bt_select,Name,Ref),!, ref_term(dba,string,Ref,Str), edit(Str,Str1,"Edit old",NAME,"",0,"PROLOG.HLP",RET), clearwindow, Str><Str1, RET=0, term_replace(dba, string, Ref, Str1). ed1(dba,Bt_select,Name):edit("",STR1,"Create New",NAME,"",0,"PROLOG.HLP",RET), clearwindow, ""><Str1, RET=0, chain_insertz(dba,file_chain,string,Str1,Ref), key_insert(dba,Bt_select,Name,Ref). open_dbase(INDEX):existfile("dd1.dat"),!, db_open(dba,"dd1.dat",in_file), bt_open(dba,"ndx",INDEX). open_dbase(INDEX):db_create(dba,"dd1.dat",in_file), bt_create(dba,"ndx",INDEX,20,4). GOAL open_dbase(INDEX), main(dba,INDEX), bt_close(dba,INDEX), db_close(dba).

External Database Programming In this section, we provide seven examples that illustrate some general principles and methods for working with Visual Prolog's external database system. This is a summary of what the following sections cover: "Scanning through a Database" shows you the way to perform a sequential scan through a chain or a B+ tree in an external database. "Displaying the Contents of a Database" defines a predicate you can use to display the current state of an external database. Chapter 170 Advanced topics

349

"Making a Database That Won't Break Down" illustrates how to protect your database from unexpected system power failure and other potential catastrophes. "Updating the Database" provides an example that makes it easy to change, add to, and protect your database. "Using Internal B+ Tree Pointers" supplies you with some predicates for positioning a pointer within an open B+ tree. "Changing the Structure of a Database" offers an alternative to the old copywhile-changing method of changing the structure of a database.

Scanning through a Database When you are using the database system, it is important to keep Visual Prolog's storage mechanisms storage mechanisms in mind. Every time Visual Prolog retrieves a term from an external database with the ref_term predicate, it places that term on the global stack. The system won't release the space occupied by that term until the program fails and backtracks to a point before the call to ref_term. This means that, to do a sequential scan through a chain in an external database, you should always use a structure like the following: /* Structure for sequentially scanning through a chain */ scan(db_selector, Chain, ....) :chain_first(db_selector, Chain, Ref), scanloop(db_selector, Ref). scanloop(db_selector, Ref) :ref_term(db_selector, mydom, Ref, Term), /* ... do your processing ... */ fail. scanloop(db_selector, _) :chain_next(db_selector, Ref, NextRef), scanloop(db_selector, NextRef).

Similarly, for a sequential scan through an index, you should use a structure like this: /* Structure for sequentially scanning through an index */ scan(db_selector, Bt_selector) :key_first(db_selector, Bt_selector, FirstRef), scanloop(db_selector, Bt_selector, FirstRef).

scanloop(db_selector, Bt_selector, Ref) :ref_term(db_selector, mydom, Ref, Term), /* ... do your processing ... */ fail. scanloop(db_selector, Bt_selector, _) :key_next(db_selector, Bt_selector, NextRef), scanloop(db_selector, Bt_selector, NextRef).

You can also carry out a sequential scan through a chain in the database by using chain_terms, like this: /* Another way to sequentially scan through a chain */ scan(db_selector, Chain) :chain_terms(db_selector, Chain, mydom, Term, Ref), /* ... do your processing ... */ fail. scan(_, _).

To scan through a B+ tree, you could have also defined and used the predicate bt_keys. During backtracking, this predicate returns (for a given B+ tree and database) each key in the tree and its associated database reference number. /* This fragment goes with Program 5 */ PREDICATES bt_keys(db_selector, bt_selector, string, ref) bt_keysloop(db_selector, bt_selector, string, ref) CLAUSES bt_keys(Db_selector, Bt_selector, Key, Ref):key_first(Db_selector, Bt_selector, _), bt_keysloop(Db_selector, Bt_selector, Key, Ref). bt_keysloop(Db_selector, Bt_selector, Key, Ref):key_current(Db_selector, Bt_selector, Key, Ref). bt_keysloop(Db_selector, Bt_selector, Key, Ref):key_next(Db_selector, Bt_selector, _), bt_keysloop(Db_selector, Bt_selector, Key, Ref).

Displaying the Contents of a Database You can use the predicate listdba, defined in the following program fragment, to display the current state of an external database. listdba has one argument: the selector of a database assumed to be open. All terms in the database must belong Chapter 170 Advanced topics

351

to the same domain. In the example, the domain is called mydom; when you use this predicate, you must replace mydom with the actual name of the appropriate domain in your program. /* Program ch249e05.pro */ CONSTANTS filename = "\\vip\\vpi\\programs\\register\\exe\\register.bin" DOMAINS db_selector = mydba mydom = city(zipcode, cityname); person(firstname, lastname, street, zipcode, code) zipcode, cityname, firstname, lastname, street, code = string PREDICATES listdba(db_selector) nondeterm bt_keys(db_selector,bt_selector,string,ref) nondeterm bt_keysloop(db_selector,bt_selector,string,ref) CLAUSES listdba(Db_selector):-nl, write("********************************************"),nl, write(" DATABASE LISTING"),nl, write("********************************************"), db_statistics(Db_selector,NoOfTerms,MemSize,DbaSize,FreeSize),nl,nl, write("Total number of records in the database: ",NoOfTerms),nl, write("Number of bytes used in main memory: ",MemSize),nl, write("Number of bytes used by the database: ",DbaSize),nl, write("Number of bytes free on disk: ",FreeSize),nl, fail. listdba(Db_selector):db_chains(Db_selector,Chain),nl,nl,nl,nl, write("******* Chain LISTING *************"),nl,nl, write("Name=",Chain),nl,nl, write("CONTENT OF: ",Chain),nl, write("------------------------------\n"), chain_terms(Db_selector, Chain, mydom,Term, Ref), write("\n", Ref, ": ",Term), fail.

listdba(Db_selector):db_btrees(Db_selector,Btree), /* Returns each B+ tree name */ bt_open(Db_selector,Btree,Bt_selector), bt_statistics(Db_selector,Bt_selector,NoOfKeys, NoOfPages,Dept,KeyLen,Order,PageSize),nl,nl,nl, write("******** INDEX LISTING **************"),nl,nl, write("Name= ", Btree),nl, write("NoOfKeys= ", NoOfKeys),nl, write("NoOfPages=", NoOfPages),nl, write("Dept= ", Dept),nl, write("Order= ", Order),nl, write("KeyLen= ", KeyLen),nl, write("PageSize= ", PageSize), nl, write("CONTENT OF: ", Btree),nl, write("-----------------------------\n"), bt_keys(Db_selector,Bt_selector,Key,Ref), write("\n",Key, " - ",Ref), fail. listdba(_). bt_keys(Db_selector,Bt_selector,Key, Ref):key_first(Db_selector,Bt_selector,_), bt_keysloop(Db_selector,Bt_selector,Key,Ref). bt_keysloop(Db_selector,Bt_selector,Key,Ref):key_current(Db_selector,Bt_selector,Key,Ref). bt_keysloop(Db_selector,Bt_selector,Key,Ref):key_next(Db_selector,Bt_selector,_), bt_keysloop(Db_selector,Bt_selector,Key,Ref). GOAL db_open(mydba,filename,in_file), listdba(mydba).

Implementing a Database That Won't Break Down If you enter a lot of new information into a database, it is important to ensure that this information won't be lost if the system goes down. In this section, we illustrate one way of doing this -- by logging all changes in another file. Making a change involves first updating the database, and then flushing it. If this operation succeeds, the system then records the change in the log file and flushes the log file itself. This means that only one file is unsafe at any given time. If the database file becomes invalid (because the system went down before the file was flushed, for example), you should be able to reconstruct it by merging the log file Chapter 170 Advanced topics

353

with a backup of the database file. If the log file becomes invalid, you should create a new log file and make a backup of the database file. If you record the date and time in the log file, together with the old values from a modification involving replacement or deletion, you should be able to reconstruct the database to its state at a given time. /* This program fragment goes with Program 6 */ DOMAINS logdom = insert(relation,dbdom,ref); replace(relation,dbdom,ref,dbdom); erase(relation,ref,dbdom) PREDICATES logdbchange(logdom) CLAUSES logdbchange(Logterm):chain_insertz(logdba,logchain,logdom,Logterm,_), db_flush(logdba).

Updating the Database As a general principle, you shouldn't spread database updating throughout the program but should keep it in some user-defined predicates. This makes it easier to change the database and/or to add new B+ trees. When you confine updating this way, it's also easier to make a robust database system because your program involves only a small piece of code in which the database is unsafe. The following example handles updating two different relations, whose objects are all strings: person(firstname, lastname, street, zipcode, code) city(zipcode, cityname)

It handles the updating with the following indexes (keys) on the person and city relations: Person's Name............Last Name plus First Name Person's Address.........Street Name City Number..............Zip Code

In this example, we assume that the B+ trees are already open, and that their bt_selectors have been asserted in the database predicate indices.

Before this program initiates the updating, it eliminates the possibility of a BREAK with the break predicate. After updating is finished, the program flushes the database with db_flush. Although db_flush makes the updating a slow process (thanks to DOS), the file will be safe after this call. To make the system as secure as possible, the program logs changes in a special file through a call to logdbchange. /* Program ch250e06.pro */ /* Logging database operations */ DOMAINS logdom = insert(relation,dbdom,ref); replace(relation,dbdom,ref,dbdom); erase(relation,ref,dbdom) PREDICATES logdbchange(logdom) CLAUSES logdbchange(Logterm):chain_insertz(logdba,logchain,logdom,Logterm,_), db_flush(logdba). DOMAINS dbdom = city(zipcode, cityname); person(firstname, lastname, street, zipcode, code) zipcode, cityname, firstname, lastname = string street, code = string indexName = person_name; person_adr; city_no relation = city; person db_selector = dba; logdba DATABASE % This takes and index name (a key) that is a person's name or address %or a city number; it also takes a B+ tree selector indices(IndexName, bt_selector) PREDICATES %and a first name (10 characters) % This predicate creates an index name from a last name (20 characters) xname(FirstName,LastName,string) CLAUSES xname(F,L,S):str_len(L,LEN),LEN>20,!, frontstr(20,L,L1,_), format(S,"%-20%",L1,F).

Chapter 170 Advanced topics

355

xname(F,L,S):format(S,"%-20%",L,F). PREDICATES ba_insert(relation, dbdom) dba_replace(relation, dbdom, Ref) dba_erase(relation, Ref) CLAUSES dba_insert(person,Term):-!, break(OldBreak), break(off), indices(person_name,I1), indices(person_adr,I2),!, Term = person(Fname,Lname,Adr,_,_), xname(Fname,Lname,Xname), chain_insertz(dba,person,dbdom,Term,Ref), key_insert(dba,I1,Xname,Ref), key_insert(dba,I2,Adr,Ref), db_flush(dba), logdbchange(insert(person,Term,Ref)), break(OldBreak). dba_insert(city,Term):break(OldBreak), break(off), indices(city_no,I),!, Term = city(ZipCode,_), chain_insertz(dba,city,dbdom,Term,Ref), key_insert(dba,I,ZipCode,Ref), db_flush(dba), logdbchange(insert(city,Term,Ref)), break(OldBreak).

dba_replace(person,NewTerm,Ref):-!, break(OldBreak), break(off), indices(person_name,I1), indices(person_adr,I2),!, ref_term(dba,dbdom,Ref,OldTerm), OldTerm=person(OldFname,OldLname,OldAdr,_,_), xname(OldFname,OldLname,OldXname), key_delete(dba,I1,OldXname,Ref), key_delete(dba,I2,Oldadr,Ref), NewTerm=person(NewFname,NewLname,NewAdr,_,_), xname(NewFname,NewLname,NewXname), term_replace(dba,dbdom,Ref,NewTerm), key_insert(dba,I1,NewXname,Ref), key_insert(dba,I2,NewAdr,Ref), db_flush(dba), logdbchange(replace(person,NewTerm,Ref,OldTerm)), break(OldBreak). dba_replace(city,NewTerm,Ref):-!, break(OldBreak), break(off), indices(city_no,I),!, ref_term(dba,dbdom,Ref,OldTerm), OldTerm=city(OldZipCode,_), key_delete(dba,I,OldZipCode,Ref), NewTerm=city(ZipCode,_), term_replace(dba,dbdom,Ref,NewTerm), key_insert(dba,I,ZipCode,Ref), db_flush(dba), logdbchange(replace(city,NewTerm,Ref,OldTerm)), break(OldBreak).

Chapter 170 Advanced topics

357

dba_erase(person,Ref):-!, break(OldBreak), break(off), indices(person_name,I1), indices(person_adr,I2),!, ref_term(dba, dbdom, Ref, OldTerm), OldTerm=person(OldFname,OldLname,OldAdr,_,_), xname(OldFname,OldLname,OldXname), key_delete(dba,I1,OldXname,Ref), key_delete(dba,I2,OldAdr,Ref), term_delete(dba,person,Ref), db_flush(dba), logdbchange(erase(person, Ref, OldTerm)), break(OldBreak). dba_erase(city,Ref):break(OldBreak), break(off), indices(city_no,I),!, ref_term(dba,dbdom,Ref,OldTerm), OldTerm=city(OldZipCode,_), key_delete(dba,I,OldZipCode,Ref), term_delete(dba,city,Ref), db_flush(dba), logdbchange(erase(city,Ref,OldTerm)), break(OldBreak).

Using Internal B+ Tree Pointers Each open B+ tree has an associated pointer to its nodes. When you open or update the B+ tree, this pointer is positioned before the start of the tree. When you call key_next with the pointer at the last key in the tree, the pointer will be positioned after the end of the tree. Whenever the pointer moves outside the tree, key_current fails. If this arrangement is not appropriate for a particular application, you can model other predicates. You can use mykey_next, mykey_prev , and mykey_search, defined in this example, to ensure that the B+ tree pointer is always positioned inside the B+ tree (provided there are any keys in the tree). PREDICATES mykey_next(db_selector, bt_selector, ref) mykey_prev(db_selector, bt_selector, ref) mykey_search(db_selector, bt_selector, string, ref)

CLAUSES mykey_prev(Dba, Bt_selector, Ref) :key_prev(Dba, Bt_selector, Ref), !. mykey_prev(Dba, Bt_selector, Ref) :key_next(Dba, Bt_selector, Ref), fail. mykey_next(Dba, Bt_selector, Ref) :key_next(Dba, Bt_selector, Ref), !. mykey_next(Dba, Bt_selector, Ref) :key_prev(Dba, Bt_selector, Ref), fail. mykey_search(Dba, Bt_selector, Key, Ref) :key_search(Dba, Bt_selector, Key, Ref), !. mykey_search(Dba, Bt_selector, _, Ref) :key_current(Dba, Bt_selector, _, Ref), !. mykey_search(Dba, Bt_selector, _, Ref) :key_last(Dba, Bt_selector, Ref).

You can use the samekey_next and samekey_prev predicates, defined in the next example, to move the index pointer to the next identical key in a B+ tree that has duplicate keys. PREDICATES samekey_next(db_selector, bt_selector, ref) try_next(db_selector, bt_selector, ref, string) samekey_prev(db_selector, bt_selector, ref) try_prev(db_selector, bt_selector, ref, string) CLAUSES Samekey_next(Dba, Bt_selector, Ref) :key_current(Dba, Bt_selector, OldKey, _), try_next(Dba, Bt_selector, Ref, OldKey). try_next(Dba, Bt_selector, Ref, OldKey) :key_next(Dba, Bt_selector, Ref), key_current(Dba, Bt_selector, NewKey, _), NewKey = OldKey, !. try_next(Dba, Bt_selector, _, _) :key_prev(Dba, Bt_selector, _), fail. samekey_prev(Dba, Bt_selector, Ref) :key_current(Dba, Bt_selector, OldKey, _), try_prev(Dba, Bt_selector, Ref, OldKey).

Chapter 170 Advanced topics

359

try_prev(Dba, Bt_selector, Ref, OldKey) :key_prev(Dba, Bt_selector, Ref), key_current(Dba, Bt_selector, NewKey, _), NewKey = OldKey, !. try_prev(Dba, Bt_selector, _, _) :key_next(Dba, Bt_selector, _), fail.

Changing the Structure of a Database One way to change the structure of a database is to write a small program that copies the old database to a new one while making external databases, changing structure of the changes. Another way, which we'll describe here, is to first dump the database into a text file, make any necessary modifications to the database with a text editor, and then read the modified database back into a new file. You can use the predicate dumpDba, defined in the next program fragment, to dump the contents of an external database into a text file if the database satisfies the following conditions: Every chain in the database models a relation. All terms in the database belong to the same domain. This method does not dump the B+ trees into the text file; we assume, given the first condition, that B+ trees can be generated from the relations. In this example, all terms belong to the generic domain mydom; when you implement this method, replace mydom with the actual name and a proper declaration. This code writes the contents of the database to a text file opened by outfile. Each line of the text file contains a term and the name of the containing chain. The term and the chain names are combined into the domain chainterm. /* Program ch251e07.pro */ CONSTANTS filename = "\\vip\\vpi\\programs\\register\\exe\\register.bin" DOMAINS Db_selector = myDba chainterm = chain(string, mydom) file = outfile mydom = city(zipcode, cityname); person(firstname, lastname, street, zipcode, code) zipcode, cityname, firstname, lastname = string street, code = string

PREDICATES wr(chainterm) dumpDba(string,string) CLAUSES wr(X):write(X),nl. dumpDba(Db_selector,OutFile):db_open(myDba,Db_selector,in_file), openwrite(outfile,OutFile), writedevice(outfile), db_chains(myDba,Chain), chain_terms(myDba,Chain,mydom,Term,_), wr(chain(Chain,Term)), fail. dumpDba(_,_):closefile(outfile), db_close(myDba). GOAL dumpDba(filename,"register.txt").

Now, using your customized version of this code, you can generate the text file by calling dumpDba, and you can reload the database by using readterm with the chainterm domain. The predicate dba_insert, which we defined in "Updating the Database" (page 354), takes care of the updating. DOMAINS chainterm = chain(string, dbdom) PREDICATES nondeterm repfile(file) copyDba loadDba(string) CLAUSES repfile(_). repfile(File) :- not(eof(File)), repfile(File).

Chapter 170 Advanced topics

361

loadDba(OutFile) :openread(Prn_file, OutFile), readdevice(Prn_file), repfile(Prn_file), readterm(Chainterm, chain(Chain, Term)), write(Term), nl, Dba_insert(Chain, Term), fail. loadDba(_) :closefile(Prn_file). copyDba :createDba, db_open(Dba, "register.bin", in_file), open_indices, loadDba("register.txt"), db_close(Dba).

Filesharing and the External Database Visual Prolog supports file-sharing the external database. This means that a file can be opened by several users or processes simultaneously, which will be useful if you are using the external database in a LAN-application or with one of the multitasking platforms. UNIX developers should take note that Visual Prolog uses advisory filelocking. Visual Prolog provides the following file-sharing facilities: opening an existing database with two different access modes and three different sharing modes for optimal speed. grouping database accesses in transactions to ensure consistency predicates that make it possible to check whether other users have updated the database.

