Related
Suppose I have a predicate foo/2 which defines a relation between its first and second argument.
What is the most idiomatic and efficient way to change the implementation of foo/2 such that:
if both of its arguments are ground, it acts as before (succeeds if the relation holds, fails otherwise).
if one of the two arguments (or both) are free, it "constrains" those two arguments so that when they will get grounded, the relation will be checked.
In other words, how to correctly implement the behaviour exhibited by dif/2 but with any kind of user-defined relation?
listing(dif/2). was of little help.
Different Prolog implementations provide different features to accomplish this. The mechanism is variously known as coroutining, delayed goals, constraints, and your Prolog system's manual will provide more information.
Here are two variants, which are available in SICStus Prolog and also some other systems.
block/1 directive
In SICStus Prolog (and possibly some other systems), one way to lift a user-defined predicate to such a constrained version is available via the declarative block declaration.
Interestingly, this does not require any changes to the predicate itself!
Suppose you have an impure version of dif/2, using the non-monotonic (\=)/2 predicate:
madif(X, Y) :-
X \= Y.
Then you can turn it into a delayed version for example with:
:- block madif(-, ?),
madif(?, -).
madif(X, Y) :-
X \= Y.
Sample queries and answers:
| ?- madif(a, b).
yes
| ?- madif(a, X).
user:madif(a,X) ? ;
no
| ?- madif(a, X), X = b.
X = b ? ;
no
| ?- madif(X, Y).
user:madif(X,Y) ? ;
no
As required, the evaluation of the goal is delayed until both arguments are instantiated.
when/2
A second way to accomplish this with SICStus Prolog (and other systems that provide this feature) is to use when/2. This requires changes to the predicate itself.
For example, using when/2, you can implement madif/2 like this:
madif(X, Y) :-
when((ground(X),
ground(Y)), X \= Y).
Sample queries and answers:
| ?- madif(X, a).
prolog:trig_ground(X,[],[X],_A,_A),
prolog:when(_A,(ground(X),ground(a)),user:(X\=a)) ? ;
no
| ?- madif(X, a), X = b.
X = b ? ;
no
First and foremostly,
Take the user's viewpoint
... and not that of an implementer. All too often this is ignored – also in existing constraint implementations. And it shows. So here are the most salient aspects to take into account.
Correctness
Obviously this should hold. It is always better to produce clean errors, mostly instantiation errors, better to flounder forever, even better to loop forever than to fail incorrectly. If all else breaks you can wrap your attempt with freeze(_, G_0). Note that you do need a working toplevel to actually see such floundering goals. SICStus has such a toplevel1, in SWI you need to wrap your query as call_residue_vars(Query_0, Vs) to see all attached constraints.
Consistency
Next you want to ensure that your constraint ensures consistency as much as possible. There are many notions of consistency like, domain and bounds consistency. To take your precise requirement think of difgrn/2 and compare it to the built-in dif/2:
difgrn(X, Y) :-
when((ground(X), ground(Y)), X \== Y).
| ?- difgrn(X, X).
prolog:trig_ground(X,[],[X],_A,_B),
prolog:trig_ground(X,[],[X],_A,_C),
prolog:trig_and(_C,[],_A,_B,_A),
prolog:when(_A,(ground(X),ground(X)),user:(X\==X)) ? ;
no
| ?- dif(X, X).
no
| ?- difgrn([], [_]).
prolog:trig_ground(_A,[],[_A],_B,_C),
prolog:trig_and(_C,[],_B,1,_B),
prolog:when(_B,(ground([]),ground([_A])),user:([]\==[_A]))
| ?- dif([], [_]).
yes
One way to implement dif/2 in full strength is to use the very special condition (?=)/2:
difwh(X,Y) :- when(?=(X,Y), X\==Y).
which should answer your question as best as one can:
In other words, how to correctly implement the behaviour exhibited by dif/2 but with any kind of user-defined relation?
But unfortunately, this does not extend to anything else.
The situation becomes even more complex if one considers consistency between various constraints. Think of X in 1..2, dif(X, 1), dif(X, 2).
Answer projections
(For lack of a better word.) Sometimes you want to see your constraints nicely on the toplevel - and the best way is to represent them as goals that themselves will reestablish the exact state required to represent an answer.
See above trig_ground answers, which certainly could be beautified a bit.
Variable projections
Same as answer projections but possible at any point in time, via frozen/2 or copy_term/3.
