--- language: prolog filename: learnprolog.pl contributors: - ["hyphz", "http://github.com/hyphz/"] --- Prolog is a logic programming language first specified in 1972, and refined into multiple modern implementations. ``` % This is a comment. % Prolog treats code entered in interactive mode differently % to code entered in a file and loaded ("consulted"). % This code must be loaded from a file to work as intended. % Lines that begin with ?- can be typed in interactive mode. % A bunch of errors and warnings will trigger when you load this file % due to the examples which are supposed to fail - they can be safely % ignored. % Output is based on SWI-prolog 7.2.3. Different Prologs may behave % differently. % Prolog is based on the ideal of logic programming. % A subprogram (called a predicate) represents a state of the world. % A command (called a goal) tells Prolog to make that state of the world % come true, if possible. % As an example, here is a definition of the simplest kind of predicate: % a fact. magicNumber(7). magicNumber(9). magicNumber(42). % This introduces magicNumber as a predicate and says that it is true % with parameter 7, 9, or 42, but no other parameter. Note that % predicate names must start with lower case letters. We can now use % interactive mode to ask if it is true for different values: ?- magicNumber(7). % True ?- magicNumber(8). % False ?- magicNumber(9). % True % Some older Prologs may display "Yes" and "No" instead of True and % False. % What makes Prolog unusual is that we can also tell Prolog to _make_ % magicNumber true, by passing it an undefined variable. Any name % starting with a capital letter is a variable in Prolog. ?- magicNumber(Presto). % Presto = 7 ; % Presto = 9 ; % Presto = 42. % Prolog makes magicNumber true by assigning one of the valid numbers to % the undefined variable Presto. By default it assigns the first one, 7. % By pressing ; in interactive mode you can reject that solution and % force it to assign the next one, 9. Pressing ; again forces it to try % the last one, 42, after which it no longer accepts input because this % is the last solution. You can accept an earlier solution by pressing . % instead of ;. % This is Prolog's central operation: unification. Unification is % essentially a combination of assignment and equality! It works as % follows: % If both sides are bound (ie, defined), check equality. % If one side is free (ie, undefined), assign to match the other side. % If both sides are free, the assignment is remembered. With some luck, % one of the two sides will eventually be bound, but this isn't % necessary. % % The = sign in Prolog represents unification, so: ?- 2 = 3. % False - equality test ?- X = 3. % X = 3 - assignment ?- X = 2, X = Y. % X = Y = 2 - two assignments % Note Y is assigned too, even though it is % on the right hand side, because it is free ?- X = 3, X = 2. % False % First acts as assignment and binds X=3 % Second acts as equality because X is bound % Since 3 does not equal 2, gives False % Thus in Prolog variables are immutable ?- X = 3+2. % X = 3+2 - unification can't do arithmetic ?- X is 3+2. % X = 5 - "is" does arithmetic. ?- 5 = X+2. % This is why = can't do arithmetic - % because Prolog can't solve equations ?- 5 is X+2. % Error. Unlike =, the right hand side of IS % must always be bound, thus guaranteeing % no attempt to solve an equation. ?- X = Y, X = 2, Z is Y + 3. % X = Y, Y = 2, Z = 5. % X = Y are both free, so Prolog remembers % it. Therefore assigning X will also % assign Y. % Any unification, and thus any predicate in Prolog, can either: % Succeed (return True) without changing anything, % because an equality-style unification was true % Succeed (return True) and bind one or more variables in the process, % because an assignment-style unification was made true % or Fail (return False) % because an equality-style unification was false % (Failure can never bind variables) % The ideal of being able to give any predicate as a goal and have it % made true is not always possible, but can be worked toward. For % example, Prolog has a built in predicate plus which represents % arithmetic addition but can reverse simple additions. ?- plus(1, 2, 3). % True ?- plus(1, 2, X). % X = 3 because 1+2 = X. ?- plus(1, X, 3). % X = 2 because 1+X = 3. ?- plus(X, 2, 3). % X = 1 because X+2 = 3. ?