Filesharing Domains The two special domains which are used for file-sharing have the alternatives: Domain

Functors

accessmode

= read; readwrite

denymode

= denynone; denywrite; denyall

Opening the Database in Sharemode In order to access the external database in share mode, you must open an already existing database file with the four arity version of db_open, specifying AccessMode and DenyMode. If AccessMode is read the file will be opened as readonly, and any attempts to update the file will result in an run-time error, if it is readwrite the file is opened for both reading and writing. AccessMode is also used with the predicate db_begintransaction. If DenyMode is denynone all other users will be able to both update and read the file, if it is denywrite, other users will not be able to open the file in accessmode = readwrite, but you will be able to update the file providing it was opened in accessmode = readwrite. If db_open is called with denymode = denyall no other users will be able to access the file at all. The first user that opens the file determines DenyMode for all subsequent attempts to open the file, and a run-time error will occur if reopened in an incompatible mode. The following table summarizes the results of opening and subsequently attempting to reopen the same file for all combinations of DenyMode and AccessMode: 2ND, 3RD, ..... REOPEN Denyall Denywrite Denynone R : AccessMode = read RW: AccessMode = readwrite Y : Open by 2ND, 3RD .. allowed N : Open by 2ND, 3RD .. not allowed

Transactions and Filesharing If a database file is opened in share mode, all database predicates that access the database file in any way, must be grouped inside "transactions" this is done by surrounding the calls to the predicates with db_begintransaction and db_endtransaction. Dependent on the combination of the chosen AccessMode and DenyMode the shared file may be locked for the duration of the transaction. Again dependent on the severity of the lock, other users may not be able to either read or update the file, while your transaction takes place. This is of course necessary to avoid conflicts between reading and writing, but if file-sharing is to have any meaning, no excessive locking ought to take place. This can be avoided by keeping the Chapter 170 Advanced topics

363

transactions small (as short as possible) and only include those predicates that access the database inside the transaction. The concept of transactions in relation to file-sharing is very important. Two often conflicting requirements, namely database consistency and a minimum of file locking, must be fulfilled at the same time. db_begintransaction ensures that database consistency is maintained and that an appropriate locking of the file is effectuated. Several readers can access the file at the same time, but only one process at the time is allowed to update the database. The predicate db_setretry can be called to set for how long db_begintransaction will wait to gain access to the file before returning with a run-time error. Calling db_begintransaction with accessmode set to readwrite with a file opened with accessmode set to read will also result in a run-time error. If db_begintransaction is called, db_endtransaction must be called before a new call to db_begintransaction for the same database, otherwise a run-time error will occur. The following table summarizes the actions taken by db_begintransaction with different combinations of AccessMode and DenyMode: AccessMode read readwrite DenyMode- denynone WLock\Reload RWLock\Reload Actions : WLock : No write. Read allowed. RWLOCK : No read or write allowed. Reload : Reloading of file descriptors.

Since reloading and locking takes time, AccessMode and DenyMode should be selected with care. If no users are going to update the database, set AccessMode to read and DenyMode to denywrite for a minimal overhead. Filesharing Predicates In this section we discuss the file sharing predicates db_open, db_begintransaction, db_endtransaction, db_updated, bt_updated, and db_setretry. db_open/4 This four arity version of db_open opens an existing database on file in share mode. db_open(Dbase, Name, AccessMode, DenyMode)

/* (i,i,i,i) */

After creating an external database (in_file) with db_create it can be opened in share mode, where Dbase is a db_selector, Name is the DOS-style file name, AccessMode is read or readwrite, and DenyMode is denynone, denywrite, or denyall. db_begintransaction/2 db_begintransaction(Dbase, AccessMode)

/* (i,i) */

This predicate marks the beginning of a transaction, and must be called prior to any form of access to a database opened in share mode, even if opened with denyall. In addition to the db_selector for the database, db_begintransaction must be called with AccessMode bound to either read or readwrite. db_endtransaction/1 db_endtransaction(Dbase)

/* (i) */

db_endtransaction marks the end of a transaction and carries out the appropriate unlocking of the database. A call of db_endtransaction without a prior call to db_begintransaction for the db_selector Dbase will result in an run-time error. db_updated/1 db_updated(Dbase)

/* (i) */

If other users have updated the database, a call of db_begintransaction will ensure that database consistency is maintained. Changes can be detected with the predicate db_updated, which succeeds if called inside a transaction where changes made by other users since your last call of db_begintransaction. If no changes have been made, db_updated will fail. If called outside a transaction a run-time error will occur. bt_updated/2 bt_updated(Dbase,Btree_Sel)

/* (i,i) */

Similar to db_updated/1, but only succeeds if the named B+ tree has been updated. db_setretry/3 db_setretry(Dbase,SleepPeriod,RetryCount)

Chapter 170 Advanced topics

/* (i,i,i) */

365

If access to a file is denied, because another process has locked the file, you can have your process wait for a period of time and then try again. The predicate db_setretry changes the default settings of SleepPeriod, which is the interval in centiseconds between retries, and RetryCount, which is the maximum number of times access will be attempted. The default settings are 100 for RetryCount and 10 for SleepPeriod.

Programming with Filesharing Great care must be taken when using the file sharing predicates. Although they, when used properly, ensure low-level consistency in a shared database, it is the application programmers responsibility to provide the demanded high level consistency for a given application. The term "transaction" is used here for a group of file accesses, but it should be kept in mind that no back out facilities are provided, and that program interruption caused by either software or hardware failure, may cause inconsistencies in the database file. When several processes share a database, special attention must also be paid to the domains involved. It's crucial that they are identical and use identical alignment. To avoid unnecessary locking of the database file the transactions should be kept fairly small, in order to ensure that the file will be locked for as short a time as possible. At the same time it is important that predicates used to locate and access an item in the database are grouped inside the same transaction: ..... db_begintransaction(dba,readwrite), key_current(dba,firstindex,Key,Ref), ref_term(dba,string,Ref,Term), db_endtransaction(dba), write(Term), .....

In this example the predicates key_current and ref_term should not be placed inside different transactions, as the term stored under Ref may be deleted by another user between transactions. If a B+ tree is updated by another user and the file buffers are reloaded, the B+ tree will be repositioned before the first element of the tree. By calling the predicate bt_updated you can detect when to reposition your B+ tree. It is still possible to list the entire index and at the same time keep the transactions small, by temporarily storing the current key in the internal database, as shown in the following program fragment. It works under the assumption that no duplicate keys exist.

DOMAINS db_selector = dba DATABASE determ currentkey(string) PREDICATES list_keys(bt_selector) list_index(bt_selector) check_update(bt_selector,string) CLAUSES check_update(Index,Key):not(bt_updated(dba,Index)),!, key_next(dba,Index,_). check_update(Index,Key):key__search(dba,Index,Key,_),!. % Will fail if current was deleted check_update(_,_). %by another user list_keys(Index):currentkey(Key), write(Key),nl, db_begintransaction(dba,read), check_update(Index,Key), key_current(dba,Index,NextKey,_), db_endtransaction(dba),!, retract(currentkey(_)), assert(currentkey(NextKey)), list_keys(Index). list_keys(_):db_endtransaction(dba). list_index(Index):db_begintransaction(dba,read), key_first(dba,Index,_), key_current(dba,Index,Key,_), db_endtransaction(dba), retractall(currentkey(_)), assert(currentkey(Key)), list_keys(Index). list_index(_).

key_search is used to reposition the B+ tree at the key that was listed previously. The my_search predicate insures that the B+ tree will be correctly positioned even if currentkey was deleted by another user.

Chapter 170 Advanced topics

367

The example above also illustrates another important point. A db_endtransaction must be used after each, and before the next, call of db_begintransaction. In the predicate list_keys above, the listing stops when key_next fails, indicating that all the keys have been listed. As db_begintransaction had to be called prior to accessing the database, db_endtransaction has to be called as well after accessing is completed. The second list_keys-clause ensures that db_endtransaction will be called when key_next fails.

Implementing highlevel locking The examples shown so far have illustrated some of the problems involved in file sharing, and how they can be avoided. You are allowed to do all the same operations on a shared database file as if you were the only user with access to the file. Grouping the accesses to the file inside db_begintransaction and db_endtransaction ensures that the Visual Prolog system has consistency in its descriptor tables. But on a higher level you must yourself ensure that the various logical constraints you have on your application are conserved over a network with multiple users. We call this high level locking or application level locking. By using the primitives db_begintransaction and db_endtransaction you have many ways of implementing a high level locking facility. A common example of where high level locking is needed is in a database system where a user wants to change a record. When he has decided that he wants to change a record he should perform some kind of action so the application will place a lock on that record until the user has finished the changes to the record so the new record can be written back to disk, and the record unlocked. Some suggestions for implementing an application-level lock of this type are: Have a special field in that record to tell whether it is locked. Have a special B+Tree or a chain where you store all references to all the records that are locked by users. Associated with a REF store a list of references to all records that are locked. You might need to implement a kind of supervisor mechanism so a special user can unlock locked records. This was just an example, you might want to implement locking on a higher level like tables or groups of tables, - or knowledge groups etc.

Note: If you want to delete a B+ tree in a database file opened in share mode, it is up to you to ensure by high level locking that no other users have opened this B+ Tree. In the Visual Prolog system there is no check for a B+Tree selector being no longer valid because the B+Tree has been deleted by another user.

A Complete Filesharing Example In the following large example it will be shown how file sharing can be done more easily by implementing your own locking system. If you manage your own locks, needless file locking can be avoided, and other users won't have to wait for access to the file because it is locked. The example is the file sharing version of the previous 4 example. The program lets several users create, edit, view and delete texts from a single shared file. When creating and editing a text, it will be locked until editing is complete. Other users cannot delete or edit a text while it is locked, but they will be able to view the text. Run the program and experiment with different settings for db_open and db_setretry. /* Program ch252e08.pro */ DATABASE - indexes determ lockindex(bt_selector) determ index(bt_selector) determ mark(real) DOMAINS my_dom = f(string) db_selector = dba PREDICATES nondeterm repeat wr_err(integer) % List texts and their status list list_texts(bt_selector,bt_selector) show_textname(string,bt_selector) CLAUSES show_textname(Key,LockIndex):key_search(dba,LockIndex,Key,_),!, write("\n*",Key). show_textname(Key,_):write("\n ",Key).

Chapter 170 Advanced topics

369

list_texts(Index,LockIndex) :key_current(dba,Index,Key,_), show_textname(Key,LockIndex), key_next(dba,Index,_),!, list_texts(Index,LockIndex). list_texts(_,_). list:-nl, write("***************** TEXTS (*=Locked) *******************"),nl, index(Index), lockindex(LockIndex), key_first(dba,Index,_),!, list_texts(Index, LockIndex),nl, write("******************************************************"),nl. list. repeat. repeat:-repeat. wr_err(E):errormsg("PROLOG.ERR",E,Errormsg,_), write(Errormsg), readchar(_). PREDICATES %Logical locking of files lock(string,bt_selector,bt_selector) CLAUSES lock(Name,Index,LockIndex):not(key_search(dba,LockIndex,Name,_)),!, key_search(dba,Index,Name,Ref), key_insert(dba, LockIndex, Name, Ref). lock(Name,_,_):db_endtransaction(dba), write(Name," is being updated by another user.\n Access denied"), fail. PREDICATES ed(db_selector, bt_selector, bt_selector, string) ed1(db_selector, bt_selector, bt_selector, string) CLAUSES % The ed predicates ensure that the edition will never fail. ed(dba,Index,LockIndex,Name):ed1(dba,Index,LockIndex,Name),!. ed(_,_,_,_).

/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * There are two choices: * * * * 1) The name already exists - modify the contents of the * * file * * 2) The name is a new name - create a new file * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */ ed1(dba, Index,LockIndex, Name) :db_begintransaction(dba,readwrite), key_search(dba, Index, Name, Ref),!, ref_term(dba, string, Ref, Str), lock(Name,Index,LockIndex), list, db_endtransaction(dba),nl, write("******************************************************"),nl, write("* EDIT ",Name," *"),nl, write("******************************************************"),nl, write(Str),nl, write("< Press 'r' to replace this string ; else any key >"),nl, readchar(X),X='r',nl, write("Enter string and press <ENTER>"),nl, readln(Str1),nl, db_begintransaction(dba,readwrite), term_replace(dba, string, Ref, Str1), key_delete(dba, LockIndex, Name, Ref), %unlock list, db_endtransaction(dba). %New file ed1(dba, Index,LockIndex, Name):chain_insertz(dba, file_chain, string, "", Ref), key_insert(dba, Index, Name, Ref), list, db_endtransaction(dba), ed1(dba,Index,LockIndex, Name). PREDICATES main(db_selector, bt_selector, bt_selector) interpret(char, bt_selector, bt_selector) check_update_view update_view get_command(char)

Chapter 170 Advanced topics

371

CLAUSES % Loop until 'Q' is pressed main(dba,Index,LockIndex) :check_update_view, get_command(Command), trap(interpret(Command,Index,LockIndex),E,wr_err(E)),!, main(dba,Index,LockIndex). main(_,_,_). check_update_view:mark(T),timeout(T),!, db_begintransaction(dba,read), update_view, db_endtransaction(dba), marktime(100,Mark), retractall(mark(_)), assert(mark(Mark)). check_update_view. update_view:-nl, write("******* COMMANDS E:Edit V:View D:DeleteQ:Quit *******"),nl, write("COMMAND>"), db_updated(dba),!, list. update_view. get_command(Command):readchar(C),!, upper_lower(Command,C), write(Command),nl. get_command(' ').

%interpret commandlineinput interpret(' ',_,_):-!. interpret('Q',_,_):-!,fail. interpret('E',Index,LockIndex):-!, write("\nFile Name: "), readln(Name),nl, ed(dba,Index,LockIndex,Name). interpret('V',Index,_):write("\nFile Name: "), readln(Name),nl, db_begintransaction(dba,read), key_search(dba,Index,Name,Ref),!, ref_term(dba,string,Ref,Str), db_endtransaction(dba), write("******************************************************"),nl, write("* VIEW ",Name," "),nl, write("******************************************************"),nl, write(Str),nl. interpret('V',_,_):-!, db_endtransaction(dba). interpret('D',Index,_):write("\nDelete file: "), readln(Name),nl, db_begintransaction(dba,readwrite), key_search(dba,Index,Name,Ref),!, % not(key_search(dba,LockIndex,Name,_)),!, key_delete(dba,Index,Name,Ref), term_delete(dba,file_chain,Ref), list, db_endtransaction(dba). interpret('D',_,_):-!, db_endtransaction(dba). nterpret(_,_,_):-beep. PREDICATES open_dbase(bt_selector,bt_selector) CLAUSES open_dbase(INDEX,LOCKINDEX):existfile("share.dba"),!, db_open(dba, "share.dba",readwrite,denynone), db_begintransaction(dba,readwrite), bt_open(dba, "locks", LOCKINDEX), bt_open(dba, "ndx", INDEX), db_endtransaction(dba).

Chapter 170 Advanced topics

373

open_dbase(INDEX,LOCKINDEX):db_create(dba,"share.dba" , in_file), bt_create(dba, "locks",TEMPLOCKINDEX,20, 4), bt_create(dba, "ndx",TEMPINDEX , 20, 4), bt_close(dba, TEMPINDEX), bt_close(dba, TEMPLOCKINDEX), db_close(dba), open_dbase(INDEX,LOCKINDEX). GOAL open_dbase(INDEX,LOCKINDEX), assert(index(INDEX)), assert(lockindex(LOCKINDEX)), marktime(10,Mark), assert(mark(Mark)), db_setretry(dba,5,20), db_begintransaction(dba,read), list,nl, db_endtransaction(dba), main(dba, INDEX,LOCKINDEX), db_begintransaction(dba,read), bt_close(dba, INDEX), bt_close(dba, LOCKINDEX), db_endtransaction(dba), db_close(dba).

Implementation Aspects of Visual Prolog Filesharing Filesharing in Visual Prolog is efficient and fast, because only the necessary parts of the database file descriptors are loaded after an update by another user. As was shown earlier in this chapter it is only under certain circumstances that any reloading of file buffers and locking of files has to be done at all, and the complex internal management of the database file ensures that after an update a minimum of disk activity is needed. The database has a serial number, which is a six byte integer, that is incremented and written to disk each time an update occurs. The db_begintransaction predicate compares the local copy of the serial number with the one on the disk, and if they differ, the descriptors are reloaded. Locking is done in an array with room for 256 readers. When a reader wishes to access the file, an unlocked space is located in this lock array, and locked for the duration of the transaction. This allows several readers to access the file simultaneously. If db_begintransaction is called with AccessMode = readwrite, it will wait until all present readers have

unlocked their space, and then lock the entire array, allowing no other users to access the file.

Miscellaneous Finally, we have provided a couple of small predicates that are handy in special circumstances. The predicate availableems will in DOS return the amount of available ems. This can be used before a call to db_open or db_create in order to see if there is enough space for placing the database in_ems. availableems(Size)

/* (real)-(o) */

Another predicate str_ref can be used to convert a database reference number to a string so it can be inserted in a B+ Tree. str_ref(Str,Ref)

/* (string,ref)-(i,o)(o,i)(i,i) */

Summary Visual Prolog's external database system adds power, speed, and efficiency to your database applications. These are the major points covered in this chapter: 1. External database terms are stored in chains, which you can access directly with database reference numbers; these reference numbers belong to the predefined ref domain. 2. Individual databases are identified by a database selector, which belongs to the standard domain db_selector. 3. You can store your external database in three locations, depending on which alternative you use for the predefined place domain: a. in_file places the database in a disk file b. in_memory places it in memory c. and in_ems places it in EMS-type expanded memory (same effect as in_memory on non-DOS platforms 4. If you want to sort the terms in your database, you'll use B+ trees. Like databases, individual B+ trees are identified by a B+ tree selector of the standard domain bt_selector.

Chapter 170 Advanced topics

375

5. Each entry in a B+ tree node consists of a key string (also known as an index), which identifies a record, and the database reference number associated with that record. 6. tree keys are grouped in pages, and the number of keys stored at a node is specified by the tree's order. 7. Filesharing is done by grouping the predicates that access the database into transactions.

CHAPTER

253

System-Level Programming 254Visual Prolog provides several predicates that allow you to access your PC's operating system and - to the extent that operating system allows - the hardware directly. We summarize those predicates in this chapter, first taking a look at the ones giving access to the OS, then those that perform bit-level logical and shifting operations. After that, we discuss a set of predicates that provide low-level support for manipulating the DOS BIOS, memory, and other hardware elements. We end this chapter with a couple of examples that demonstrate how to use some of these predicates within a Visual Prolog application.

Access to the operating system With a handful of predicates, you can access the operating system while running the Visual Prolog integrated environment, as well as build the ability to access the run-time computer's operating system right into your Visual Prolog applications. You can execute any external program with a call to system, call the date and time facilities with date and time, investigate the environment table with envsymbol, and read the command-line arguments with comline. Furthermore, you can establish the start-up directory and exe-filename of the program by calling syspath, and the marktime, the timeout and the sleep predicates provide time-tunneling capacity. Then there's the inevitable sound and beep predicates, and finally osversion returning the operating system version, diskspace returning the amount of free disk space, and three versions of storage used to determine memory used.