Subsumption checking
This is useful for diagnostic purposes and loop checks.
For purely syntactic terms, there is subsumes_term/2 which ignores constraints. A prerequisite to perform an effective test is to connect each involved variable to the actual constraint. Consider the goal freeze(X, Y = a) and imagine some subsumption checking with Y as an argument. If Y is no longer attached to the information (as it usually is with current implementations of freeze/2) you will come to the wrong conclusion that this Y subsumes b.
Note as for the actual example of dif/2, this was the very first constraint ever (1972, Prolog 0). A more elaborate description gives Michel van Caneghem in L'anatomie de Prolog, InterÉditions 1986 and Lee Naish in Papers about MU-Prolog.
1 Half-true. For library(clpfd) you need assert(clpfd:full_answer).
I have a predicate that may unify its arguments, for example:
foo(X) :- X = 42.
How can I tell if, while proving foo(X), unification changed X? For example, I would like to know if writeln(X), foo(X), writeln(X) would print the same value for X twice, without actually doing the printing.
My actual implementation of foo/1 is actually much more complex, so please don't suggest specific to the simplified version above. In my program, foo(X) simplifies X using unification, but foo(X) may need to be proven several times until all simplifications have been performed. I would like to be able to write a foohelper(X) predicate that invokes foo(X) until X stops being unified.
Assuming we have only syntactic unification - that is, no constraints:
:- meta_predicate(call_instantiated(0,?)).
call_instantiated(Goal_0, Instantiated) :-
copy_term(Goal_0, Copy_0),
Goal_0,
( subsumes_term(Goal_0, Copy_0) -> % succeeds iff equal u.t.r.
Instantiated = false
; Instantiated = true
).
Note that Goal_0 will or will not be further instantiated. The above subsumes_term/2 tests whether or not Goal_0 is now "more general" than Copy_0. Of course, it cannot be more general, so effectively that test tests whether or not the terms are identical up to renaming of variables.
Compared to using term_variables/2, as #PauloMoura has indicated, this may more may not be more efficient. It primarily depends on the efficiency of subsumes_term/2.
Maybe you can use the standard term_variables/2 predicate? You can call it with your goal before and after calling the goal and check if the returned lists of variables are different. Something like:
...,
term_variables(foo(X), Vars0),
foo(X),
term_variables(foo(X), Vars),
( Vars0 == Vars ->
write(simplified)
; write(not_simplified)
),
...
I am looking for a reliable way to pass an implicit parameter among predicates, e.g.,
p(Context, Args) :- goal(Args).
where goal(Args) should be expanded to something like interpret(Context, goal, Args).
However, since I don't know whether a term will be evaluated as a goal or used as data, I'd like to attach Context to goal as extra data without modifying its actual structure.
Is there any way to do this?
Also, I'd need some way to hook into the evaluation of a term, then fetch its context and call interpret.
Any Ideas?
I am using SWI Prolog, a portable solution would be fine, but is not required.
Edit:
In pseudocode, what I am roughly looking for is the following:
term_expansion((Head :- Body), (Head :- Body)) :-
arg(1, Head, Context),
forall T: CompoundTerm in Body =>
set_term_attribute(T, context, Context).
on_evaluate(T) :-
get_term_attribute(T, context, Context) -> interpret(Context, T) ;
call(T).
Check out the important predicates term_expansion/2 and goal_expansion/2.
They allow you to rewrite clauses at compilation time, and these or similar constructs are supported by all serious Prolog systems to provide an expansion mechanism for Prolog code.
Definite clause grammars (see dcg for more information) are often implemented using such mechanisms. In fact, to pass around a context, DCG notation is itself often already sufficient, so you may not even have to rewrite the code yourself in your use case.
See the DCG resources for more information.
Easy. Use a Logtalk parametric object. You can use this solution with twelve Prolog systems, including SWI-Prolog. For example:
:- object(foo(_Parameter1, _Parameter2)).
:- public(p1/1).
p1(Parameter1) :-
parameter(1, Parameter1).
:- public(p2/1).
p2(Parameter2) :-
parameter(2, Parameter2).
:- end_object.
?- foo(1,2)::p1(P1).
P1 = 1.
?- foo(1,2)::p2(P2).
P2 = 2.
The parameters are logical variables, with all the benefits and potential that entails. The parameters are implicitly shared by all object predicate clauses. In SWI-Prolog, you can even use dicts for parameter values.
