Module qualifications not resolved at runtime

[bakaq]

It seems that module qualifications are resolved at compile-time¹, and variable module qualifications are just ignored:

%% a.pl
:- module(a, []).

a(1).

?- use_module(a).
   true.
?- [user].
a(2).

?- a:a(A).
   A = 1.
?- a(A).
   A = 2.
?- Module = a, Module:a(A).
   Module = a, A = 2, unexpected.
   Module = a, A = 1. % expected, but not found

This quite limits the kind of things you can do with that, and makes this pattern currently used in the test suite useless, because only the tests visible in the unqualified module are found, instead of only the tests visible in the qualified module (which is the intended behavior).

I found this while adapting that test framework into something to be provided by the Scryer Prolog standard library.

Footnotes

1. Actually, I think at expansion-time, because if I remember correctly that is when module qualifications are dealt with. I think I even remember seeing the specific code that ignores variables once. ↩

[triska]

Regarding the footnote, maybe #2811?

[triska]

The WAM instructions look sensible, so there is hope that the issue is limited to the interpreter:

For instance, when I define:

p(M) :-
        M:p.

I get:

?- wam_instructions(p/1, Is), maplist(portray_clause, Is).
dynamic_else(1,inf,0).
put_structure(:,2,x(1)).
set_local_value(x(1)).
set_constant(p).
execute(call,1).

It seems to correctly call (:)/2.

But, I unexpectedly get:

?- p(a).
   error(type_error(atom,... :a),call/0), unexpected.

[UWN]

Even more so:

?- Module = a, call(Module:a(X)).
   Module = a, X = 2, unexpected.
   Module = a, X = 1. % expected, but not found
?- Module = a, G_0=Module:a(X), G_0.
   Module = a, G_0 = a:a(1), X = 1. % expected

So there is some premature expansion of call/1.

[bakaq]

This seems to be a good workaround for now:

?- Module = a, G0 = Module:a(A), call(G0).
   Module = a, G0 = a:a(1), A = 1.