This section describes each of these predicates in detail and provides some practical examples that demonstrate how to use them. system/1 Visual Prolog programs provide access to the OS through the system predicate, which takes the following form: system("command")

/* (i) */

If the argument is an empty string (""), a new command interpreter will be run in interactive mode. Examples 1. To copy the file B:ORIGINAL.FIL to a file A:NEWCOPY.FIL from within the Visual Prolog system, you could give the goal system("").

then copy the file using the usual command, copy b:original.fil newcopy.fil

You could then return to Visual Prolog by typing exit

after which you are back in your program again. 2. To rename the file (without going out to the OS), you could give the command system("ren newcopy.fil newcopy.txt").

system/3 This extended version of the system predicate provides two extra features: one for returning the OS error level, and one for resetting the run-time system's video mode. The latter has no effect in OS/2 or Windows. In UNIX, this argument is used to indicate that the process has no interaction with the terminal, and hence that there's no need to clear and reset it. This is a somewhat unfortunate dual use of the same argument, but it fulfills the typical needs of users. system/3 takes this format: system(CommandString, ResetVideo, ErrorLevel)

Chapter 170 Advanced topics

/* (i,i,o) */

377

The error level is returned in ErrorLevel. This is the program return code known by the OS at the time control returns to the program issuing the system call. In DOS and OS/2 this is only available for .COM and .EXE files. In textmode DOS, ResetVideo controls whether your program should reset the video hardware to the state it was in before system/3 was called. ResetVideo = 1 resets the video mode; ResetVideo = 0 does not. When ResetVideo = 0, your program will run in the new video mode you set, even if that's a mode not specifically supported by Visual Prolog. (For information about setting the runtime system's video mode, refer to the reference manual for the video hardware.) In other words, if your external program MYSETMD sets the video hardware to a mode not specifically supported by Visual Prolog, and you place the following calls to system in your Visual Prolog program (running from the development environment), you can actually make your program run in that unsupported mode: system("mysetmd", 0, ErrorLevel),

Note: The external program must be compatible with the hardware at least at the BIOS level (updating the BIOS variables rows and columns on-screen). envsymbol/2 The envsymbol predicate searches for environment symbols in the application's environment table; the SET (OS) commands set these symbols. envsymbol takes this format: envsymbol(EnvSymb, Value)

/* (i,o) */

For example, the command SET SYSDIR=C:\FOOL

sets the symbol SYSDIR to the string C:\FOOL, and the goal /*...*/ envsymbol("SYSDIR", SysDir), /*...*/

searches the environment for the symbol SYSDIR, binding SetValue to C:\FOOL. envsymbol will fail if the symbol does not exist.

date and time Visual Prolog has three more handy OS-related standard predicates: two forms of date and time. The date/3 and time/3 predicates can be used in two ways, depending on whether their parameters are free or bound on entry. With input flow, time and date will set the internal system clock to the time specified (in UNIX you need root privileges to do this). If all variables are free, the system will bind them to the internal clock's current values. time(Hours, Minutes, Seconds, Hundredths) /* (i,i,i,i), (o,o,o,o) */

Note that the UNIX version of time doesn't return anything useful in the Hundredths argument. date/3 also relies on the internal system clock and operates similarly to time; it takes the following form: date(Year, Month, Day)

/* (i,i,i), (o,o,o) */

date/4 only has an output flow version. The fourth argument is the weekday number, but what numbering scheme is used is operating system dependent. However, it's fairly common that 0 is Sunday, 1 is Monday, etc. date(Year, Month, Day, WeekDay)

/* (o,o,o,o) */

Example Program 2 uses time to display the time elapsed during a listing of the default directory. /* Program ch255e02.pro */ GOAL time(H1,M1,S1,_),nl, write("Start time is: ",H1,":",M1,":",S1),nl, /* This is the activity that is being timed */ system("dir /s/b c:\\*.*"), time(H2,M2,S2,_), Time = S2-S1 + 60*(M2-M1 + 60*(H2-H1)), write("Elapsed time: ",Time," seconds"),nl, time(H3,M3,S3,_), write("The time now is: ",H3,":",M3,":",S3).

Chapter 170 Advanced topics

379

comline/1 comline reads the command-line parameters used when invoking a program; this is its format: comline(CommandLine)

/* (o) */

where CommandLine is a string. syspath/2 syspath returns the start-up directory and name of the program calling it. syspath looks as follows: syspath(HomeDir,ExeName)

/* (o,o) */

The main use for syspath is to provide programs the possibility of loading files from their home directory, as well as constructing helpful command-line error messages: <progname>: Usage: [-foul] <blah> <blah> <blah>. On UNIX, the start-up directory is not directly available to a program. In order to use syspath on UNIX, an initialization predicate, initsyspath, must be called. In particular, this must be called before the program changes its working directory, if this becomes necessary. If initsyspath isn't called, syspath will exit with an error code of 1138.

Timing Services Visual Prolog provides two different timing services: execution suspension, and elapsed-time testing. Some special notes apply to UNIX, see the description of difftime below. sleep/1 sleep suspends program execution for a specified length of time. It looks like this sleep(CSecs)

/* (i) */

where Csecs is the time (in centiseconds, i.e. 1/100ths) to sleep. The exact length of time the program will wait may vary, depending on CPU / OS activity, and you shouldn't expect greater accuracy than 20-50 milliseconds. In UNIX, sleep uses the nap(S) system call for delays and fractions of delays less than 1 second. This call uses the kernel's callout table, and it may be necessary to increase the size of this (kernel parameter NCALL) to prevent overflows if more than 10-20 processes simultaneously use sleep with fractional delays or nap(S).

marktime/2 marktime returns a time-stamp which may later be tested for expiration using the timeout predicate. marktime has the following format: marktime(CSecs,Ticket)

/* (i,o) */

where CSecs is the required length of time Ticket should last. The Ticket is an implementation-defined structure holding the timing information, currently masquerading as a real number. timeout/1 timeout tests a time-ticket returned by marktime for expiration. If it has expired, timeout succeeds, otherwise it fails. timeout looks like this: timeout(Ticket)

/* (i) */

As with sleep, don't expect too fine a granularity. difftime On UNIX, the standard predicate time doesn't provide a resolution in 100ths, so any timing calculations will be rather rough. However, the UNIX version of Visual Prolog has a standard predicate difftime: difftime(real,real,real)

/* (i,i,o) */

which returns the difference between the 1st and the 2nd timemark, in hundredths of seconds as a floating-point number. The first timemark should be the younger, and the second the older, i.e. the usage is marktime(0,M1), lengthy_process, marktime(0,M2), difftime(M2,M1,Diff).

In order for marktime and difftime to work, they must know how many clockticks the machine has per second. For UNIX executables, they establish this by calling the sysconf library function (see sysconf(S)), which is a very safe mechanism. However, for XENIX executables they have to call the library function gethz (see gethz(S)), which in it's current implementation simply examines a shell variable called HZ. Thus it is critical that this variable has the correct value, which, unless it's a whole new world when you read this, is 60. If gethz fails (e.g. because HZ doesn't exist), marktime will exit with error 1136. The same applies to difftime if either marktime has never been called, or if marktime exited due to failure in gethz. Chapter 170 Advanced topics

381

The granularity of sleep and the marktime and timeout predicates is systemdefined, currently being 1/60th of a second. Note that timemarks do not survive reboots. Under UNIX they're the number of machine clock-ticks since "an arbitrary point in the past" which in practice means system start-up. With 60 ticks/second, this also means that the tick count wraps around zero after approx. 2.26 years. Example Program 4 below demonstrates marktime and timeout. /* Program ch256e04.pro */ PREDICATES ttimeout(real) CLAUSES ttimeout(TM):-timeout(TM),!. ttimeout(TM):write("No timeout, sleep 0.5 secs"),nl, sleep(50), ttimeout(TM). GOAL marktime(400,TM), ttimeout(TM), write("\nBINGO!\n").

% 4 secs

sound/2 sound generates a sound in the PC's speaker: sound(Duration,Frequency)

/* (i,i) */

where Duration is the time in 1/100ths of a second. On UNIX, sound works only on the ANSI console; whether you're running on this is established by examining the TERM shell variable. On other terminals, sound is equivalent to beep. beep/0 beep

/* (no arguments) */

In the DOS-related versions of Visual Prolog, beep is equivalent to sound(50,1000).

On UNIX, beep writes the bell character to the file used for terminal output. If the program is in terminal mode, all buffering will be bypassed. osversion/1 osversion returns the current operating system version and looks like this: osversion(VerString)

/* (o) */

The format for VerString is operating system defined. For DOS and OS/2, it consists of the major and minor version numbers, separated by a dot (full stop), e.g. "3.30". Note that the major version number currently returned by OS/2 is 10, rather than 1. In UNIX, the string contains the information returned by uname(S). diskspace/2 diskspace returns as an unsigned long the available disk space, using the following format: diskspace(Where,Space)

/* (i,o) */

The space is reported in bytes. In the DOS-related versions of Visual Prolog, Where should be a character specifying the drive letter. In the UNIX version, it should be the name of a file residing on the file system you want to query (see statfs(S)). You may use simply "/" for the root file system, or an empty or null string in which case information is retrieved for the file system holding the current working directory. The space reported will be the smaller of the actual space and the ulimit for the process (see ulimit(S)). storage/3 The standard predicate storage returns information about the three run-time memory areas used by the system (stack, heap, and trail, respectively) as unsigned longs: storage(StackSize,HeapSize,TrailSize)

/* (o,o,o) */

The values are all in bytes. In all versions of Visual Prolog, TrailSize contains the amount of memory used by the trail. In the DOS-related versions, StackSize indicates how much stack space is left. In UNIX, StackSize is the exact opposite, namely how much stack that's been used so far. Chapter 170 Advanced topics

383

Finally, the HeapSize shows how much memory is available to the process. In UNIX this is the difference between the current break value and the maximum possible break value (see ulimit(S) and brk(S)), which again is set by the kernel parameter MAXUMEM. It does not include memory held in freelists in the heap. In DOS, the HeapSize is the unallocated physical memory between the top of the GStack and the bottom of the heap. It does not include memory held in freelists in the heap. The storage predicate returns the size that you can be sure of having when you're loading a file or going out to the operating system. In OS/2 the storage will be limited by the virtual memory which is limited by the size of the disk where the swap file resides. Since there is no easy way to find where the swap file resides, the storage predicate will for the heap just return the size of the largest block OS/2 can allocate without compacting and/or swapping memory. You can't really use that for anything, but it gives you an indication of memory fragmentation. If you feel certain that you know where the swap file resides, you can use the diskspace standard predicate to check the free space available. storage/0 The 0-arity version of storage is primarily intended for debugging purposes. It prints in the current window an overview of the amount of memory in use by the different parts of Visual Prolog's memory management, as well as the number of backtrack points.

Bit-Level Operations Visual Prolog provides six predicates for bit-level operations; bitor, bitand, bitnot, bitxor, bitleft, and bitright. These predicates have one flow variant each, operate on unsigned integers, and must be used in prefix notation. bitnot/2 bitnot performs a bit-wise logical NOT. bitnot(X, Z)

/* (i,o) */

With X bound to some integral value, Z will be bound to the bit-wise negation of X. Operator

bitnot

bitand/3 bitand performs a bit-wise AND. bitand(X, Y, Z)

/* (i,i,o) */

With X and Y bound to some integral values, Z will be bound to the result of bitwise ANDing the corresponding bits of X and Y.

Operator

bitand

bitor/3 bitor performs a bit-wise OR. bitor(X, Y, Z)

/* (i,i,o) */

With X and Y bound to some integral values, Z will be bound to the result of bitwise ORing the corresponding bits of X and Y. Operator

bitor

bitxor/3 bitxor performs a bit-wise XOR. bitxor(X, Y, Z)

Chapter 170 Advanced topics

/* (i,i,o) */

385

With X and Y bound to some integral values, Z will be bound to the result of bitwise XORing the corresponding bits of X and Y. Operator

bitxor

bitleft/3 bitleft performs a bit-wise left shift. bitleft(X, N, Y)

/* (i,i,o) */

With X and N are bound to some integral values, Y is bound to the result of shifting the bit-wise representation of X N places to the left. The new bits will be zero-filled. bitright/3 bitright performs a bit-wise right shift. bitright(X, N, Y)

/* (i,i,o) */

With X and N are bound to some integral values, Y is bound to the result of shifting the bit-wise representation of X N places to the right. The new bits will be zero-filled. Exercise Write a Visual Prolog program to test the theory that myxor(A, B, Result) :bitnot(B, NotB), bitand(A, NotB, AandNotB), bitnot(A, NotA), bitand(NotA, B, NotAandB), bitor(AandNotB, NotAandB, Result). behaves like bitxor(A, B, Result)

Access to the Hardware: Low-Level Support The DOS ROM-BIOS (Read Only Memory-Basic Input/Output System) provides an interface between programs and the operating system to perform various functions, including disk, file, printer, and screen I/O. For specific information on the ROM-BIOS, refer to the DOS Technical Reference Manual. Visual Prolog provides six built-in predicates that give low-level access to the operating system, I/O ports, and hardware. These predicates are bios (2 versions), ptr_dword, memword, membyte, and port_byte. This section describes each of these predicates in detail and provides some practical examples that demonstrate how to use them. bios/3 and bios/4 bios gives access to the PC's low-level BIOS (Basic I/O System) routines. For information about these routines, refer to your DOS Reference Manual. Note that the bios predicates only relate to DOS. Under UNIX, it's possible to access routines in shared libraries using the nlistnlist library call (see nlist(S)). However, the process is rather involved and won't be described here. See NLIST.PRO in the PROGRAMS directory for an example. Information passes to and from the BIOS functions through the predefined compound object reg(...). The bios predicate takes the following forms: bios(InterruptNo, RegistersIn, RegistersOut) bios(InterruptNo, RegistersIn, RegistersOut, OutFlags)

/* (i,i,o) */ /* (i,i,o,o) */

where RegistersIn and RegistersOut are data structures defined as follows: /* RegistersIn */ reg(AXi, BXi, CXi, DXi, SIi, DIi, DSi, ESi) /* (i,i,i,i,i,i,i,i) */ /* RegistersOut */ reg(AXo, BXo, CXo, DXo, SIo, DIo, DSo, ESo) /* (o,o,o,o,o,o,o,o) */

The bios predicates use the arguments AXi, BXi, CXi, DXi, SIi, DIi, DSi, and ESi to represent the PC's hardware register values passed to the BIOS. AXo, ... , ESo for those register values returned by the BIOS. Chapter 170 Advanced topics

387

The flow pattern for bios/3 is (i,i,o); for bios/4 it's (i,i,o,o). When you make a call to the BIOS, each argument of RegistersIn must be instantiated (bound to a value), and each argument of RegistersOut must be free (not bound to a value). The domain for the RegistersIn and RegistersOut compound objects (reg(AX, BX, ...)) is the reg domain, a predefined data structure created by Visual Prolog specifically for the bios predicate. Internally, Visual Prolog defines this data structure as DOMAINS reg = reg(integer, integer, integer, ..., integer)

The optional OutFlag argument in the bios/4 predicate is packed coding for the 8086 flag register (see Figure 257.1). OutFlag allows you to read the contents of the status flags after return from the interrupt. The flags are packed in an integer value as shown here: Figure 258.2: Packing the 8086 Flag Register in an Integer2593 _________________________________________________________________ | | | | | | | | | | | | | | | | | | U | U | U | U | O | D | I | T | S | Z | U | A | U | P | U | C | |___|___|___|___|___|___|___|___|___|___|___|___|___|___|___|___| | | | | | | | | | | | | | | | | | |15 |14 |13 |12 |11 |10 | 9 | 8 | 7 | 6 | 5 | 4 | 3 | 2 | 1 | 0 | |___|___|___|___|___|___|___|___|___|___|___|___|___|___|___|___|

Code

Flag

Flag's Purpose

Undefined; indeterminate value

Not used

Overflow flag

Indicates arithmetic overflow

Direction flag

Controls left/right direction in repeated operations

Interrupt enable flag

Enables interrupts (when set)

Trap flag

Generates a trap at end of each instruction (for trace)

Sign flag

Indicates negative result or comparison

if set Z

Zero flag

Indicates zero result or equal comparison

Auxiliary flag

Need adjustment in BCD (Binary Coded Decimal) operations

Parity flag

Indicates even number of bits set

Carry flag

Indicates arithmetic carry out bit

ptr_dword ptr_dword returns the internal address of StringVar, or creates the string ("the char pointer") StringVar based on the supplied address. ptr_dword(StringVar, Seg, Off)

/* (o,i,i), (i,o,o) */

When StringVar is bound, ptr_dword returns the internal segment and offset for the string. When Seg and Off are bound, ptr_dword binds StringVar to the string stored at that location. On 32-bit platforms the segment is ignored. ptr_dword has to a considerable extent been superseded by the cast function. A string in Visual Prolog is a series of ASCII values terminated by a zero value. You can use the low-level routines in this chapter on abnormal strings (those that contain several zero bytes). However, you can't write abnormal strings out or assert them in the database. membyte, memword, memdword Visual Prolog provides three predicates for looking at (peeking) and modifying (poking) specific elements in memory. membyte, memword and memdword access byte, word and dword sized locations respectively. All predicates have two formats: membyte(Segment, Offset, Byte) memword(Segment, Offset, Word) memdword(Segment, Offset, DWord)

/* (i,i,i), (i,i,o) */ /* (i,i,i), (i,i,o) */ /* (i,i,i), (i,i,o) */

and membyte(StringVar, Byte) memword(StringVar, Word) memdword(StringVar, DWord)

Chapter 170 Advanced topics

/* (i,i), (i,o) */ /* (i,i), (i,o) */ /* (i,i), (i,o) */

389

The Segment is an ushort, the Offset is an unsigned, and Byte, Word and DWord are byte, word and dword respectively. Many of the bios calls require pointers to be passed as Segment:Offset pairs. membyte and memword also require pointers in this format. In realmode DOS, Memory locations are calculated as ((Segment ) 16) + Offset). The mem* predicates have to a large extent been superseded by the get*entry and set*entry predicates for the binary datatype. port_byte/2 The port_byte predicate allows you to read or write a byte to a specific I/O port. The DOS format for port_byte is port_byte(PortAddress, Byte)

/* (i,i), (i,o) */

where PortAddress and Byte are defined as unsigneds. If you don't know what to use port_byte for, don't worry and don't think about using it. It's intended for access to (custom) hardware using ports for I/O.

Summary These are the major points covered in this chapter: 1. Visual Prolog includes several predicates that a. give access to the OS b. perform bit-level logical and shifting operations c. provide low-level support for manipulating the BIOS, memory, and other hardware elements 2. These are the predicates giving access to the OS: a. system (execute external program) b. time (get or set the system clock) c. date (get or set the internal calendar) d. envsymbol (look up an environment variable) e. comline (read the command-line arguments) f. syspath (return start-up directory and name of .EXE file) g. osversion (returns operating system version number)

h. diskspace (returns disk space available) 3. These are the predicates that perform bit-level operations: a. bitor (bit-wise OR) b. bitand (bit-wise AND) c. bitnot (bit-wise NOT) d. bitxor (bit-wise XOR) e. bitleft (bit-wise LEFT SHIFT) f. bitright (bit-wise RIGHT SHIFT) 4. These are the predicates that provide low-level support for various hardware elements: a. bios (accesses the PC's low-level BIOS routines) b. ptr_dword (returns the internal address of its argument or places the argument at a specified memory location) c. membyte (peeks or pokes a one-byte value) d. memword (peeks or pokes a two-byte value) e. memdword (peeks or pokes a four-byte value) f. port_byte (reads or writes a byte to a specific I/O port)

CHAPTER

260

Example Prolog Programs 261In this final tutorial chapter, we present some small example programs intended to stimulate your own ideas and to further illustrate the topics covered in the earlier tutorial chapters. Nearly all of the examples offer plenty of room for expansion; your own ideas can grow into full-blown programs using one of these programs as a starting point.