It's also easy to provide default parameter values. For example:
:- object(foo, extends(foo(a,b))).
:- end_object.
?- foo::p1(P1).
P1 = a.
?- foo::p2(P2).
P2 = b.
You can find several examples of using parametric objects in the Logtalk distribution.
I'm new to prolog; I'm coming from a structured programming background, as will become obvious :)
I am building up a prolog query that involves reversing a number; eg. reverse_num(123,X) results in X = 321. I came up with the following definition, but it only works when I provide a number as the first parameter.
reverse_num(Num, Revnum) :-
number_chars(Num, Atoms),
reverse(Revatoms, Atoms),
number_chars(Reversed, Revatoms),
Reversed = Revnum.
the number_chars/2 predicate doesn't like an unsubstantiated variable if I do: reverse_num(X,123) (where I'm expecting X to be 321).
Am I trying too hard to make reverse_num do something it shouldn't (should it be understood to work only with a number as the first parameter and variable as the second)?
Or is there an easy / straight-forward way to handle a variable as the first parameter?
Relational naming
Before jumping into coding, let's take a step back. After all, the idea in Prolog is to define relations. Your name reverse_num/2 rather suggests some actions, num_reversed/2 might be a better name.
Determine the relation
Your definition is not that bad, let me rewrite it to1:
num_reversed(Num, Reversed) :-
number_chars(Num, Chars),
reverse(Chars, Revchars),
number_chars(Reversed, Revchars).
?- num_reversed(123,X).
X = 321.
?- num_reversed(1230,X).
X = 321.
?- num_reversed(12300,X).
X = 321.
Do you see the pattern? All numbers N*10^I have the same result!
Now, let's ask some more:
?- num_reversed(Num, 321).
error(instantiation_error,number_chars/2).
Hm, what did we expect? Actually, we wanted all 123*10^I to be printed. That's infinitely many solutions. So above query, if correctly answered, would require infinitely many solutions to be printed. If we print them directly, that will take all our universe's lifetime, and more!
It is for this reason, that Prolog produces an instantiation error instead. By this, Prolog essentially states:
This goal is too general that I can make a good answer. Maybe there are infinitely many solutions, maybe not. I know not. But at least I indicate this by issuing an error. To remove this error you need to instantiate the arguments a bit more.
So the answer Prolog produced was not that bad at all! In fact, it is much better to produce a clean error than to, say, fail incorrectly. In general, Prolog's errors are often a very useful hint to what semantic problems you might have. See all error classes how.
Coroutining
As have other answers suggested, coroutining, using when/2 might solve this problem. However, coroutining itself has many semantic problems. Not without reason, systems like XSB do not offer it, due to the many problems related to subsumption checking. An implementation that would be compatible to it would be unexpectedly inefficient.
But for the sake of the point, we could make our definition more versatile by querying it like
?- when(nonvar(Num), num_reversed(Num, Reversed)).
when(nonvar(Num), num_reversed(Num, Reversed)).
Now we get back as an answer exactly the query we entered. This is also known as floundering. So there is a way to represent infinitely may solutions in a compact manner! However, this comes at a rather high price: You no longer know whether a solution exists or not. Think of:
?- when(nonvar(Num), num_reversed(Num, -1)).
when(nonvar(Num), num_reversed(Num, -1)).
Others have suggested to wait also for nonvar(Reversed) which would only be correct if we would produce infinitely many answers - but, as we have seen - this just takes too much time.
Coroutining looked as a very promising road at the beginning of the 1980s. However, it has never really caught on as a general programming methodology. Most of the time you get much too much floundering which is just a pain and even more difficult to handle than, say instantiation errors.
However, a more promising offspring of this development are constraints. There, the mechanisms are much cleaner defined. For practical purposes, programmers will only use existing libraries, like CLPFD, CLPQ, or CHR. Implementing your own library is an extremely non-trivial project in its own right. In fact it might even be possible to provide an implementation of num_reversed/2 using library(clpfd) that is, restricting the relation to the integer case.
Mode dependent conditionals
Traditionally, many such problems are solved by testing for instantiations explicitly. It is good style to perform this exclusively with nonvar/1 and ground/1 like the condition in when/2- other type test predicates lead easily to errors as exemplified by another answer.
num_reversed(Num, Reversed) :-
( nonvar(Num)
-> original_num_reversed(Num, Reversed)
; original_num_reversed(Reversed, Base),
( Base =:= 0
-> Num is 0
; length(_, I),
Num is Base*10^I
)
).