- plus(X, 5, Y). % Error - although this could be solved, % the number of solutions is infinite, % which most predicates try to avoid. % When a predicate such as magicNumber can give several solutions, the % overall compound goal including it may have several solutions too. ?- magicNumber(X), plus(X,Y,100). % X = 7, Y = 93 ; % X = 9, Y = 91 ; % X = 42, Y = 58 . % Note: on this occasion it works to pass two variables to plus because % only Y is free (X is bound by magicNumber). % However, if one of the goals is fully bound and thus acts as a test, % then solutions which fail the test are rejected. ?- magicNumber(X), X > 40. % X = 42 ?- magicNumber(X), X > 100. % False % To see how Prolog actually handles this, let's introduce the print % predicate. Print always succeeds, never binds any variables, and % prints out its parameter as a side effect. ?- print("Hello"). % "Hello" true. ?- X = 2, print(X). % 2 true. ?- X = 2, print(X), X = 3. % 2 false - print happens immediately when % it is encountered, even though the overall % compound goal fails (because 2 != 3, % see the example above). % By using Print we can see what actually happens when we give a % compound goal including a test that sometimes fails. ?- magicNumber(X), print(X), X > 40. % 7 9 42 X = 42 . % MagicNumber(X) unifies X with its first possibility, 7. % Print(X) prints out 7. % X > 40 tests if 7 > 40. It is not, so it fails. % However, Prolog remembers that magicNumber(X) offered multiple % solutions. So it _backtracks_ to that point in the code to try % the next solution, X = 9. % Having backtracked it must work through the compound goal % again from that point including the Print(X). So Print(X) prints out % 9. % X > 40 tests if 9 > 40 and fails again. % Prolog remembers that magicNumber(X) still has solutions and % backtracks. Now X = 42. % It works through the Print(X) again and prints 42. % X > 40 tests if 42 > 40 and succeeds so the result bound to X % The same backtracking process is used when you reject a result at % the interactive prompt by pressing ;, for example: ?- magicNumber(X), print(X), X > 8. % 7 9 X = 9 ; % 42 X = 42. % As you saw above we can define our own simple predicates as facts. % More complex predicates are defined as rules, like this: nearby(X,Y) :- X = Y. nearby(X,Y) :- Y is X+1. nearby(X,Y) :- Y is X-1. % nearby(X,Y) is true if Y is X plus or minus 1. % However this predicate could be improved. Here's why: ?- nearby(2,3). % True ; False. % Because we have three possible definitions, Prolog sees this as 3 % possibilities. X = Y fails, so Y is X+1 is then tried and succeeds, % giving the True answer. But Prolog still remembers there are more % possibilities for nearby() (in Prolog terminology, "it has a % choice point") even though "Y is X-1" is doomed to fail, and gives us % the option of rejecting the True answer, which doesn't make a whole % lot of sense. ?- nearby(4, X). % X = 4 ; % X = 5 ; % X = 3. Great, this works ?- nearby(X, 4). % X = 4 ; % error % After rejecting X = 4 prolog backtracks and tries "Y is X+1" which is % "4 is X+1" after substitution of parameters. But as we know from above % "is" requires its argument to be fully instantiated and it is not, so % an error occurs. % One way to solve the first problem is to use a construct called the % cut, !, which does nothing but which cannot be backtracked past. nearbychk(X,Y) :- X = Y, !. nearbychk(X,Y) :- Y is X+1, !. nearbychk(X,Y) :- Y is X-1. % This solves the first problem: ?- nearbychk(2,3). % True. % But unfortunately it has consequences: ?- nearbychk(2,X). % X = 2. % Because Prolog cannot backtrack past the cut after X = Y, it cannot % try the possibilities "Y is X+1" and "Y is X-1", so it only generates % one solution when there should be 3. % However if our only interest is in checking if numbers are nearby, % this may be all we need, thus the name nearbychk. % This structure is used in Prolog itself from time to time (for example % in list membership). % To solve the second problem we can use built-in predicates in Prolog % to verify if a parameter is bound or free and adjust our calculations % appropriately. nearby2(X,Y) :- nonvar(X), X = Y. nearby2(X,Y) :- nonvar(X), Y is X+1. nearby2(X,Y) :- nonvar(X), Y is X-1. nearby2(X,Y) :- var(X), nonvar(Y), nearby2(Y,X). % We can combine this with a cut in the case where both variables are % bound, to solve both problems. nearby3(X,Y) :- nonvar(X), nonvar(Y), nearby2(X,Y), !. nearby3(X,Y) :- nearby2(X,Y). % However