Chapter 170 Advanced topics

391

Building a Small Expert System In this first example, we show you how to construct a small expert system expert system, sample that figures out which of seven animals (if any) the system's user has in mind. The expert system will figure out the animal by asking questions then making deductions from the replies given. This example demonstrates backtracking--using facts--and how to use not effectively. A typical user dialogue with this expert system might be: has it hair? yes does it eat meat? yes has it a fawn color? yes has it dark spots? yes

Your animal may be a cheetah! Visual Prolog's ability to check facts and rules will provide your program with the reasoning capabilities germane to an expert system. The first step is to provide the knowledge with which the system can reason; this is known as the inference engine and is shown in ch262e01.pro. /* Program ch263e01.pro */ DATABASE xpositive(symbol,symbol) xnegative(symbol,symbol) PREDICATES nondeterm animal_is(symbol) nondeterm it_is(symbol) ask(symbol,symbol,symbol) remember(symbol,symbol,symbol) positive(symbol,symbol) negative(symbol,symbol) clear_facts run

CLAUSES animal_is(cheetah):it_is(mammal), it_is(carnivore), positive(has,tawny_color), positive(has,dark_spots). animal_is(tiger):it_is(mammal), it_is(carnivore), positive(has, tawny_color), positive(has, black_stripes). animal_is(giraffe):it_is(ungulate), positive(has,long_neck), positive(has,long_legs), positive(has, dark_spots). animal_is(zebra):it_is(ungulate), positive(has,black_stripes). animal_is(ostrich):it_is(bird), negative(does,fly), positive(has,long_neck), positive(has,long_legs), positive(has, black_and_white_color). animal_is(penguin):it_is(bird), negative(does,fly), positive(does,swim), positive(has,black_and_white_color). animal_is(albatross):it_is(bird),positive(does,fly_well). it_is(mammal):positive(has,hair). it_is(mammal):positive(does,give_milk). it_is(bird):positive(has,feathers). it_is(bird):positive(does,fly), positive(does,lay_eggs).

Chapter 170 Advanced topics

393

it_is(carnivore):positive(does,eat_meat). it_is(carnivore):positive(has,pointed_teeth), positive(has, claws), positive(has,forward_eyes). it_is(ungulate):it_is(mammal), positive(has,hooves). it_is(ungulate):it_is(mammal), positive(does,chew_cud). positive(X,Y):xpositive(X,Y),!. positive(X,Y):not(xnegative(X,Y)), ask(X,Y,yes). negative(X,Y):xnegative(X,Y),!. negative(X,Y):not(xpositive(X,Y)), ask(X,Y,no). ask(X,Y,yes):!, write(X," it ",Y,'\n'), readln(Reply),nl, frontchar(Reply,'y',_), remember(X,Y,yes). ask(X,Y,no):!, write(X," it ",Y,'\n'), readln(Reply),nl, frontchar(Reply,'n',_), remember(X,Y,no). remember(X,Y,yes):assertz(xpositive(X,Y)). remember(X,Y,no):assertz(xnegative(X,Y)). clear_facts:write("\n\nPlease press the space bar to exit\n"), retractall(_,dbasedom),readchar(_).

run:animal_is(X),!, write("\nYour animal may be a (an) ",X), nl,nl,clear_facts. run :write("\nUnable to determine what"), write("your animal is.\n\n"), clear_facts. GOAL run.

Each animal is described by a number of attributes that it has (or has not). Those questions that the user is to reply to are the positive(X,Y) and negative(X,Y) ones. The system, therefore, might ask something like this: Does it have hair?

Having received a reply to such a question, you want to be able to add the answer to the database, so the system will be able to use the previously gathered information when reasoning.* For simplicity, this example program will only consider positive and negative replies, so it uses a database containing two predicates: DATABASE xpositive(symbol, symbol) xnegative(symbol, symbol)

The fact that the animal doesn't have hair is represented by xnegative(has,hair).

The rules of positive and negative then check to see if the answer is already known, before asking the user. askable positive(X,Y) :xpositive(X,Y), !. positive(X,Y) :not(xnegative(X,Y)), ask(X,Y,yes). negative(X,Y) :xnegative(X,Y), !. negative(X,Y) :not(xpositive(X,Y)), ask(X,Y,no).

Chapter 170 Advanced topics

395

Notice that the second rule for both positive and negative ensures that a contradiction won't arise before asking the user. The ask predicate asks the questions and organizes the remembered replies. If a reply begins with the letter y, the system assumes the answer is Yes; if it begins with n, the answer is No. /* Asking Questions and Remembering Answers */ ask(X, Y, yes) :- !, write(X, " it ", Y, '\n'), readln(Reply), frontchar(Reply, 'y', _), remember(X, Y, yes). ask(X, Y, no) :- !, write(X, " it ", Y, '\n'), readln(Reply), frontchar(Reply, 'n', _), remember(X, Y, no). remember(X, Y, yes) :- assertz(xpositive(X, Y)). remember(X, Y, no) :- assertz(xnegative(X, Y)). /* Clearing Out Old Facts */ clear_facts :- write("\n\nPlease press the space bar to exit\n"), retractall(_,dbasedom), readchar(_).

For practice, type in the preceding inference engine and knowledge clauses. Add appropriate declarations to make a complete program, and then try out the result. The completed animal expert system is provided as ch264e01.pro. An example expert systems shell (GENI.PRO) is also provided with Visual Prolog in the PROGRAMS directory; this shell is based on the same techniques introduced in this example, with the added feature that it allows you to dynamically change the rules.

Prototyping: A Simple Routing Problem Suppose you want to construct a computer system to help decide the best route between two U.S. cities. You could first use Visual Prolog to build a miniature version of the system (see 2), since it will then become easier to investigate and explore different ways of solving the problems involved. You will use the final system to investigate questions such as: Is there a direct road from one particular town to another?

Which towns are situated less than ten miles from a particular town? The following program is a classic example of using backtracking and recursion to solve route planning. /* Program ch265e02.pro */ DOMAINS town = symbol distance = integer PREDICATES nondeterm road(town,town,distance) nondeterm route(town,town,distance) CLAUSES road(tampa,houston,200). road(gordon,tampa,300). road(houston,gordon,100). road(houston,kansas_city,120). road(gordon,kansas_city,130). route(Town1,Town2,Distance):road(Town1,Town2,Distance). route(Town1,Town2,Distance):road(Town1,X,Dist1), route(X,Town2,Dist2), Distance=Dist1+Dist2, !.

Figure 266.1 shows a simplified map for the prototype. Kansas City

Houston Tampa

Gordon

Figure 267.2: Prototype Map2683 Chapter 170 Advanced topics

397

Each clause for the road predicate is a fact that describes a road of a certain length (in miles) that goes from one town to another. The route clauses indicate that it is possible to make a route from one town to another over several stretches of road. Following the route, the driver travels a distance given by the third parameter, distance. The route predicate is defined recursively; a route can simply consist of one single stretch of road, as in the first clause. In this case, the total distance is merely the length of the road. You can also construct a route from Town1 to Town2 by driving first from Town1 to X, then following some other route from X to Town2. The total distance is the sum of the distance from Town1 to X and the distance from X to Town2, as shown in the second clause for route. Try the program with the goal: route(tampa, kansas_city, X).

Can the program handle all possible combinations of starting point and destination? If not, can you modify the program to avoid any omissions? The next example will give you ideas about how to get this routing program to make a list of towns visited enroute. Making such a list prevents Visual Prolog from choosing a route that involves visiting the same town twice, thereby avoiding going around in circles, and ensures that the program doesn't go into an infinite loop. When you've solved problems of this type, you can enlarge the program by adding more cities and roads.

Adventures in a Dangerous Cave You're an adventurer, and you've heard that there is a vast gold treasure hidden inside a cave. Many people before you have tried to find it, but to no avail. The cave is a labyrinth of galleries connecting different rooms in which there are dangerous beings, like monsters and robbers. In your favor is the fact that the treasure is all in one room. Which route should you follow to get to the treasure and escape unhurt with it? Consider the following map of the cave:

entry

hell mermaid

robbers

fountain

food

monsters

gold treasure

exit

Figure 269.4: The Labyrinth of Galleries270 You can construct a Visual Prolog representation of the map to help you find a safe route. Each gallery is described by a fact. The predicates go and route give rules. Give the program the goal go(entry, exit).

The answer will consist of a list of the rooms you should visit to capture the treasure and return safely with it. An important design feature of this program is that the rooms already visited are collected in a catalog. This happens thanks to the route predicate, which is defined recursively. If you're standing in the exit room, the third parameter in the route predicate will be a list of the rooms you've already visited. If the gold_treasure room is a member of this list, you'll have achieved your aim. Otherwise, the list of rooms visited is enlarged with Nextroom, provided Nextroom is not one of the dangerous rooms and has not been visited before. /* Program ch271e03.pro */ DOMAINS room = symbol roomlist = room*

Chapter 170 Advanced topics

399

PREDICATES nondeterm gallery(room,room) % There is a gallery between two rooms % Necessary because it does not matter % which direction you go along a gallery nondeterm neighborroom(room,room) avoid(roomlist) nondeterm go(room,room) nondeterm route(room,room,roomlist) % This is the route to be followed. % roomlist consists of a list of rooms already visited. nondeterm member(room,roomlist) CLAUSES gallery(entry,monsters). gallery(fountain,hell). gallery(exit,gold_treasure). gallery(robbers,gold_treasure). gallery(food,gold_treasure). gallery(monsters,gold_treasure).

gallery(entry,fountain). gallery(fountain,food). gallery(fountain,mermaid). gallery(fountain,robbers). gallery(mermaid,exit). gallery(gold_treasure,exit).

neighborroom(X,Y):-gallery(X,Y). neighborroom(X,Y):-gallery(Y,X). avoid([monsters,robbers]). go(Here,There):-route(Here,There,[Here]). go(_,_). route(Room,Room,VisitedRooms):member(gold_treasure,VisitedRooms), write(VisitedRooms),nl. route(Room,Way_out,VisitedRooms):neighborroom(Room,Nextroom), avoid(DangerousRooms), not(member(NextRoom,DangerousRooms)), not(member(NextRoom,VisitedRooms)), route(NextRoom,Way_out,[NextRoom|VisitedRooms]). member(X,[X|_]). member(X,[_|H]):-member (X,H).

After verifying that the program does find a solution to the goal go(entry, exit).

you might want to try adding some more galleries, for example, gallery(mermaid, gold_treasure).

Or you can add some additional nasty things to avoid. Even though--once you've made these additions--there is more than one possible solution to the problem, your program will only come up with one solution. To obtain all the solutions, you must make Visual Prolog backtrack as soon as it has found one solution. You can do this by adding the fail predicate to the first rule for route: route(Room, Room, VisitedRooms) :member(gold_treasure, VisitedRooms), write(VisitedRooms), nl, fail.

To get a neater output, you could use a list-writing predicate, write_a_list, to write the list of names without the containing square brackets ([ and ]) or the separating commas. However, the rooms you've visited are collected in the VisitedRooms list in reverse order (exit first and entry last). Therefore, you need to reverse the list or make the list-writing predicate write the list in reverse.

Hardware Simulation Every logical circuit can be described with a Visual Prolog predicate, where the predicate indicates the relationship between the signals on the input and output terminals of the circuit. The fundamental circuits are described by giving a table of corresponding truth values (see the and_, or_, and not_ predicates in Program 4). Fundamental circuits can be described by indicating the relationships between the internal connections, as well as the terminals. To see how this works, construct an exclusive OR circuit from AND, OR, and NOT circuits, and then check its operation with a Visual Prolog program. The circuit is shown in Figure 272.5. Input1

NOT

N1 AND

N4 OR

Input2

NOT

AND

Output

Figure 273.6: Fundamental XOR Circuit2747 Chapter 170 Advanced topics

401

In Program 4, this network is described by the xor predicate. /* Program ch275e04.pro */ DOMAINS d = integer PREDICATES nondeterm not_(D,D) and_(D,D,D) or_(D,D,D) nondeterm xor(D,D,D) CLAUSES not_(1,0). and_(0,0,0). and_(1,0,0). or_(0,0,0). or_(1,0,1).

not_(0,1). and_(0,1,0). and_(1,1,1). or_(0,1,1). or_(1,1,1).

xor(Input1,Input2,Output):not_(Input1,N1), not_(Input2,N2), and_(Input1,N2,N3), and_(Input2,N1,N4), or_(N3,N4,Output).

Given the following goal in interactive mode: xor(Input1, Input2, Output).

this program yields the following result: Input1=1, Input2=1, Input1=1, Input2=0, Input1=0, Input2=1, Input1=0, Input2=0, 4 Solutions

Output=0 Output=1 Output=1 Output=0

Interpreting this result as a truth table, you can see that the circuit does indeed perform as expected.

Towers of Hanoi The solution to the Towers of Hanoi puzzle is a classic example of recursion. The ancient puzzle of the Towers Of Hanoi consists of a number of wooden disks

mounted on three poles, which are in turn attached to a baseboard. The disks each have different diameters and a hole in the middle large enough for the poles to pass through. In the beginning, all the disks are on the left pole as shown in Figure 276.8.

Figure 277.9: The Towers of Hanoi27810 The object of the puzzle is to move all the disks over to the right pole, one at a time, so that they end up in the original order on that pole. You can use the middle pole as a temporary resting place for disks, but at no time is a larger disk to be on top of a smaller one. It's easy to solve the Towers of Hanoi with two or three disks, but the process becomes more difficult with four or more disks. A simple strategy for solving the puzzle is as follows: You can move a single disk directly. You can move N disks in three general steps: Move N-1 disks to the middle pole. Move the last (Nth) disk directly over to the right pole. Move the N-1 disks from the middle pole to the right pole. The Visual Prolog program to solve the Towers Of Hanoi puzzle uses three predicates: hanoi, with one parameter that indicates the total number of disks you are working with. move, which describes the moving of N disks from one pole to another-using the remaining pole as a temporary resting place for disks. inform, which displays what has happened to a particular disk. /* Program ch279e05.pro */ DOMAINS loc =right;middle;left

Chapter 170 Advanced topics

403

PREDICATES hanoi(integer) move(integer,loc,loc,loc) inform(loc,loc) CLAUSES hanoi(N):move(N,left,middle,right). move(1,A,_,C):inform(A,C),!. move(N,A,B,C):N1=N-1, move(N1,A,C,B), inform(A,C),move(N1,B,A,C). inform(Loc1, Loc2):-nl, write("Move a disk from ", Loc1, " to ", Loc2).

To solve the Towers of Hanoi with three disks, give the goal hanoi(3). The output is: Move Move Move Move Move Move Move

a a a a a a a

disk disk disk disk disk disk disk

from from from from from from from

left to right left to middle right to middle left to right middle to left middle to right left to right

Dividing Words into Syllables Using a very simple algorithm that involves looking at the sequence of vowels and consonants a word contains, a computer program can decide how to divide words into syllables. For instance, consider the two sequences: 1) vowel consonant vowel In this case, the word is divided after the first vowel. For example, this rule can be applied to the following words: ruler

> ru-ler

prolog > pro-log 2) vowel consonant consonant vowel

In this case, the word is divided between the two consonants. For example, number > num-ber panter

> pan-ter

console > con-sole These two rules work well for most words but fail with words like handbook and hungry, which conform to neither pattern. To divide such words, your program would have to use a library containing all words. Write a Visual Prolog program to divide a word into syllables. The program will first ask for a word to be typed in, and then attempt to split it into syllables using the two rules just given. As we've mentioned, this will not always produce correct results. First, the program should split the word up into a list of characters. You therefore need the following domain declarations: DOMAINS letter = symbol word= letter*

You must have a predicate that determines whether the letter is a vowel or a consonant. However, the two rules given can also work with the vocals (the usual vowels--a, e, i, o, and u--plus the letter y). The letter y sounds like (and is considered to be) a vowel in many words, for example, hyphen, pity, myrrh, syzygy, and martyr. To account for the vocals, you have the clauses vocal(a). vocal(o).

vocal(e). vocal(u).

vocal(i). vocal(y).

for the predicate vocal. A consonant is defined as a letter that is not a vocal: consonant(L) :- not(vocal(L)).

You also need two more predicates. First, you need the append predicate. append(word, word, word)

Second, you need a predicate to convert a string to a list of the characters in that string: string_word(string, word)

This predicate will use the standard predicate frontstr (described in chapter 280), as well as the standard predicates free and bound (where free(X) succeeds if X is Chapter 170 Advanced topics

405

a free variable at the time of calling, and bound(Y) succeeds if Y is bound), to control which clause to activate, dependent on the flow-pattern. Now you're ready to attack the main problem: defining the predicate divide that separates a word into syllables. divide has four parameters and is defined recursively. The first and second parameters contain, respectively, the Start and the Remainder of a given word during the recursion. The last two arguments return, respectively, the first and the last part of the word after the word has been divided into syllables. As a example, the first rule for divide is: divide(Start, [T1, T2, T3|Rest], D, [T2, T3|Rest]) :vocal(T1), consonant(T2), vocal(T3), append(Start, [T1], D).

where Start is a list of the first group of characters in the word to be divided. The next three characters in the word are represented by T1, T2, and T3, while Rest represents the remaining characters in the word. In list D, the characters T2 and T3, and the list Rest represent the complete sequence of letters in the word. The word is divided into syllables at the end of those letters contained in D. This rule can be satisfied by the call: divide([p, r], [o, l, o, g], P1, P2)

To see how, insert the appropriate letters into the clause: divide([p, r], [o, l, o|[g]], [p, r, o], [l, o | [g]]) :vocal(o), consonant(l), vocal(o), append([p, r], [o], [p, r, o]).

The append predicate concatenates the first vocal to the start of the word. P1 becomes bound to [p, r, o], and P2 is bound to [l, o, g]. The second rule for divide is shown in the complete program, 6. /* Program ch281e06.pro */ DOMAINS letter = char word_ = letter*

PREDICATES nondeterm divide(word_,word_,word_,word_) vocal(letter) consonant(letter) nondeterm string_word(string,word_) append(word_,word_,word_) nondeterm repeat CLAUSES divide(Start,[T1,T2,T3|Rest],D1,[T2,T3|Rest]):vocal(T1),consonant(T2),vocal(T3), append(Start,[T1],D1). divide(Start,[T1,T2,T3,T4|Rest],D1,[T3,T4|Rest]):vocal(T1),consonant(T2),consonant(T3),vocal(T4), append(Start,[T1,T2],D1). divide(Start,[T1|Rest],D1,D2):append(Start,[T1],S), divide(S,Rest,D1,D2). vocal('a'). vocal('e'). vocal('i'). vocal('o'). vocal('u'). vocal('y'). consonant(B):not(vocal(B)),B <= 'z','a' <= B. string_word("",[]):-!. string_word(Str,[H|T]):bound(Str),frontchar(Str,H,S),string_word(S,T). string_word(Str,[H|T]):free(Str),bound(H),string_word(S,T),frontchar(Str,H,S). append([],L,L):-!. append([X|L1],L2,[X|L3]):append(L1,L2,L3). repeat. repeat:-repeat. GOAL repeat, write("Write a multi-syllable word: "), readln(S),nl, string_word(S,Word), divide([],Word,Part1,Part2), string_word(Syllable1,Part1), string_word(Syllable2,Part2), write("Division: ",Syllable1,"-",Syllable2),nl, fail.

Chapter 170 Advanced topics

407

The N Queens Problem In the N Queens problem, the object is to place N queens on a chessboard in such a way that no two queens can take each other. Accordingly, no two queens can be placed on the same row, column, or diagonal. To solve the problem, you'll number the rows and columns of the chessboard from 1 to N. To number the diagonals, you divide them into two types, so that a diagonal is uniquely specified by a type and a number calculated from its row and column numbers: Diagonal = N + Column - Row (Type 1) Diagonal = Row + Column - 1 (Type 2)

When you view the chessboard with row 1 at the top and column 1 on the left side, Type 1 diagonals resemble the backslash (\) character in shape, and Type 2 diagonals resemble the shape of slash (/). Figure 282.11 shows the numbering of Type 2 diagonals on a 4x4 board.