Above code breaks very soon for floats using base 2 and somewhat later for base 10. In fact, with classical base 2 floats, the relation itself does not make much sense.
As for the definition of number_chars/2, ISO/IEC 13211-1:1995 has the following template and mode subclause:
8.16.7.2 Template and modes
number_chars(+number, ?character_list)
number_chars(-number, +character_list)
The first case is when the first argument is instantiated (thus nonvar). The second case, when the first argument is not instantiated. In that case, the second argument has to be instantiated.
Note, however, that due to very similar problems, number_chars/2 is not a relation. As example, Chs = ['0','0'], number_chars(0, Chs) succeeds, whereas number_chars(0, Chs), Chs = ['0','0'] fails.
Very fine print
1 This rewrite is necessary, because in many Prologs reverse/2 only terminates if the first argument is known. And in SWI this rewrite is necessary due to some idiosyncratic inefficiencies.
The number_chars/2 predicate has the signature:
number_chars(?Number, ?CharList)
But although not fully specified by the signature, at least Number or CharList have to be instantiated. That's where the error occurs from.
If you call:
reverse_num(Num,123)
You will call number_chars/2 with both uninstatiated at that time so the predicate will error.
A not very nice solution to the problem is to ask whether Num or RevNum are number/2s. You can do this by writing two versions. It will furthermore filter other calls like reverse_num(f(a),b), etc.:
reverse_num(Num,Revnum) :-
\+ number(Num),
\+ number(Revnum),
throw(error(instantiation_error, _)).
reverse_num(Num, Revnum) :-
ground(Num),
number(Num),
!,
number_chars(Num, Atoms),
reverse(Revatoms, Atoms),
number_chars(Revnum, Revatoms).
reverse_num(Num, Revnum) :-
ground(Revnum),
number(Revnum),
reverse_num(Revnum,Num).
Or you can in case you use two nongrounds (e.g. reverse_num(X,Y).) an instantiation error instead of false as #false says:
reverse_num(Num,Revnum) :-
\+ number(Num),
\+ number(Revnum),
!,
throw(error(instantiation_error, _)).
reverse_num(Num, Revnum) :-
number(Num),
!,
number_chars(Num, Atoms),
reverse(Revatoms, Atoms),
number_chars(Revnum, Revatoms).
reverse_num(Num, Revnum) :-
reverse_num(Revnum,Num).
The cut (!) is not behaviorally necessary, but will increase performance a bit. I'm not really a fan of this implementation, but Prolog cannot always fully make predicates reversible since (a) reversibility is an undecidable property because Prolog is Turing complete; and (b) one of the characteristics of Prolog is that the body atoms are evaluated left-to-right. otherwise it will take ages to evaluate some programs. There are logic engines that can do this in an arbitrary order and thus will succeed for this task.
If the predicate/2 is commutative, a solution that can be generalized is the following pattern:
predicate(X,Y) :-
predicate1(X,A),
predicate2(A,B),
% ...
predicaten(C,Y).
predicate(X,Y) :-
predicate(Y,X).
But you cannot simply add the last clause to the theory, because it can loop infinitely.
Nice to see someone is also worried about define flexible rules with no restrictions in the set of bound arguments.
If using a Prolog system that supports coroutining and the when/2 built-in predicate (e.g. SICStus Prolog, SWI-Prolog, or YAP), try as:
reverse_num(Num, Reversed) :-
when( ( ground(Num); ground(Atoms) ), number_chars(Num, Atoms) ),
when( ( ground(Reversed); ground(Revatoms) ), number_chars(Reversed, Revatoms) ),
reverse(Atoms , Revatoms).
that gives:
?- reverse_num( 123, X ).
X = 321.
?- reverse_num( X, 123 ).
X = 321 .
( thanks to persons who provided theses answers: Prolog: missing feature? )
This SWISH session shows my effort to answer.
Then I've come back here, where I found I was on #PasabaPorAqui' mood (+1), but I didn't get it right.
But, such an interesting topic: notice how regular is the join pattern.
reverse_num(X, Y) :-
when((nonvar(Xs);nonvar(Ys)), reverse(Xs, Ys)),
when((nonvar(X) ;nonvar(Xs)), atomic_chars(X, Xs)),
when((nonvar(Y) ;nonvar(Ys)), atomic_chars(Y, Ys)).