when writing a predicate it is not normally necessary to go to % these lengths to perfectly support every possible parameter % combination. It suffices to support parameter combinations we need to % use in the program. It is a good idea to document which combinations % are supported. In regular Prolog this is informally in structured % comments, but in some Prolog variants like Visual Prolog and Mercury % this is mandatory and checked by the compiler. % Here is the structured comment declaration for nearby3: %! nearby3(+X:Int, +Y:Int) is semideterministic. %! nearby3(+X:Int, -Y:Int) is multi. %! nearby3(-X:Int, +Y:Int) is multi. % For each variable we list a type. The + or - before the variable name % indicates if the parameter is bound (+) or free (-). The word after % "is" describes the behaviour of the predicate: % semideterministic - can succeed once or fail % ( Two specific numbers are either nearby or not ) % multi - can succeed multiple times but cannot fail % ( One number surely has at least 3 nearby numbers ) % Other possibilities are: % det - always succeeds exactly once (eg, print) % nondet - can succeed multiple times or fail. % In Prolog these are just structured comments and strictly informal but % extremely useful. % An unusual feature of Prolog is its support for atoms. Atoms are % essentially members of an enumerated type that are created on demand % whenever an unquoted non variable value is used. For example: character(batman). % Creates atom value batman character(robin). % Creates atom value robin character(joker). % Creates atom value joker character(darthVader). % Creates atom value darthVader ?- batman = batman. % True - Once created value is reused ?- batman = batMan. % False - atoms are case sensitive ?- batman = darthVader. % False - atoms are distinct % Atoms are popular in examples but were created on the assumption that % Prolog would be used interactively by end users - they are less % useful for modern applications and some Prolog variants abolish them % completely. However they can be very useful internally. % Loops in Prolog are classically written using recursion. % Note that below, writeln is used instead of print because print is % intended for debugging. %! countTo(+X:Int) is deterministic. %! countUpTo(+Value:Int, +Limit:Int) is deterministic. countTo(X) :- countUpTo(1,X). countUpTo(Value, Limit) :- Value = Limit, writeln(Value), !. countUpTo(Value, Limit) :- Value \= Limit, writeln(Value), NextValue is Value+1, countUpTo(NextValue, Limit). ?- countTo(10). % Outputs 1 to 10 % Note the use of multiple declarations in countUpTo to create an % IF test. If Value = Limit fails the second declaration is run. % There is also a more elegant syntax. %! countUpTo2(+Value:Int, +Limit:Int) is deterministic. countUpTo2(Value, Limit) :- writeln(Value), Value = Limit -> true ; ( NextValue is Value+1, countUpTo2(NextValue, Limit)). ?- countUpTo2(1,10). % Outputs 1 to 10 % If a predicate returns multiple times it is often useful to loop % through all the values it returns. Older Prologs used a hideous syntax % called a "failure-driven loop" to do this, but newer ones use a higher % order function. %! countTo2(+X:Int) is deterministic. countTo2(X) :- forall(between(1,X,Y),writeln(Y)). ?- countTo2(10). % Outputs 1 to 10 % Lists are given in square brackets. Use memberchk to check membership. % A group is safe if it doesn't include Joker or does include Batman. %! safe(Group:list(atom)) is deterministic. safe(Group) :- memberchk(joker, Group) -> memberchk(batman, Group) ; true. ?- safe([robin]). % True ?- safe([joker]). % False ?- safe([joker, batman]). % True % The member predicate works like memberchk if both arguments are bound, % but can accept free variables and thus can be used to loop through % lists. ?- member(X, [1,2,3]). % X = 1 ; X = 2 ; X = 3 . ?- forall(member(X,[1,2,3]), (Y is X+1, writeln(Y))). % 2 3 4 % The maplist function can be used to generate lists based on other % lists. Note that the output list is a free variable, causing an % undefined value to be passed to plus, which is then bound by % unification. Also notice the use of currying on the plus predicate - % it's a 3 argument predicate, but we specify only the first, because % the second and third are filled in by maplist. ?- maplist(plus(1), [2,3,4], Output). % Output = [3, 4, 5]. ``` ##Ready For More? * [SWI-Prolog](http://www.swi-prolog.org/)