Figure 283.12: The N Queens Chessboard28413 To solve the N Queens Problem with a Visual Prolog program, you must record which rows, columns, and diagonals are unoccupied, and also make a note of where the queens are placed. A queen's position is described with a row number and a column number as in the domain declaration: queen = q(integer, integer)

This declaration represents the position of one queen. To describe more positions, you can use a list: queens = queen*

Likewise, you need several numerical lists indicating the rows, columns, and diagonals not occupied by a queen. These lists are described by: freelist = integer*

You will treat the chessboard as a single object with the following domain declaration: board = board(queens, freelist, freelist, freelist, freelist)

The four freelists represent the free rows, columns, and diagonals of Type 1 and Type 2, respectively. To see how this is going to work, let board represent a 4 )4 chessboard in two situations: (1) without queens, and (2) with one queen at the top left corner. 1. board without queens board([], [1,2,3,4], [1,2,3,4], [1,2,3,4,5,6,7], [1,2,3,4,5,6,7])

2. board with one queen board([q(1,1)], [2,3,4], [2,3,4], [1,2,3,5,6,7], [2,3,4,5,6,7])

You can now solve the problem by describing the relationship between an empty board and a board with N queens. You define the predicate placeN(integer, board, board)

with the two clauses following. Queens are placed one at a time until every row and column is occupied. You can see this in the first clause, where the two lists of freerows and freecols are empty: placeN(_, board(D, [], [], X, Y), board(D, [], [], X, Y)) :- !. placeN(N, Board1, Result) :place_a_queen(N, Board1, Board2), placeN(N, Board2, Result).

In the second clause, the predicate place_a_queen gives the connection between Board1 and Board2. (Board2 has one more queen than Board1). Use this predicate declaration: place_a_queen(integer, board, board)

Chapter 170 Advanced topics

409

The core of the N Queens Problem lies in the description of how to add extra queens until they have all been successfully placed, starting with an empty board. To solve this problem, add the new queen to the list of those already placed: [q(R, C)|Queens]

Among the remaining free rows, Rows, you need to find a row R where you can place the next queen. At the same time, you must remove R from the list of free rows, resulting in a new list of free rows, NewR. This is formulated as: findandremove(R, Rows, NewR)

Correspondingly, you must find and remove a vacant column C. From R and C, you can calculate the numbers of the occupied diagonals. Then you can determine if D1 and D2 are among the vacant diagonals. This is the place_a_queen clause: place_a_queen(N, board(Queens, Rows, Columns, Diag1, Diag2), board([q(R, C)|Queens], NewR, NewS, NewD1, NewD2)) :findandremove(R, Rows, NewR), findandremove(C, Columns, NewC), D1=N+S-R, findandremove(D1, Diag1, NewD1), D2=R+S-1, findandremove(D2, Diag2, NewD2).

Program 7 is the complete program. It contains a number of smaller additions to define nqueens, so you only need to give a goal like: nqueens(5)

to obtain a possible solution (in this case, for placing five queens on a 5 )5 board). /* Program ch285e07.pro */ DOMAINS queen queens freelist board

= = = =

q(integer, integer) queen* integer* board(queens, freelist, freelist, freelist, freelist)

PREDICATES nondeterm placeN(integer, board, board) nondeterm place_a_queen(integer, board, board) nondeterm nqueens(integer) nondeterm makelist(integer, freelist) nondeterm findandremove(integer, freelist, freelist) nextrow(integer, freelist, freelist)

CLAUSES nqueens(N):makelist(N,L),Diagonal=N*2-1,makelist(Diagonal,LL), placeN(N,board([],L,L,LL,LL),Final), write(Final). placeN(_,board(D,[],[],D1,D2),board(D,[],[],D1,D2)):-!. placeN(N,Board1,Result):place_a_queen(N,Board1,Board2), placeN(N,Board2,Result). place_a_queen(N,board(Queens,Rows,Columns,Diag1,Diag2), board([q(R,C)|Queens],NewR,NewC,NewD1,NewD2)):nextrow(R,Rows,NewR), findandremove(C,Columns,NewC), D1=N+C-R,findandremove(D1,Diag1,NewD1), D2=R+C-1,findandremove(D2,Diag2,NewD2). findandremove(X,[X|Rest],Rest). findandremove(X,[Y|Rest],[Y|Tail]):findandremove(X,Rest,Tail). makelist(1,[1]). makelist(N,[N|Rest]) :N1=N-1,makelist(N1,Rest). nextrow(Row,[Row|Rest],Rest).

PA RT

Programmer’s Guide

Chapter 170 Advanced topics

411

CHAPTER

286

Elements of the Language 287In this chapter, we summarize the elements of the Visual Prolog compiler and language. We discuss some fundamental elements of the language: names, program sections, compiler directives, and memory management. After this we give an introduction to handling modules in Visual Prolog, and how a program can be split up into several modules, which you can then compile separately and link together. We've written this chapter for programmers who've already worked some with Visual Prolog. To get the most benefit out of this chapter, you should be familiar with the material in the first chapters of the Visual Prolog Language.

Names In Prolog, names are used to denote symbolic constants, domains, predicates, and variables. A name consists of a letter, or an underscore character, followed by any combination of zero or more letters, digits, and underscores. Two important restrictions are imposed on names: Names of symbolic constants must start with a lower-case letter. Names of variables must start with an upper-case letter or an underscore. Except for these restrictions, you can use upper-case and lower-case letters in your programs as you please. For instance, you could make a name more readable by using mixed upper-case and lower-case, as in the variable MyLongestVariableNameSoFar

or by using underscores, as in pair_who_might_make_a_happy_couple(henry_viii, ann_boleyn)

The Visual Prolog compiler does not make a distinction between upper and lower case letters, except for the first letter. This means that the two variables: SourceCode

and

SOURCECODE

are the same. Keywords The following are reserved words; you must not employ them as user-defined names: and clauses constants database

domains elsedef enddef global

goal if ifdef ifndef

include or predicates

Specially-Handled Predicates The following list of predicates are handled specially by the compiler. assert asserta assertz bound chain_inserta chain_insertafter chain_insertz

chain_terms consult db_btrees db_chains fail findall format

free msgrecv msgsend not readterm ref_term retract

retractall save term_bin term_replace term_str trap write writef

Program Sections A Visual Prolog program consists of several program sections. Each program section is identified by a keyword, as shown in this table. Table 288.1: Contents of Program Sections289 Section

Contents

compiler options

Options are given at the top of a program.

constants section

Zero or more constants.

domains section

Zero or more domain declarations.

facts section

Zero or more database predicates.

Class section Zero or more class declarations Chapter 170 Advanced topics

413

Implement section

Zero or more implementations of classpredicates

predicates section

Zero or more predicate declarations.

goal section

Zero or one goal.

clauses section

Zero or more clauses.

To generate an executable stand-alone application, your program must contain a goal. Usually, a program requires at least a predicates and a clauses section. For most programs, a domains section is needed to declare lists, compound structures and your own names for the basic domains. For modular programming, you can prefix the keywords domains, predicates and database with the word global, indicating that the subsequent declarations affect several program modules globally. (Modular programming is discussed on page 438). A program can contain several domains, predicates, database or clauses sections, provided you observe the following restrictions: Constants, domains and predicates should be defined before you use them. However, within the domains section you can refer to domains that are declared at a later point. Only one goal must be met during compilation. However, the goal can appear anywhere. All clauses that describe the same predicate must occur in sequence (one after the other). All global declarations must come before any local declarations. The database sections can be named, but a given name can only appear once. Because the default name is dbasedom, there can only be one unnamed database section.

The Domains Section A domains section contains domain declarations. Five generic formats are used:

name = d mylist = elementDom* my_CompDom = f1(d11,d12,...,d1n); f2(d21,d22,...,d2n); ... predefdom = name1;name2;...;nameN

/*standard domain*/ /*list domain*/ /*compound object domain*/

/*eg db_selector and file domains*/ pclass = determspec args flow langspec /* predicate class declaration */

Standard Domains name = d

This declaration specifies a domain, name, which consists of elements from a standard domain type d; the domain type d must be char, real, ref, string, symbol, or one of the integral domains. This declaration is used for objects that are syntactically alike but semantically different. For instance, NoOfApples and HeightInFeet could both be represented as integers, and consequently be mistaken for one another. You can avoid this by declaring two different domains of integer type, like this: apples, height = integer

Declaring different domains in this way allows Visual Prolog to perform domain checks to ensure, for example, that apples and height are never inadvertently mixed. However both domains can interchangeably be mixed with integers, and you can use the equal sign to convert between NoOfApples and HeightInFeet. List Domains mylist = elementDom*

This is a convenient notation for declaring a list domain. mylist is a domain consisting of lists of elements, from the domain elementDom. The domain elementDom can be either a user-defined domain, or one of the standard types of domain. You read the asteriskasterisk as "list". For example, this domain declaration: numberlist = integer*

declares a domain for lists of integers, such as [1, -5, 2, -6].

Chapter 170 Advanced topics

415

Compound Object Domains myCompDom=f1(d11, .., d1N); f2(d21, d22 ..); ...

To declare a domain that consists of compound objects, you state a functor and the domains for all the subcomponents. For example, you could declare a domain of owners made up of elements like this: owns(john, book(wuthering_heights, bronte))

with this declaration: owners = owns(symbol, book) book = book(symbol,symbol)]

where owns is the functor of the compound object, and symbol and book are domains of the subcomponents. The right side of this type of domain declaration can define several alternatives, separated by a semicolon (;). Each alternative must contain a unique functor and a description of the domains for the actual subcomponents of the functor. For example, the following domain declaration could be used to say, "For some predicates a key is either up, down, left, right or a character value." key = up; down; left; right; char(char)

There is a possibility to include a comment after the domain, for instance person= p(string name, integer age).

File Domain file = name1;name2;...;nameN

A file domain must be defined when you need to refer to files (other than the predefined ones) by symbolic names. A program can have only one domain of this type, which must be called file. Symbolic file names are then given as alternatives for the file domain. For example, this declaration: file = sales ; salaries

introduces the two symbolic file names sales and salaries.

The following alternatives are predefined in the file domain: keyboard screen stderr

stdin stdout

Specially Handled Predefined Domains There are several predefined domains; some are handled specially, like the file domain and the db_selector domain. Here's a summary of these special predefined domains: Table 290.2: Specially Handled Predefined Domains291 dbasedom

generated domain for terms in the global database

bt_selector

returned binary tree selector

db_selector

user-defined external database selectors

place

in_memory; in_ems; in_file

accesmode

read; readwrite

denymode

denynone; denywrite; denyall

ref

domain for database reference numbers

file

symbolic file names

reg

reg(AX,BX,CX,DX,SI,DI,DS,ES) used with bios/4

bgi_ilist

list of integers used in the BGI predicates.

Shortening Domain Declarations As shown in the standard domain declaration name=d, the left side of a domain declaration (except for a file domain) can consist of a list of names, like this: mydom1, mydom2, ... , mydomN = ...

This feature allows you to declare several domains at the same time. firstname, lastname, address = string

Chapter 170 Advanced topics

417

Declaring Reference Domains A reference domain is one that can carry unbound variables as input arguments. To declare a reference domain, precede the right side of the domain declaration with the keyword reference. When you declare a compound domain as a reference domain, all its subdomains are automatically declared as reference domains. DOMAINS reflist refint term refsymb

= = = =

reference reference reference reference

refint* integer int(refint); symb(refsymb) symbol

Declaring Predicate Domains A predicate domain declares a group or class of predicates. In a subsequent predicate declaration you may then declare one or more predicates as belonging to such a group, and these may then be specified as arguments to other predicates. Those other predicates will hence be able to do a variable call. The declaration for a predicate domain is of the form: pdom = { determ | nondeterm } [ domain ] arglist [ - flowpattern ] [ language ]

(curly braces indicate "choose one", square brackets indicate optional items) where domain is the return domain, if you're declaring a function arglist is of the form ( [ domain [ , domain ]* ] )

flowpattern is of the form ( flow ) where flow is { i | o | functor flowpattern | listflow } where listflow is '[' flow [ , flow ]* [ '|' { i | o | listflow } ] ']'

language is of the form language { prolog | c | pascal | asm | stdcall | syscall }

defaults to pascal if omitted, but this should not be relied upon if a particular convention is desired. The flowpattern specifies how each argument is to be used. It should be the letter i for an argument with input flow, the letter o for one with output flow, a functor and flowpattern for a compound term (e.g. (i,o,myfunc(i,i),o) ), or a listflow (e.g. [i,myfunc(i,o),o] or [o,o|i] ). You can have no more than one flowpattern declaration for a predicate pointer domain, and it must be given unless the argument list is empty or all arguments have input flow. Hence, the declaration for a group of deterministic predicates taking an integer as argument and returning an integer, would be: DOMAINS list_process = determ integer (integer) - (i)

This group, or class, is now known as list_process.

The Predicates Section In Visual Prolog, the sections introduced by the keyword predicates contain predicate declarations. You declare a predicate by its name and the domains of its arguments, like this:) PREDICATES predname(domain1, domain2,...,domainN)

In this example, predname represents the new predicate name and domain1, ..., domainN stand for user-defined domains or pre-defined domains. Multiple declarations for one predicate are also allowed. As an example, you could declare that the predicate member works both on numbers and names by giving the following declarations: PREDICATES member(name, namelist) member(number, numberlist)

In this example, the arguments name, namelist, number, and numberlist are userdefined domains. You can declare a predicate with several different arities. hanoi hanoi(integer)

% chooses 10 slices as default % moves N slices

Chapter 170 Advanced topics

419

If you give more than one declaration for the same name, these declarations must come right after each other. You can declare predicates as being deterministic by preceding the predicate declaration with determ, or you can declare a predicate as being non-deterministic by preceding the declaration by nondeterm. If you declare a predicate to be deterministic, the compiler will issue a warning if it finds any non-deterministic clauses for the predicate. This functions exactly as if you had used the general compiler directive check_determ. On the other hand, when you declare a predicate as non-deterministic, the compiler will not complain when you add check_determ for checking the other predicates. nondeterm repeat /*repeat is non-deterministic by design*/ determ menuact(Integer,String) /*menuact is deterministic*/

Note that predicates also can be preceded with the following keywords: Multi: The keyword multi defines non-deterministic predicates that can backtrack and generate multiple solutions. Predicates declared with the keyword multi always succeed (never fail) and, therefore, always have at least one solution. Failure: A predicate declared with the keyword failure should always fail. Therefore, such a predicate does not produce a solution. In Visual Prolog failure predicates always enforce a program to backtrack to the nearest backtracking point. Erroneous: A predicate declared with the keyword erroneous should never fail and should not produce solution. Typical used for errorhandling purposes. Predicate Classes If you have declared a predicate domain in the domain section, you may declare one or more predicates as belonging to that domain. The syntax for this is. PREDICATES pred1: p_domain pred2: p_domain ...

where pred1, pred2 etc. are the predicate names and p_domain is the predicate domain declared in the domain section. Functions By prefixing a predicate declaration with a domain name, you declare a function. The return value is taken from the last argument in the final clause executed, and

this argument must not be present in the predicate declaration. A function returning the cube of its argument would hence be declared as: PREDICATES integer cube(integer)

And the clause for this function would be: CLAUSES cube(In,Out):- Out = In*In*In.

A function can return any domain.

The Facts/Database Section A facts database section declares predicates just as the predicates section does. However, the clauses for database predicates can only consist of plain facts, they cannot have an associated body. These facts can be inserted at run time by assert, asserta, assertz, or consult, and you can remove them again with retract or retractall. You can have a number of database sections in your program; some of them can be global and some local. You should name your program's database sections, and each name must be unique within the module. If you don't give a name for a database section, the compiler will give it the default name dbasedom. Only one unnamed database is possible. You can precede a database predicate with determ if you know that there will be only one fact for that predicate. This enables the compiler to produce better code, and you will not get nondeterministic warnings for calling such a predicate. This is useful for flags, counters, and other things that are essentially global variables. When a database section is declared, the compiler will internally declare a corresponding domain with the same name as the name of the database section; this allows predicates to handle facts as terms. The form of a facts section is: [global] DATABASE[ - <databasename> ] [determ|single|nocopy] dbpred1(....) dbpred2(.....)

An example is: FACTS - tables part(name,cost) salesperson(name,sex) PREDICATES write_table_element(tables)

Chapter 170 Advanced topics

421

CLAUSES write_table_element(part(Name,Cost)):writef("\nPart's Name= % Cost = %",Name,Cost). write_table_element(salesperson(Name,Sex)):writef("\nSalesperson's Name= % Sex = %",Name,Sex).

The Clauses Section A clause is either a fact or a rule corresponding to one of the declared predicates. In general, a clause consists of either 1) a fact or 2) a clause head followed first by a colon and hyphen (:-), then by a list of predicate calls separated by commas or semicolons. Both facts and rules must be terminated by a period (.). The fact: same_league(ucla, usc).

consists of a predicate name (same_league), and a bracketed list of arguments (ucla, usc). Simple Constants Simple constants belong to one of the following standard domains: char

A character (an 8-bit ASCII character enclosed between a pair of single quotation marks) belongs to the char domain. An ASCII character is indicated by the escape character (\) followed by the ASCII code for that character. \n, \t, \r produce a newline , a tab and a carriage return character, respectively. A backslash (\) followed by any other character produces that character ('\\' produces \ and '\'' produces ').

integral numbers

positive and negative numbers can be represented in the Visual Prologs integral number domains shown in the following table.

real

A real number belongs to the real domain and is a number in the range -)1e-307 to -)1e+308. real numbers. Real numbers are written with a sign, a mantissa, a decimal point, a fractional part, an e, a sign, and an

exponent, all without included spaces. For example, the real value -12345.6789 * 1014 can be written as -1.23456789e+18. The sign, fractional, and exponent parts are optional (though if you omit the fractional part, you must leave out the decimal point, too). Visual Prolog automatically converts integers to real numbers when necessary. string

A string (any sequence of characters between a pair of double quotation marks) belongs to the string domain. Strings can contain characters produced by an escape sequence (as mentioned under char); strings can be up to 64 K in length.

symbol

A symbolic constant (a name starting with a lowercase letter) belongs to the symbol domain type. Strings are accepted as symbols too, but symbols are kept in an internal table for quicker matching. The symbol table takes up some storage space, as well as the time required to make an entry in the table. However, if the same symbols are frequently compared, it's well worth the investment.

binary

A binary constant belongs to the binary domain. It is written as a comma-separated list of integral values, each less than or equal to 255, enclosed in square brackets prefixed with a dollar sign: $[1,0xff,'a'].

predicate pointer

A predicate pointer is the name of a predicate previously declared as belonging to a predicate pointer domain. It is written simply as the name of the predicate, with no argument list or brackets.

Table 292.3: Integral Standard Domains293 Domain short

Description and implementation A small, signed, quantity. All platforms

Chapter 170 Advanced topics

16 bits,2s comp

32768 .. 32767

423

ushort

A small, unsigned, quantity. All platforms

long

unsigned

32 bits,2s comp

-2147483648 .. 2147483647

A large, unsigned quantity All platforms

integer

0 .. 65535

A large signed quantity All platforms

ulong

16 bits

32 bits

0 .. 4294967295

A signed quantity, having the natural machine/platform architecture in question.

size

16bit platforms

16 bits,2s comp

-32768 .. 32767

32bit platforms

32 bits,2s comp

-2147483648 .. 2147483647

for

the

An unsigned quantity, having the natural size for the machine/platform architecture in question. 16bit platforms 32bit platforms

16 bits 32 bits

0 .. 65535 0 .. 4294967295

byte All platforms

³ 8 bits

All platforms

16 bits

All platforms

32 bits

0 .. 255

word 0 .. 65535

dword

An integral value may be preceded by syntax respectively.

0 .. 4294967295

or 0o, indicating hexadecimal and octal

Terms A term is, strictly speaking, any Prolog entity. In practice we tend to mean those (variable) entities holding data or non-compiled information, or compound terms (consisting of a functor and optional arguments).