So, we can generalize in a simple way (after accounting for PasabaPorAqui correction, ground/1 it's the key):
% generalized... thanks Pasaba Por Aqui
:- meta_predicate when_2(0).
when_2(P) :-
strip_module(P,_,Q),
Q =.. [_,A0,A1],
when((ground(A0);ground(A1)), P).
reverse_num(X, Y) :-
maplist(when_2, [reverse(Xs, Ys), atomic_chars(X, Xs), atomic_chars(Y, Ys)]).
I think I understand why nonvar/1 was problematic: the list bound for reverse get 'fired' too early, when just the head get bound... too fast !
maplist/2 is not really necessary: by hand we can write
reverse_num(X, Y) :-
when_2(reverse(Xs, Ys)),
when_2(atomic_chars(X, Xs)),
when_2(atomic_chars(Y, Ys)).
this seems an ideal application of term rewriting... what do you think about -:- ? Implementing that we could write bidirectional code like
reverse_num(X, Y) -:-
reverse(Xs, Ys),
atomic_chars(X, Xs),
atomic_chars(Y, Ys).
edit SWISH maybe is not 'term_rewrite' friendly... so here is a lower level approach:
:- op(900, xfy, ++).
A ++ B ++ C :- when_2(A), B ++ C.
A ++ B :- when_2(A), when_2(B).
reverse_num(X, Y) :-
reverse(Xs, Ys) ++ atomic_chars(X, Xs) ++ atomic_chars(Y, Ys).
Setting aside the problem of trailing zeroes turning into leading zeroes, it doesn't seem like it should be much more complicated than something like this (made somewhat more complicated by dealing with negative numbers):
reverse_number(X,Y) :- number(X) , ! , rev(X,Y) .
reverse_number(X,Y) :- number(Y) , ! , rev(Y,X) .
rev(N,R) :-
N < 0 ,
! ,
A is abs(N) ,
rev(A,T) ,
R is - T
.
rev(N,R) :-
number_chars(N,Ns) ,
reverse(Ns,Rs) ,
number_chars(R,Rs)
.
Note that this does require at least one of the arguments to reverse_number/2 to be instantiated.
I have a database of facts like this:
li(a,2).
li(b,3).
li(b,1).
li(c,2).
li(d,1).
li(d,1).
I need to write a predicate more(+Let) that succeeds if it exists more than one fact li(Let,_).
For example the queries more(b) and more(d) will succeed, but more(a) and more(c) will not.
My idea was to check if li(Let,_) succeeds more than once, but I do not know how to do it.
Try findall/3:
findall(X, li(d,X), L), length(L,N), N>1.
Abstracting the d out and making a predicate is trivial. Right? :)
If you don't want to use any of the predicates like findall, you can change the representation of your knowledge - bring it down one level, so to speak:
my_knowledge(li, [a-2,b-3,b-1,c-2,d-1,d-1]).
and then you can use SWI Prolog's predicate select/3 to handle it:
select_knowledge(kn, key, R):-
my_knowledge(kn,L),
select_key(L,key,R).
select_key(L,K,R):-
select(K-X,L,L1) -> R=[X|R1], select_key(L1,K,R1)
; R = [].
You can rewrite the last predicate as basic recursion over lists, and then tweak it to stop after getting first N results.
more_than_once(Goal) :-
\+ \+ call_nth(Goal,2).
With call_nth/2 as defined in this answer.
The big advantage of this solution compared to the other solutions proposed is that it will succeed rapidly even if there is a very large sequence of answers. In fact, it will even succeed for an infinite sequence of answers:
?- more_than_once(repeat).
true.
?- more_than_once(between(1,100000,_)).
true.
(The implementation of call_nth/2 uses some non-standard, low-level built-ins of SWI. It is possible to avoid that, but with even more headache.)
SWI-Prolog has library(aggregate).
:- [library(aggregate)].
more(Key) :- aggregate_all(count, li(Key, _), C), C > 1.
test:
?- more(b).
true.
?- more(a).
false.
It's not very easy to learn, but useful to handle such common tasks. If you have a very large code base, then findall (and aggregate as well, that uses findall inside) could be inefficient, building a list only to count its elements.
Then you could use a side effect based predicate: in this related answer you'll find such utility. For max efficiency, see the comments, where is explained how to use nb_setval/nb_getval...