Variables Variables are names starting with an upper-case letter or underscore or, to represent the anonymous variable, a single underscore character underscore character). The anonymous variable is used when the value of that variable is not of interest. A variable is said to be free when it is not yet associated with a term, and bound or instantiated when it is unified with a term. The Visual Prolog has a option so it can give a warning when it detects that a variable has been used only once in a clause. This warning will not be given if the variable starts with an underscore. Note that when a variable name starts with an underscore like _Win, it is still a normal variable, that unlike the anonymous variable can be used to pass values from one call to another. Compound Objects A compound object is a single object that consists of a collection of other objects (called subcomponents) and a describing name (the functor). The subcomponents are enclosed in parentheses and separated by commas. The functor is written just before the left parenthesis. For example, the following compound term consists of the functor author and three subcomponents: author(emily, bronte, 1818)

A compound object belongs to a user-defined domain. The domain declaration corresponding to the author compound object might look like this: DOMAINS author_dom = author(firstname, lastname, year_of_birth) firstname, lastname = symbol year_of_birth = integer

Functorless Compound Objects By prefixing a compound object declaration with the directive struct, you declare a functorless object. DOMAINS author_dom = struct author(firstname, lastname, year_of_birth)

The internal representation of such an object has no functor and there can be no alternatives in a functorless domain. Functorless terms can be used just like other terms in your source code, but their primary aim is to be directly compatible with C structs.

Chapter 170 Advanced topics

425

Lists--A Special Kind of Compound Object Lists are a common data structure in Prolog and is actually a form of compound object. Syntactically, it is written as a sequence of comma-separated arguments, enclosed in square brackets. A list of integers would appear as follows: [1, 2, 3, 9, -3, 2]

Such a list belongs to a user-defined domain, such as: DOMAINS ilist = integer*

If the elements in a list are of mixed types (for example, a list containing both characters and integers), you must state this in a corresponding domain declaration. For example, the following declarations DOMAINS element = c(char) ; i(integer) list = element*

would allow lists like this one: [i(12), i(34), i(-567), c('x'), c('y'), c('z'), i(987)]

Memory Alignment By prefixing a compound or list declaration with an alignment specification, you can override the default alignment. The syntax is: DOMAINS dom = align { byte | word | dword } domdecl

where domdecl is a normal domain declaration: DOMAINS element = align byte c(char) ; i(integer) list = align dword element*

This would make the internal representation for elements byte-aligned and list dword-aligned. If you want to override the default alignment for a functorless domain, the struct directive must precede the align directive. DOMAINS bbdom = struct align byte blm(char,integer)

The primary aim of overriding alignment is to make compound objects compatible with external code using a different alignment than the default for your platform. If several program share an external database or communicate over pipes, the domains involved must use the same alignment.

The Constants Section You can define and use constants in your Visual Prolog programs. A constant declaration section is indicated by the keyword constants, followed by the declarations themselves, using the following syntax: <Id> =

Each <definition> is terminated by a newline character, so there can be only one constant declaration per line. Constants declared in this way can then be referred to later in the program. Consider the following program fragment: CONSTANTS blue = 1 green = 2 red = 4 grayfill = [0xaa, 0x55, 0xaa, 0x55, 0xaa, 0x55, 0xaa, 0x55 ] language = english project_module = true

Before compiling your program, Visual Prolog will replace each constant with the actual string to which it corresponds. For instance: ... menu_colors(red,green,blue), my_fill_pattern(grayfill), text_convert(prolog, language), status(project_module), ...

will be handled by the compiler in exactly the same way as: ... menu_colors(4, 2, 1), my_fill_pattern([0xaa, 0x55, 0xaa, 0x55, 0xaa, 0x55, 0xaa, 0x55]), text_convert(prolog, english), status(true), ...

Chapter 170 Advanced topics

427

There are a few restrictions on the use of symbolic constants. The definition of a constant can't refer to itself. For example: list = [1, 2|list].

/* Is not allowed

will generate the error message Recursion in constant definition. The system does not distinguish between upper-case and lower-case in a constant declaration. Consequently, when a constant identifier is used in the clauses section of a program, the first letter must be lower-case to avoid ambiguity with variables. So, for example, the following is a valid construction: CONSTANTS Two = 2 GOAL A=two, write(A).

There can be several constants sections in a program, but each constant must be declared before it is used. Constant identifiers are global for the rest of the file and can only be declared once. Multiple declarations of the same identifier will result in an error message. You can use constants to redefine names of domains and predicates, except the specially-handled predicates. Refer to "SpeciallyHandled Predicates" earlier in this chapter. Predefined Constants Depending on the target platform selected for compilation, one or more constants will be predefined:

Table 294.4: Predefined Constants295 Constant

Target selections causing it to be defined

os_dos os_os2 os_nt os_unix ws_win ws_pm ws_motif dosx286 platform_16bit platform_32bit

DOS, Phar Lap and Windows OS/2 or PM Windows 95 or Windows NT 32bit mode XENIX, UNIX and Motif MS Windows Presentation Manager Motif Phar Lap286 16-bit platforms 32-bit platforms

Selecting DOS as your target will cause os_dos to be defined, and selecting MS Windows will cause both os_dos and ws_win to be defined. These predefined constants enable you to control platform-dependent conditional compilation.

Conditional Compilation You use conditional compilation when you need to generate different versions of the same program; for example, one version that uses graphics and another that only uses text mode. The syntax for conditional compilation directives is: [ifdef | ifndef] <constantID> ... elsedef ... enddef

<constantID> represents a constant identifier declared in a constants section. The value of the constant is irrelevant; only its presence matters. The ifdef directive succeeds if the constant is defined, while the ifndef directive succeeds if the constant is not defined. The elsedef part is optional. The following program shows a typical use of the conditional compilation directives. CONSTANTS restricted = 1 ifdef restricted

/* if restricted is defined, use this */

savebase(_):write("\nBase cannot be saved in demo version"), readchar(_). elsedef

Chapter 170 Advanced topics

/* otherwise, use this */

429

savebase(Name):write("\nSaving ",Name), save(Name). enddef

Including Files in Your Program You use include to include the contents of another file in your program during compilation. The syntax is: include "OSFileName"

The OSFileName can include a path name, but you must remember that the backslash character used to give subdirectories in the DOS-related versions of Visual Prolog is an escape character in Visual Prolog. Because of this, you must always give two backslash characters when you use the backslash in a path inside the source text. include "\\vip\\include\\error.con"

Under Options | Project | Directories you can give one or more paths separated by semicolons (colons under UNIX) to indicate where the Prolog system should look for the include files (Here, of course, only a single backslash is required). If you don't give an absolute path in your OSFileName, the compiler will in turn try to concatenate each of the paths given in the include directory to your filename in order to locate the file. You can only use include files on section boundaries in a program, so include can appear only where one of the keywords domains, predicates, goal, database, or clauses is permitted. An include file itself can contain further include directives. However, include files must not be used recursively in such a way that the same file is included more than once during compilation. Include files can contain any sections, provided the restrictions on program structure are observed (see page 414).

Compiler Directives A number of compiler features are controlled through compiler directives. You can introduce one or more of the following directives at the beginning of the program text:

bgidriver bgifont check_determ code

config diagnostics errorlevel heap

nobreak nowarnings printermenu project

shorttrace trace

Many of the compiler directives can be set both in the Visual Prolog development environment (from the menus), through command-line options and in the source code. If a compiler directive exists in the source code, its setting will override values set elsewhere. Note, that most of the compiler directives are now obsolete for VPI Programs. bgidriver When you want to link a particular BGI graphics driver directly into your executable BGI program, use the bgidriver compiler directive followed by the public name for the graphics driver file. This directive is only relevant for plain DOS. bgidriver "_CGA_driver_far"

bgifont When you want to link BGI stroked character fonts directly into your executable BGI program, use the bgifont compiler directive followed by the public name for the font file. This directive is only relevant for plain DOS. bgifont "_gothic_font_far"

check_determ Options|Project|Compiler Options|Nondeterm Check When you specify check_determ, the Visual Prolog system will give a warning for each program clause that results in a non-deterministic predicate. There are two kinds of non-deterministic clauses: 1. If a clause does not contain a cut, and there are one or more clauses that can match with the same input arguments for that flow pattern. 2. If a clause calls a non-deterministic predicate, and that predicate call is not followed by a cut. You can use check_determ to guide the setting of cuts. Visual Prolog itself performs extensive tests to decide whether a predicate is deterministic or nondeterministic, so you don't need to fill your programs with cuts merely to save stack space (as is necessary in many other Prolog implementations). Chapter 170 Advanced topics

431

If some predicates are non-deterministic, and you wish to squelch the warning messages, you can precede a predicate declaration by determ or nondeterm; these state that the predicate is to be deterministic or non-deterministic, respectively. For example, PREDICATES nondeterm repeat determ readname(string)

By default, all global predicates are treated as being deterministic. nondeterm allows you to declare a global predicate as non-deterministic. You can also use determ on database predicates intended to have no more than one clause. In this case, the compiler generates simpler code when accessing the predicate. code Options|Project|Compiler Options|Default for Local Calls The code directive specifies the size of the internal code array. The default is 4000 paragraphs (16-byte units) for the 16-bit versions of Visual Prolog, otherwise 10000 paragraphs. For larger programs you might need to specify a larger size. code = Number_of_paragraphs

where Number_of_paragraphs represents the number of memory paragraphs (16 bytes each) required in the code array. For example, the directive: code = 1024

sets the size of the code array to 16 Kbytes. The code directive has no influence on the size of an executable file, it simply controls how much memory the compiler should allocate for the compilation. When the code size exceeds the value 4095, the compiler will switch over to generating FAR calls inside that module. For this reason, you should only use a code size above 4095 if it is really needed. For 32-bit code generation, the size of the code array is practically irrelevant. All code is NEAR, and the operating system will only allocate physical memory when the allocated virtual memory is actually used. config This option is only relevant for old DOS textmode windowing applications ! To let a stand-alone application read a configuration file that defines default window attributes, keyboard setup, etc., place the directive:

config "<ConfigFileName>.cfg"

in your program. The application will read <ConfigFileName>.cfg and set the configurations the same way Visual Prolog does with its configuration file. Various options are passed from the environment to the executable file; these will either be built in during compilation or read from the configuration file during start-up. Some of the values built into the execuatble file can be overwritten by corresponding settings in the configuration file. Here's an overview of these parameters: Passed during compilation Stack Size

Heap Size

Snow Check

Read only from <Config>.cfg

Overwritten from <Config>.cfg x x

Keyboard Layout

Help Lines

Xedit Setup

diagnostics Options|Project|Compiler Options|Diagnostics output When you specify diagnostics, the compiler will display an analysis of your program containing the following information: the names of the predicates used whether a predicate is local, global or defined externally whether a predicate is deterministic or non-deterministic the size of the code for each predicate the domain types of the parameters the flow patterns for each predicate The diagnostics will also produce a listing of which domains are treated as reference domains, and for which domains the compiler generates internal unification predicates. These predicates are generated when you unify two terms in certain non-simple ways. As an example if you are writing L1=L2 where both L1 and L2 are bound to a list, the compiler needs to test all the elements of the list for equality. Chapter 170 Advanced topics 433

Here's an example of a diagnostics display: DIAGNOSTICS FOR MODULE: /usr/pdev/test/flow1.pro Predicate Name ---------------_PROLOG_Goal _p1_0 _p1_1 ---------------Total size

Type -----local global global ------

Determ Size ------ ----yes 216 yes 112 yes 128 ------ ----460

Domains -- flowpattern ----------------------------integerlist -- [i,o,i|o] integerlist -- o ----------------------------

Size of symbol table= 324 bytes Size of PROCONST segment=1705 bytes

Under Options | Global | Environment, it is possible to log the diagnostics output into a file. errorlevel Options | Project | Compiler Options | Errorlevel The compiler directive errorlevel enables you to control how detailed the error reporting should be. The syntax is: errorlevel = d

where d is one of 0, 1, or 2, representing the following levels: d

Level of Error Reporting

Generates the most efficient code. No cursor information will be placed in the code and only the error number will be displayed in an error occurs.

This is the default. When an error occurs, its origin (module name and include file, if applicable) will be displayed. The place where the error was detected within the relevant source file will also be displayed, expressed in terms of the number of bytes from the beginning of the file.

At this level, certain errors not reported at level 1, including stack overflow heap overflow, trail overflow, etc., are also reported.

In a project, it is the error-level option in each module that controls that module's detail of saving the cursor information. If, however, the error-level option in the main module is higher than that of the sub-modules, Visual Prolog might generate misleading error information. For example, if an error occurs in a module compiled with error level 0, which is included in a main module compiled with error level 1 or 2, the system will be unable to show the correct location of the error. Instead, it will indicate the position of some previously executed code. heap

Options | Project | Compiler Options | Heap Size

Note: this is relevant only if you are going to implement a DOS TSR program. The heap directive specifies how much memory your .EXE file should allocate when it is started from DOS. If you don't use the heap directive, or if you set it to the value 0, the program will allocate all available memory. This is normally the right way to do it, but if you want to implement a RAM-resident Visual Prolog program, your program should only allocate the necessary memory. The format is: heap = Number_of_paragraphs

nobreak

Options | Project | Compiler Options | Break Check

Note: this is only relevant for DOS textmode programs. In the absence of the nobreak compiler directive, the Visual Prolog system will generate code to check the keyboard before each predicate call, to ensure that the Ctrl+Break key combination has not been pressed. This slows down program execution slightly and takes up a little extra program space. The nobreak directive prevents this automatic generation of code. When nobreak is in operation, the only way to escape an endless loop is to reboot the computer (DOS, Phar Lap) or kill the process in some other way. nobreak should only be used after a program has been thoroughly tested. nowarnings Options|Project|Compiler Options|Single variable The nowarnings directive suppresses the warnings given when a variable occurs only once in a clause. Note: This directive is only included for completeness. If a variable occurs only once, either it is a mistake or it should be replaced by the anonymous variable - or a variable starting with an underscore. Chapter 170 Advanced topics

435

printermenu

Options|Project|Compiler Options|Print menu in DOS .EXE

Note: this is only relevant for DOS textmode programs. When this compiler directive appears in the program, Visual Prolog will place code in the executable file for handling the Alt-P key. This means that the user will be able to send screen output to the printer or capture it to a log file. project Options|Project|Compiler Options|.SYM File Name The project compiler directive is used in modular programming modular programming. All Visual Prolog modules involved in a project need to share an internal symbol table. If the project is called MYPROJ, the symbol table will be placed in a file called MYPROJ.SYM. The project directive must appear on the first line of a module to specify which project that module belongs to. For example, the following line of code states that a module belongs to the MYPROJ project: project "myproj"

The project name is not allowed to have a path or an extension. If the name of the .SYM file is given in the VDE or as option to the commandline compiler, the project directive will be ignored. See page 438 for complete details about modular programming.

Visual Prolog Memory Management Visual Prolog uses the following memory area’s: Stack

the stack is used for transferring arguments and return addresses for predicate calls. The stack also holds the information for backtrackpoints.

Heap

the heap holds all objects that are more or less permanent, such as database facts, window buffers, file buffers etc.

GStack the global stack, normally called gstack, is the place where lists, compound structures and strings are placed. The Gstack is only released during backtracking. Trail

The trail is only used when the program uses reference variables. It holds information about which reference variables must be unbound during backtracking. The trail is allocated in the heap.

Releasing Spare Memory Resources During program execution, the memory is used for several different purposes; depending upon the purpose, spare memory resources can be released in separate ways. To minimize stack use, avoid unnecessary non-determinism; use the check_determ directive to guide the setting of cuts. Also, take advantage of tail-recursion elimination by writing your predicates so they are tailrecursive. The global stack is used for building strings and structures. In order to save global stack space, write your program so that the outer loop is a repeat...fail loop. The trail will seldom be a problem in Visual Prolog. In all versions of Visual Prolog, the trail is dynamically allocated (in the heap), and will be increased in size when necessary. However, in the 16-bit versions the trail is limited to 64K and the first thing to do if the system complains about trail overflow is to avoid using reference domains. If you want to use reference domains, you should decrease the number of backtrack points by using some cuts (use check_determ). The repeat...fail combination will also release the trail. As a last resort, rearrange your predicate calls so that you create less reference variables. Chapter 170 Advanced topics

437

The heap is used when facts are inserted in a database and to store window buffers, file buffers, graphic drivers, database tables, etc. These areas are automatically released when facts are retracted, windows are closed, and so on.

Modular Programming A Visual Prolog program can be broken up into modules. You can write, edit, and compile the modules separately, and then link them together to create a single executable program. If you need to change the program, you only need to edit and recompile individual modules, not the entire program--a feature you will appreciate when you write large programs. Also, modular programming allows you to take advantage of the fact that, by default, all predicate and domain names are local. This means different modules can use the same name in different ways. Visual Prolog uses two concepts to manage modular programming: projects and global declarations. Among other things, these features make it possible to keep a record of which modules make up a program (this record is called a project), and to perform type-checking across module boundaries. In this section, we'll define the two concepts; then, using a simple example, we'll show you how some modules can be combined into a single, stand-alone program.

Global Declarations By default, all names used in a module are local. Visual Prolog programs communicate across module boundaries using the predicates defined in the global predicates and global database sections. The domains used in these global sections must be defined as global domains, or else they must be pre-defined domains. ALL THE MODULES IN A PROJECT NEED TO HAVE EXACTLY THE SAME GLOBAL DATABASE AND GLOBAL DOMAINS DECLARATIONS. If you mix this up, all sorts of strange things will probably happen, such as a hung computer under DOS or MS Windows, or a protection violation on other platforms. The easiest way to ensure this is correct is by writing all global declarations in a single file, which you can then include in every relevant module with an include directive. For example, if all your global declarations are in a file called global.inc, you can include that file in every relevant module by adding the directive:

include "global.inc"

to the top of each module. All global declarations must appear before any local declarations local declarations. Global Domains You make a domain global by writing it in a global domains section. In all other respects, global domains are the same as ordinary (local) domains. Note: If any global domain definition is changed, all modules in that project must be recompiled. Global Database You make a database section global to a project by preceding the database keyword with the keyword global. You can only give initializing facts for global databases in the main module, which is the one containing the goal section. The goal section must appear before the global database clauses in the main module. Note: If any global database definition is changed, all modules in that project must be recompiled. Global Predicates Global predicate declarations differ from ordinary (local) predicate declarations because they must contain a description of the flow pattern(s) by which each given predicate can be called. If such one is not specified, all arguments will be input. The syntax for a global predicate declarations is: [ { determ | nondeterm | single | nocopy} ] [ domain ] name arglist [ - flowpattern [ [,] flowpattern ]* ] [ language ] [ namespec ]

(curly braces indicate "choose one", square brackets indicate optional items) where Note, that predicate types can also appear before the flowpattern. domain is the return domain, if you're declaring a function name is the name of the predicate arglist is of the form ( [ domain [ , domain ]* ] )

Chapter 170 Advanced topics

439

flowpattern is of the form ( flow [ , flow ]* ) where flow is { i | o | functor flowpattern | listflow } where listflow is '[' flow [ , flow ]* [ '|' { i | o | listflow } ] ']'

language is of the form language { c | asm | pascal | prolog | stdcall | syscall }

namespec is of the form as "extname"

The namespec may be used to specify the public object-code name, overriding the default naming used by Visual Prolog. The main use of this is when you're linking in modules written in other languages. The language directs the calling convention used when calling the predicate. This defaults to "prolog". In the following global predicate declaration, name and home are of type string, and age is of type integer; the arguments to first_pred can either be all bound (i, i, i) or all free (o, o, o): first_pred(name,home,age) - (i,i,i) (o,o,o)

Here is the declaration for a predicate with either compound flow of an integer list, or plain output flow: p1(integerlist) - ([i,o,i|o]),(o)

Finally, this declaration specifies compound flow for an object declared as func(string,integer) coming from a domain called mydom: pred(mydom) - (func(i,o)) (func(o,i))

Note: If any global predicate definition is changed, only the modules that refer to this predicate need to be recompiled. However, it is rather critical that this recompilation is done; if you change the flow pattern of a predicate the calls using it will need different code. It doesn't matter in which module the clauses for global predicates appear, but--as with local predicates--all clauses must appear together.

CHAPTER

296

Interfacing with Other Languages 297Although Visual Prolog is an excellent tool for many purposes, there are still reasons to use other languages. For example, it's easier to perform numeric integration in C, and interrupt-handling and low-level stuff is perhaps better done in Assembly language. Moreover, if you've developed a large program in another language that already solves some aspect of the problem, this work should not be wasted. For these reasons, Visual Prolog allows you to interface your programs with other languages, as long as those languages produce standard object files and follows the conventions outlined in this chapter. In this chapter you will find a number of examples of interfacing C and Visual Prolog. Their source files are in the DOC\EXAMPLES directory or the FOREIGN directory of your distribution. In order to run them, you need to have the appropriate development system and/or C compiler and libraries installed. The process to compile and link the examples varies considerably between the different operating systems and the different C compilers. In the foreign subdirectory of your distribution you will find thorough instructions and examples for the different platforms and compilers. Read these instructions carefully. When using the Visual Prolog Development Environment, you don't, strictly speaking, need to know how to compile C programs, how to run the linker, or which libraries to specify and how to do it. This is handled automatically. However, you should have a fairly thorough understanding about C, and be able to write, compile and link multi-module C programs yourself.

Using DLL’s A dynamic-link library (DLL) is a binary file that acts as a shared library of predicates that can be used simultaneously by multiple applications. Visual Prolog can generate dll’s and link in dll’s staticly or load dll’s dynamicly. For more information about Visual Prolog and dll’s please see the examples VPI\EXAMPLES\DLL and VPI\TOOLEXAMP\BUILD.

Chapter 170 Advanced topics

441

Calling Other Languages from Visual Prolog In this section, we cover what you need to know to call C, Pascal and assembler routines from Visual Prolog. Before calling routines and functions written in other languages, you need to declare them as external predicates in Visual Prolog. You also need to understand the correct calling conventions and parameter-pushing sequences, and you need to know how to name the different flow variants of your external predicates.

Declaring External Predicates To inform the Visual Prolog system that a given global predicate is implemented in another language, you need to append a language specification to the global predicates declaration, as briefly mentioned in chapter 298: GLOBAL PREDICATES add(integer,integer,integer) - (i,i,o),(i,i,i) language c scanner(string,token) - (i,o) language pascal triple(integer,real) - (i,o) language asm

In Visual Prolog, you explicitly list the interfaced language; this simplifies the problems inherent in calling conventions, such as activation record format, naming convention and returning conventions.

Calling Conventions and Parameter Passing The 80x86 processor family gives programmers a choice between NEAR and FAR subroutine calls, when running 16-bit programs. Visual Prolog requires all global routines to be FAR. The same applies to pointers to data objects. Many 16bit compilers for the 80x86 family require you to choose between 16-bit and 32bit pointers, where the 16-bit pointers refer to a default segment. In order to access all of memory, Visual Prolog always uses 32-bit pointers. For 32-bit programs, "NEAR" means 32 bits and the above considerations are irrelevant. Input parameters For input parameters, the value is pushed directly, and the size of the parameter depends on its type.

Output parameters An output parameter is pushed as a 32-bit pointer to where a values must be assigned. Return Values Visual Prolog follows the most widely adopted register convention for function values on the 80x86 CPU family. This should not be of any concern in most cases, but is included here for completeness. Table 299.1: Registers for Return Values300 Operand Size

Program Type

16 bit

32 bit

byte (8 bits)

word (16 bits)

dword (32 bits)

DX:AX

EAX

Pointers are 32 bits in size and are handled as dwords. The Program Type is determined by the operating system, Floating point values are exceedingly troublesome to handle. They may be returned in registers, on the (emulated) coprocessor stack, and the pascal calling convention will frequently return them through pointers. Currently pascal functions cannot return floating point values. See the notes in the FOREIGN subdirectory of your distribution for any special considerations for your platform. In any case, floating point values can always be returned in arguments. However, take special note that Visual Prolog's real corresponds to a C double (8 bytes). You should also be aware that currently external C functions cannot return C structs (but they may of course return pointers to structs). Multiple declarations In Visual Prolog, a predicate can have several type variants, arities, and flow variants, and a separate procedure is needed for each type and flow variant. When you implement predicates, having several versions, in C, each C function must have a name corresponding to the name generated by Visual Prolog. The naming convention used by Visual Prolog is straightforward; the predicate name is used Chapter 170 Advanced topics

443

as the root, and the suffix _X is appended to signify the variant number, where X is an integer starting at 0. If there is only one variant, no suffix is appended. Consider the following program: GLOBAL PREDICATES add(integer,integer,integer) - (i,i,o),(i,o,i),(o,i,i),(i,i,i) language c square(integer,integer) - (i,o) GOAL add(2,3,X), write("2 + add(2,Y,5), write("5 add(Z,3,5), write("5 add(2,3,5), write("2 + square(5,Sq), write("5

3 = ",X), nl, 2 = ",Y), nl, 3 = ",Z), nl, 3 is 5"), nl, squared is ",Sq).

A module linked with this program should contain the following C functions: add_0 for the first flow pattern (i,i,o) add_1 for the (i,o,i) flow pattern add_2 for (o,i,i) add_3 for (i,i,i) square As an example, the following C module implements square as well as all flow patterns for add: add_0(int x, int y, int *z) { *z = x + y; }

/* (i,i,o) flow pattern */

add_1(int x, int *y, int z) { *y = z - x; }

/* (i,o,i) flow pattern */

add_2(int *x, int y, int z) { *x = z - y; }

/* (o,i,i) flow pattern */

add_3(int x, int y, int z) { if ( (x + y) != z ) RUN_Fail(); }

/* (i,i,i) flow pattern */

square(int i,int *i_sq) { *i_sq = i*i; }

Parameter pushing order When interfacing to a routine written in C, the parameters are pushed onto the stack in reverse order and, after return, the stack pointer is automatically adjusted by Visual Prolog. When calling languages other than C, the parameters are pushed in the normal order, and the called function is responsible for removing the parameters from the stack. Leading underscored On the 16bit platforms, C compilers will prefix the name of public C functions with an underscore. Therefore, global predicates declared as language C will also have their name prefixed with an underscore if the target platform is one of these. NT naming convention For the Win32 API, the number of bytes pushed on the stack will together with a dollar sign be suffixed to the predicate name. this means that a predicate p which has two integer arguments will be names p$8. When choosing the calling convention stdcall under Win32, this convention will be used. Converting the name to Uppercase (Pascal) PASCAl uses the convention, that the name is converted to uppercase. So if language PASCAl is used, the name will during .OBJ module generation be converted to uppercase. Adjustment of stackpointer There are two possibilities of adjusting the SP register . This can be done either by the called function or the calling function. Traditionally PASCAL does this in the called function, while C does it in the calling function. Table 301.2: Calling conventions302 Convert name to upper case pascal c

Add leading under Score

Push args Reversed

Adjust SP after return

NT naming convention

Chapter 170 Advanced topics

445

stdcall

syscall

X X

The AS "external_name" Declaration As an alternative to the automatic naming convention, you can use the as keyword in the global declaration, like this: GLOBAL PREDICATES scanner(string,token) - (i,o) language c as "_myscan"

The result of this is that Visual Prolog will refer to the name _myscan in the object file instead of _scanner. You would still refer to the name scanner in your Visual Prolog source. You can only use the as option if there is a single flow variant for the predicate.

Domain Implementation Most types normally used in C form a subset of Visual Prolog domains, and hence have direct equivalents. Below we discuss both simple and complex domain equivalents in C.

Simple Domains The implementation of Visual Prolog's simple domains are outlined in the following table: Table 303.3: Visual Prolog Simple Domains304 Domain

Implementation 16-bit OS

32-bit OS

char, byte

1 byte (see note)

(u)short, word

2 bytes

(u)long, dword

4 bytes

unsigned, integer

2 bytes

4 bytes

real

8 bytes (IEEE format)

ref

4 bytes

Note: The char and byte domains occupy a machine word when pushed on the stack (2 bytes for 16-bit programs, 4 bytes for 32-bit programs).

Complex Domains All non-simple domains are implemented as pointers to things. The string and symbol domains are pointers to null-terminated character arrays, with symbols being hashed and stored in the symbol table in the heap. The binary domain is a pointer to a block of memory, prefixed by a dword (32bit platforms) or word (16bit platforms) indicating the net size of the block. Size

bytes ^ | Pointer

Special consideration must be given to allocation of memory for complex domains. This will be detailed in a later section.

Chapter 170 Advanced topics

447

Ordinary Compound Objects and Structures User-defined compound objects are pointers to records (structs and unions in C). The general format of these is a byte representing the functor (domain alternative), followed by the individual components of the term. These will vary, depending on which alternative we're dealing with. In any case, components belonging to simple domains are stored directly in the term record itself, while complex components are themselves stored as pointers. For example, in this code fragment: DOMAINS mydom = i(integer); c(char); s(string)

the functor number will be 1 for the first alternative, i(integer), 2 for the second, c(char), and 3 for the third. A suitable C typedef for mydom would be: typedef struct { unsigned char func; union { int i; char c; char *s; } u; } MYDOM;

Here func will have the value 1, 2 or 3, depending on which domain alternative we're dealing with. This then indicates which of the union's components it's appropriate to access. Functorless Terms (structs) By prefixing a compound domain declaration with the compiler directive struct, the terms belonging to that domain will not carry functors with them: DOMAINS rec = struct record(d1,d2,...)

Apart from the struct directive, terms belonging to a functorless domain are used and written exactly like other terms in your program, except that there can be no alternatives in a functorless domain. Functorless terms allow you to duplicate C structs when interfacing to C routines and libraries using predefined structs. Apart from that, they'll save you a bit of memory if you don't need alternatives in the domain.

Lists Lists are implemented exactly like ordinary compound domains, with a field at the end of the record pointing to the next. This is known as linked lists in C terminology. From a Prolog perspective lists are merely a notational convenience. Given for instance a declaration for a list of strings: DOMAINS strlist = string*

the C structures relevant for this are identical to those for: DOMAINS strlist = elem(string,strlist); endoflist()

The records for the elem alternative will contain: 1. a functor 2. a pointer to a string 3. a pointer to the next element in the list and the record for the endoflist alternative will only contain a functor. This collection of fields is reflected in the C data structure for strlist: struct node { unsigned char functor; char *value; struct node *next; } strlist;

/* The type */ /* A string pointer */ /* A pointer to struct node */

The functor field indicates the type of list record. The value is 1 if it's a list element, and 2 if it's the end of the list.

Memory Considerations While all memory considerations are handled automatically when you write pure Prolog code, you need to take special care when interfacing to foreign languages. In this section we'll describe several of these aspects.

Memory Alignment C compilers for 32-bit platforms will usually align data on dword boundaries, while those for 16-bit platforms will usually align on byte boundaries. The reason Chapter 170 Advanced topics

449

for aligning on word or dword boundaries is speed. On a 32-bit platform, dword alignment will give up to 10-12 percent faster execution than byte alignment. For simple variables alignment isn't important, but in order for Prolog terms and C structures to be compatible, the data contained in the records must be identically aligned. To this end, Visual Prolog gives you the option of selecting a different alignment than what's the default for your platform. The default is dword if you're using a 32-bit version of Visual Prolog, otherwise byte. The alignment scheme may be selected with the help of the Options | Project | Compiler Options menu item, or through the -A command line option. Additionally, the align compiler directive may be used to override alignment on selected domains, like this: DOMAINS dom = align { byte | word | dword } func(d1,d2,...) [; func1(...); ... ]

The align directive must appear before any alternatives in the domain, and all alternatives will have the alignment specified. It's not possible to specify different alignment for individual alternatives. For functorless terms, the align directive should appear after the struct directive. Note that when several processes share a database or communicate over pipes, it's crucial that the domains involved use identical alignment. Example Byte alignment is easy: each element is simply put right after the previous one. Given the declaration dom = struct my_struct(char,short,char,long) (recall that the struct directive declares the term to be functorless), the term my_struct('P',29285,'B',1702063209) is stored in memory like this: Byte number: | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | _______________________________ | | | | | |'P'| 29285 |'B'| 1702063209 | |___|_______|___|_______________|

Word and dword alignment is a bit trickier. Here, items are stored in memory so that accessing them won't cross a word or dword boundary. That means that the individual elements of terms may be followed by a number of unused bytes,

depending on the size of the following element. With dword alignment, the term above would be stored like this: Byte number: | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10| 11| _______________________________________________ | | | | | | | | | |'P'|PAD| 29285 |'B'|PAD|PAD|PAD| 1702063209 | |___|___|_______|___|___|___|___|_______________|

The PADs indicate unused bytes, allowing the values following them to be stored on suitable boundaries. Notice that it's sufficient for the value 29285 to be aligned on a word boundary, because it's a short (16 bits); accessing it on a word boundary won't cross any undesirable boundaries.

Memory Allocation When you create and return compound objects to Visual Prolog, memory for the objects must normally be allocated on the Global Stack. This memory will automatically be released if you fail back to a point previous to its allocation. GStack memory is allocated using: void *MEM_AllocGStack(unsigned size);

You would typically use C's sizeof function to determine how much memory to allocate. Given for instance the mydom domain discussed previously, the Prolog declarations for a C routine returning a term belonging to that domain in an argument would be: /* Program mydom_p.pro */ project "mydom" global domains mydom = i(integer); c(char); s(string) global predicates determ make_mydom(mydom) - (o) language C goal make_mydom(MD), write(MD), nl.

Chapter 170 Advanced topics

451

And the C code for mydom and make_mydom could be: /* Program mydom_c.c */ typedef struct { unsigned char func; union { int i; char c; char *s; } u; } MYDOM; void *MEM_AllocGStack(unsigned); char *MEM_SaveStringGStack(char *); void make_mydom(register MYDOM **md) { *md = MEM_AllocGStack(sizeof(MYDOM)); (*md)->func = 3; (*md)->u.s = MEM_SaveStringGStack("wombat"); }

Notice that, as terms are handled through pointers in Prolog, the argument to make_mydom is a pointer to a term pointer. This example also makes use of another GStack-related function, MEM_SaveStringGStack, which allocates GStack space for the string (based on its length), then copies the string into the allocated space, returning a pointer to it. There's a few other handy functions in Visual Prolog's library: char *MEM_SaveStringHeap(char *String); /* Copies String to heap */ unsigned STR_StrLen(char *String); /* Returns length (excluding terminating null byte) of String */ void MEM_MovMem(void *Source,void *Dest,unsigned Len); /* Moves Len bytes from Source to Dest; these may overlap */

Pre-allocation of Memory Many C library functions require you to specify a pointer to a structure, which the C routine then fills in. In this case the compound flow pattern for global predicates should be used to specify what's happening:

GLOBAL DOMAINS off_t, time_t = long dev_t = short stat = struct stat(dev_t,ushort,ushort,short,ushort,ushort, dev_t,off_t,time_t,time_t,time_t) GLOBAL PREDICATES determ integer stat(string,stat) (i,stat(o,o,o,o,o,o,o,o,o,o,o)) language C

When you call stat ..., 0 = stat("/unix",Stat), !, write(Stat).

Visual Prolog will allocate memory for the stat structure before the call. The sizeof function Visual Prolog has a sizeof function that duplicates C's sizeof function, returning the size of the specified domain or variable. For a compound domain with alternatives, sizeof will return the size of the largest alternative. Given a second argument of a functor from one of the domain alternatives, sizeof will return the size of that particular alternative. Given a variable, sizeof will return the size of the corresponding domain (alternative), except that for string variables or constants, sizeof will return the number of bytes in the string including the terminating zero byte. The program align.pro illustrates alignment selection and the use of the sizeof function: /* Program align.pro */ DOMAINS dom = struct f(char,integer) dom1 = align word f(integer,integer,long); g(string) refint = reference integer predicates refint(refint) clauses refint(_). goal A = sizeof(dom), write("\nSize=",A),

Chapter 170 Advanced topics

% Find the size of a functorless domain

453

% when there are alternatives, the largest is returned B = sizeof(dom1), write("\nSize=",B), % Find size of a single alternative C = sizeof(dom1,g), write("\nSize=",C), % Find size of a term pointed to by a variable X = f(1,1,1), D = sizeof(X), write("\nSize=",D),

% This is from dom1

% Find size of a string pointed to by a variable Y = "hello there", E = sizeof(Y), write("\nSize=",E), % Find size of a reference variable refint(Z), F = sizeof(Z), write("\nSize=",F).

Load and run this program. Try changing the domains and their alignment, and watch the results. malloc and free When writing functions in other languages, you often need to allocate dynamic memory. You've already seen MEM_AllocGStack, but this allocates memory on Prolog's Global Stack, which is released automatically. Permanent allocations should be done in the heap, and because Visual Prolog already has suitable memory allocation routines, it's generally preferable to use these. In fact, in DOS it's mandatory to use them, since a foreign memory allocation package would be allocating memory from the same physical memory as Visual Prolog. On other platforms, you can use C's malloc and free, but this would duplicate an amount of code and data, and the two packages would both be holding released memory in separate pools. Moreover, Visual Prolog's heap allocation system has a performance far superior to that supplied with most C compilers. Therefore, on all platforms except UNIX, public routines for malloc and free, consisting of bindings to Visual Prolog's heap allocation routines, are provided in the initialization assembler and object files. These files are found in the subdirectories for the different platforms and compilers in the FOREIGN directory of your distribution. Note that when linking, it's essential that the

appropriate initialization file appears before the (C) library containing malloc and free.

Examples List Handling In this section we give a more useful example that shows how to convert a list to an array and back to a list again. The C routine ListToArray takes a list of integers, converts this to an array placed on the Global Stack, and returns the number of elements. The conversion is done in three steps: 1. The list is traversed in order to count the number of elements. 2. The array with the needed number of elements is allocated. 3. The list is traversed again while the elements are transferred to the array. The C routine ArrayToList takes an integer array and the size of the array as arguments, then converts these to a list of integers. This routine only makes one pass, building the list as it indexes through the array. All of this is used in the C-coded predicate inclist. When given a list of integers, inclist first converts the input list to an array, increments the elements of the array by 1, then converts the array back to a list of integers. /* Program lstar_p.pro */ project "lstar" global domains ilist = integer* global predicates inclist(ilist,ilist) - (i,o) language c goal inclist([1,2,3,4,5,6,7],L), write(L).

Here is the C program defining the two C procedures ListToArray and ArrayToList, and the external Visual Prolog predicate inclist.

Chapter 170 Advanced topics

455

/* Program lstar_c.c */ #define listfno 1 #define nilfno 2 typedef unsigned char BYTE; void *MEM_AllocGStack(unsigned); typedef struct ilist { BYTE Functor; int Value; struct ilist *Next; } INTLIST; int ListToArray(INTLIST *List,int **ResultArray) { INTLIST *SaveList = List; int *Array, len; register int *ArrP; register int i; /* Count the number of elements in the list */ i = 0; while ( List->Functor == listfno ) { i++; List = List->Next; } len = i; Array = MEM_AllocGStack(i*sizeof(int)); ArrP = Array; /* Transfer the elements from the list to the array */ List = SaveList; while ( i != 0 ) { *ArrP++ = List->Value; List = List->Next; i--; } *ResultArray = Array; return(len); }

void ArrayToList(register int *ArrP,register int n, register INTLIST **ListPP) { while ( n != 0 ) { *ListPP = MEM_AllocGStack(sizeof(INTLIST)); (*ListPP)->Functor = listfno; (*ListPP)->Value = *ArrP++; ListPP = &(*ListPP)->Next; n--; } *ListPP = MEM_AllocGStack(sizeof((*ListPP)->Functor)); /* End of list */ (*ListPP)->Functor = nilfno; } void inclist(INTLIST *InList,INTLIST **OutList) { register int *ArrP, i, len; int *Array; len = ListToArray(InList,&Array); ArrP = Array; for ( i = 0; i < len; i++) ++*ArrP++; ArrayToList(Array,len,OutList); }

This program belongs to the kind where memory alignment can be critical. If you intend to compile to several platforms, you're well advised to keep an eye on this. As a first step, check that the sizes of the structures shared by C and Prolog are the same; the padding applied when aligning on non-byte boundaries will make things a bit bigger. The sizeof function comes in handy here. You can write a small C function: unsigned c_ilsize(void) { return(sizeof(INTLIST)); }

returning the size of the INTLIST structure. This can then be used by a Prolog predicate to verify that the sizes of INTLIST and ilist are identical: GLOBAL PREDICATES unsigned c_ilsize() language C PREDICATES scheck

Chapter 170 Advanced topics

457

CLAUSES scheck:- ILSize = sizeof(ilist), ILSize = c_ilsize(), !. scheck:- write("ilist element sizes differ\n"), exit(1).

Calling Prolog from Foreign Languages If you supply Prolog clauses for global predicates declared as being of foreign language, those predicates may be called from foreign languages. They will have parameter access and entry and exit code, including register preservation, as for the language specified. Hello This small project is hello-world, with a twist. /* Program hello_p.pro */ global predicates char prowin_msg(string) - (i) language c hello_c - language c clauses prowin_msg(S,C) :write(S," (press any key)"), readchar(C). goal prowin_msg("Hello from PDC Prolog"), hello_c.

The global predicate prowin_msg is now accessible from C and can be called just like any other C function: /* Program hello_c.c */ char prowin_msg(char *); void hello_c() { while ( prowin_msg("Hello from C (press 'C')") != 'C' ) ; }

As is evident, values may be returned to foreign languages. Standard Predicates Most of Visual Prolog's standard predicates can be called from C, but their public names and exact functionality are subject to change without notice. It's therefore

strongly recommended that you write a small set of interface routines if you want to call Visual Prolog standard predicates from C. The following illustrates bindings to a number of Visual Prolog's DOS Textmode I/O predicates: /* Program spred_p.pro */ project "spred" global predicates myfail language c as "_fail" mymakewindow(integer,integer,integer,string,integer,integer, integer,integer) - (i,i,i,i,i,i,i,i) language c as "_makewindow" myshiftwindow(integer) - (i) language c as "_shiftwindow" myremovewindow language c as "_removewindow" write_integer(integer) - (i) language c as "_write_integer" write_real(real) - (i) language c as "_write_real" write_string(string) - (i) language c as "_write_string" myreadchar(char) - (o) language c as "_readchar" myreadline(string) - (o) language c as "_readline" extprog language c clauses myfail:- fail. mymakewindow(Wno, Wattr, Fattr, Text, Srow, Scol, Rows, Cols):makewindow(Wno, Wattr, Fattr, Text, Srow, Scol, Rows, Cols). myshiftwindow(WNO):- shiftwindow(WNO). myremovewindow:- removewindow. write_integer(I):- write(I). write_real(R):- write(R). write_string(S):- write(S). myreadchar(CH):- readchar(CH). myreadline(S):- readln(S). goal extprog.

These may be accessed freely by C, as illustrated by extprog:

Chapter 170 Advanced topics

459

/* Program spred_c.c */ void extprog(void) { char dummychar; char *Name; makewindow(1,7,7,"Hello there",5,5,15,60); write_string("\n\nIsn't it easy"); readchar(&dummychar); write_string("\nEnter your name: "); readline(&Name); write_string("\nYour name is: "); write_string(Name); readchar(&dummychar); removewindow(); }

Calling an Assembler Routine from Visual Prolog You can also call assembler routines from Visual Prolog. The activation record is the same as for pascal (that is, parameters are pushed left to right), and the called routine should pop the stack itself. If you have a C compiler supporting inline assembler, things will be considerably easier than if you have to do everything yourself. In any case there seems to be little point in using assembler since C handles most things, but a small example is included here for completeness. For obvious reasons, the code differs between 16 and 32 bit platforms. Suppose you want to write a routine returning a 32-bit sum of the characters in a string, and also verifies that all characters are within a certain range, say A-Z. The Prolog code for this could be: /* Program csum_p.pro */ project "csum" global predicates integer sum_verify(char,char,string,ulong) - (i,i,i,o) language asm predicates uc_check(string)

clauses uc_check(S):0 = sum_verify('A','Z',S,Sum), !, write('"',S,"\" OK, sum = ",Sum,'\n'). uc_check(S):- write('"',S,"\" fails\n"). goal uc_check("UNIX"), uc_check("Windows").

where we have adopted the convention that a return value of 0 means the string was OK. Here is the suitable 16-bit assembler code: /* Program csum_a16.asm */ ;/* Copyright (c) 1986, '92 by Prolog Development Center */ ; 16-bit version CSUM_A16_TEXT CSUM_A16_TEXT _DATA _DATA CONST CONST _BSS _BSS DGROUP

CSUM_A16_TEXT ASSUME PUBLIC sum_verify push ov

SEGMENT WORD PUBLIC 'CODE' ENDS SEGMENT WORD PUBLIC 'DATA' ENDS SEGMENT WORD PUBLIC 'CONST' ENDS SEGMENT WORD PUBLIC 'BSS' ENDS GROUP CONST, _BSS, _DATA ASSUME CS: CSUM_A16_TEXT, DS: DGROUP, SS: DGROUP SEGMENT CS: CSUM_A16_TEXT sum_verify PROC FAR bp bp,sp

lolim equ 16 hilim equ 14 string equ 10 sum equ 6

Chapter 170 Advanced topics

461

xor xor les mov mov xor

dx,dx bx,bx di,[bp+string] cl,byte ptr [bp+lolim] ch,byte ptr [bp+hilim] ax,ax

ALIGN 2 loopy: add adc mov inc cmp jb cmp jbe

bx,ax dx,0 al,byte ptr es:[di] di al,cl end_check al,ch loopy

end_check: or jnz les mov mov inc

al,al go_home di,[bp+sum] es:[di],bx es:[di+2],dx ax; ax: 0 -> 1

go_home: dec mov pop ret sum_verify

ax sp,bp bp 12 ENDP

CSUM_A16_TEXT END

; Do sum in dx:bx ; Pointer to string ; Low limit in cl ; High limit in ch

; Add sum

; ax: 1 -> 0, or 0 -> -1

ENDS

When writing assembler code, take special care that the sizes of things on the stack follow the machine's natural word-size. This is 2 bytes on 16-bit machines and 4 bytes on 32-bit machines. A good first attempt is to compile a dummy C routine, with the correct parameters and local variables, to assembler, and then use the entry, exit, and variable access code generated by the C compiler. It isn't necessary to preserve any of the usual registers when foreign language routines are called from Prolog, but if you're calling from C or assembler it's assumed that you preserve si and di (esi and edi on 32-bit platforms). On 32-bit platforms, ebx must also be preserved.

Index 8 80x86 processor family 442 A abs 197 absolute values 196 access external database via B+ trees347 access modes 296 accessmode 362 adding facts at run time 179 addition 192 alignment memory databases and pipes 450 alternate solutions lxviii and 216 append 166, 229, 405 approximate 202 arc tangent 198 arctan 198 arguments xxx compound data objects 104 known 222 reference domain 226 arithmetic 191 operations 192 assembler routines calling 460 assert 180 asserta 180 assertz 180 assignment statements 201 atoms 103 automatic type conversion 103, 228, 320

B B+ trees 325, 343 key length 343 multiple scans 344 backslash 422 backtracking lxviii, 120 basic principles lxxiv point lxviii basic concepts of Prolog xviii string-handling procedures 311 beep 382 bgidriver compiler directive 431 bgifont compiler directive 431 binary search trees 145 trees reference domains and 230 binary terms accessing 246 comparing 247 creating 245 unifying 247 binding flowpatternstopredicatecalls 222 bios 387 bit-level operations 384 bitand 385 bitleft 386 bitnot 384 bitor 385 bitright 386 bitxor 306, 385 break 257 breakpressed 258 bt_copyselector 345

bt_create 344 bt_delete 345 bt_selector 227, 329 bt_statistics 345 bt_updated 365 byte 424 C C lists 455 calculating 126 calling conventions 442 calls non-deterministic lxxxiii carriage return 422 chain_delete 337 chain_first 337 chain_inserta 336 chain_insertafter 337 chain_next 338 chain_terms 337 chains manipulating 336 of terms 327 char 102, 422 characters 102, 204 converting to integers 318 converting to strings 318 check_determ compiler directive 266, 271, 431, 437 class 219 Classes 218 clause 168 clauses xxvi Horn xviii non-deterministic 271 section xliii, 422 closefile 287 code compiler directive 432 codes 102 command line 380 comments xxxvii

comparing 191 arithmetic expressions 200 characters 204 comparison 204 compilation conditional 429 compiler 132 Compiler lxi compiler directives lxi, 430 check_determ 266, 271 determ 179 include lxi composebinary 246 compound 154 data objects 104 lists 170 mixed-domain declarations 116 objects declaring domains 111 compound flow pattern 223 compund object 425 concatenation strings 314 conditional lviii conditional compilation 429 config compiler directive 432 CONFIG.SYS 331 configuration file 432 constants 102 declaring lix predefined 428 section lix, 427 consult 181, 255 consulterror 255 controlling the flow analysis 225 conversion case 319 character to integer 318 type lviii copyfile 295 cos 197

counting list elements 158 criticalerror 263 cursor predicate 222 cutbacktrack 265 cuts 132 as goto xcvi using lxxxiv D data security 334 database global 439 predicates restrictions 177 section 177, 421 databases lix date 108, 379 db_begintransaction 363, 365, 374 db_btrees 335 db_chains 335 db_create 331 db_endtransaction 363, 365 db_flush 333 db_open 364 db_openinvalid 333 db_reuserefs 330 db_saveems 333 db_selector 227, 329 db_setretry 366 db_statistics 335 db_updated 365 dbasedom 178, 417 declarations xliv as external name 446 database selectors 329 denymode 362 different domains 415 domains as reference 226 facts section 177 local 439 predicate domains 418 predicates xliv reference domains 418

declarative language xviii definition 221 deletefile 294 denymode 362 depth-first search 142 determ 432 determinism lxxxiii cut and lxxxvii diagnostics compiler directive 433 difference lists 171 difftime 381 dirclose 302 dirfiles 303 dirmatch 302 diropen 301 disabling breaks 257 diskspace 383 displaying external database contents 351 div 196 dividing words into syllables 404 division 192 domains file 289 binary 244 compound object 416 converting reference 228 db_selector 327 internal 421 ref 330 section 414 shortening declarations 417 specially handled 417 user-defined 419, 425 Domains 210 DOS critical error 263 double quotation marks 423 dumpDba 360 duplettes 344 dword 424 dynamic cutting 265

dynamic memory allocation 454 E environment symbols 378 eof 291 equal predicate 201 sign unifying compound objects105 equality 201, 202 equations 202 error memory overflow 230 error reporting 253 errorcodes 250 reserved 250 errorlevel 253 errorlevel compiler directive 434 errormsg 252 errors constant definition recursion 428 consult 255 control in EXE files 263 readterm 256 term reader 255 trapping 250 existdir example 302 existfile 293 exit 251 exp 198 expressions 191 external predicates declaring 442 external databases accessmode 362 B+ tree names 335 chain names 335 closing 334 copying 332 deleting 334 denymode 362 filesharing 374

flushing 333 location of 331 locking of 364, 368, 374 non-breakdown 353 opening 332 programming 349 RAM requirements 324 reopening 363 scanning through 350 sharemode 363 statistics 335 structure of 325 system 324 external program 377 F facts xix, 214, 219 section lix facts databases using 178 fail lxxxii, 123 failing l FAR subroutines 442 file_bin 284 file_str 281 fileattrib 305 fileerror 264 filemode 287 filenameext 300 filenamepath 299 filepos 290 files attributes 295 opening 286 domain 416 dumped (external databases) 360 external databases file-sharing 362 object 441 symbolic file names 423 filesharing predicates 364 transaction 363 findall 168

flag 388 floating-point numbers 423 flow pattern xxxix, 167, 222 non-existent 225 flush 293 formatted output examples 279 to string variable 315 formatting arguments into a string 312 frontchar 312 frontstr 314 fronttoken 313 functions 193 declaring 420 return values 233 functors 416 G games adventures in a cave 398 Towers of Hanoi 402 getbacktrack 265 getbinarysize 246 getentry binary access 246 global declarations 438 domains 439 stack 437 global sections lxi goal trees lxxiii goals xxxiv H heap compiler directive 435 heapsize 384 Horn clauses xviii I I/O redirecting 289 ports 390 IEEE standard format 103 ifdef 429 implementation 208

in 156 in_file 331 in_memory 331 include compiler directive 438 include file error.con 250 including files in your program 430 inference engine xviii, 392 infix predicate 105 input argument xxxix instances 209 instantiating reference variables 227 integer 424 integers arithmetic 196 converting to characters 318 converting to strings 318 division 196 random 195 interchangeability of unknowns 161 internal databases 176 goals 414 system time clock 379 isname 315 K key_current 347 key_first 346 key_insert 346 key_next 347 keyboard 417 keyword lii, 208 keywords 413 L lasterror 254 length of a string 315 less than 200 listdba 351 lists 117 appending 165

as compound objects 426 declaring 154 defined 153 domains 415 handling 455 length 158 linked 449 membership 164 mixed types 426 passing to C 449 processing 155 using 156 ln 198 loading facts from a file at run time 181 log 198 log file external databases 353 logarithm 198 logic program defined xviii logical AND 385 circuit 401 inference xviii NOT 384 OR 385 XOR 385 long 424 loop variables 134 loops backtracking 124 low-level support 387 M macro definition 427 makebinary 245 malloc 454 manipulation external databases 330 marktime 381 matching lxv member 164, 229, 419 membyte 389

memory access 389 alignment 426, 449 databases and pipes 366 allocation 451 freeing 437 management 437 overflow error 230 regaining (external databases)334 merging free space 334 mod 196 modular arithmetic 196 modular programming 438 modules 438 moving external databases 332 multiple arity lvii solutions lxxxiii multiple-type arguments 116 multiplication 192 multitasking 362 mykey_next 358 mykey_prev 358 mykey_search 358 N N Queens problem 408 names 412 external database predicates 325 redefining 428 restrictions in 412 naming conventions 443 extended database predicates 325 natural logarithm 198 negation lxxxviii newline 422 nl 273 nobreak compiler directive 435 non-determinism lxxxiii non-deterministic clause warning431 not lxxxviii, 392 notation 200 nowarnings compiler directive 435

numbers 103 converting 200 hexadecimal 191 octal 191 O objects xix compound 425 Objects 210 of lxvi, 158, 192 openappend 287 openfile 297 opening B+ trees 345 invalid external database 333 openmodify 287 openread 150, 286 openwrite 150, 286 operands 191 operations 192 bit-level 384 operators precedence of 193 optimization 128 order B+ trees 343 of evaluation 192 OS accessing from applications 376 output argument xxxix arguments 222 echoing to file or printer 436 formatted to string variable 315 P parameters input 443 parser 321 parsing 174 by different lists 171 pathname in include files 430 pattern matcher xviii peeking 389

performed 137 place 227 pointers B+ trees (internal) 358 stack 445 poking 389 post-actions 122 predicate logic xviii predicates xxx arity 419 as arguments 238 C functions 444 declaring as deterministic 420 equal 201 global 439 names xliv section xliii, 419 specially handled 413 Predicates 212 preventing backtracking lxxxiv printermenu compiler directive 436 procedural perspective xciii program sections 413 program structure restrictions on 414 programming efficiency 267 style 267 programs 438 different versions of same 429 large 438 sections domains xlvi project compiler directive 436 projects 438 error level in 435 Prolog predicate logic syntax xviii Q queries xxi questions xxi quotation marks 422 R

random 195 random numbers generating 194 randominit 195 readblock 284 readchar 281 readdevice 150, 288 reading user-edited files 255 readint 280 readln 280 readreal 281 readterm 255, 281, 306 readtermerror 256 real 103 converting to string 319 random 195 recursion from a procedural viewpoint 166 recursive data structures 138 procedures 126 red cuts lxxxiv ref 329 ref_term 339 reference domains 226 trail array and 227 numbers 329 variable 226 variables 227 reg 227 reg domain 388 registers preserving 458, 462 relational operators 200 relations xix removing backtrack points 265 facts at run time 181 several facts at once 183 renamefile 295 repeat 124

repeat...fail 438 repetitive processes 120 replacing terms 338 reserved words 413 restrictions symbolic constants 428 restrictions to using database predicates 177 results 126 retract 181 retractall 183 RetryCount 366 return values registers for 443 round 199 rounding 196, 199 route planning example 396 rules xx as procedures xciii syntax lvii using like case statements xciv run-time errors 250 S samekey_next 359 samekey_prev 359 save 186, 189, 255 saving facts at run time 186 scanner 320 scanning 174 scope constant identifiers 428 predicates 433 Scopes 214 search database for record 343 searchchar 316 searchfile 294 searching lxvii searchstring 317 selectors external databases 326

sentence structure 115 set 167 setentry binary access 247 setting cuts 431 sharing modes 296 short 423 signal 258 simple constants 422 sin 197 sizeof function 453 sleep 380 SleepPeriod 366 solutions controlling the search lxxxi finding all at once 168 sorting with reference domains 231 sound 382 sqrt 199 square roots 196, 199 standard domains 415 storage (predicate) 383 str_char 318 str_int 318 str_len 315 str_real 319 strings 103, 205, 423 blank spaces 315 building 312 concatenation 313, 314 creating blank 312 dividing 311 internal address 389 length 315 manipulations 311 returning 312 verifying 312 struct compiler directive 448 structure external databases 360 structures

passing to C 448 subcomponents 425 substring 316 subtraction 192 symbol 436 symbolic constants lix symbolic constants 412, 423 symbols 103, 205 syspath 380 T tab 422 table 423 tail recursion 160 optimization 128 tan 197 telephone directory lv term_bin 249 term_delete 338 term_str 319 termination l terms 424 binary conversion of 248 functorless 448 manipulating 338 replacing 338 tests external database system external 339 text files external databases dump 360 timeout 381 to 217 transcendental functions 196 trap 251 traversing trees 141 tree-based sorting 147 trees as data type 139 creating 143 trigonometry 196 trunc 199

truncation 196, 199 type conversion 318 type conversion 266 type implementation 447 type variants 419 typing arguments in predicate declarations liv U ulong 424 underscore symbol xxxiv unification lxv, lxvi of compound objects 105 unsigned 424 updating external databases 354 facts section 179 upper_lower 319

ushort 424 V values 197 variables xxi, xxx, lxv, 425 anonymous xxxiii, 425 constants and 428 efficient programming with 267 verifying the length of a string 312 version OS 383 W word 424 write 272 write_a_list 401 writeblock 284 writedevice 150, 288 writef 277 title Visual Prolog xlii