Different results with and without { true } in a DCG

[Kiyoshi364]

I have a DCG (cbor_minor_value//2), that in some cases only unify arguments,
and in another cases it immediately calls another DCG (numbytes_number//2) and return.

cbor_minor_value( 0, val(i,  0)) --> [].
cbor_minor_value( 1, val(i,  1)) --> [].
...
cbor_minor_value(23, val(i, 23)) --> [].
cbor_minor_value(24, val(x1, V)) --> numbytes_number(1, V).
cbor_minor_value(25, val(x2, V)) --> numbytes_number(2, V).
cbor_minor_value(26, val(x4, V)) --> numbytes_number(4, V).
cbor_minor_value(27, val(x8, V)) --> numbytes_number(8, V).

The predicate by itself works fine.
But when I call it from another predicate (cbor_item//1), it always fails at calling numbytes_number//2, even though a solution exists.
If we add any predicate (such as { true } or []), before or after calling numbytes_number//2, it starts to behave as expected.

I don't know why that is happening.
I use library(clpz) to implement numbytes_number//2.

There is the minimal file with a test case.
min-cbor.pl.txt
It is enough for main/0 to succeed.

The test file is a modified version of cbor.pl.
This problem was first mentioned in #2800.

My scryer-prolog version:

$ scryer-prolog --version
8ac663d

[hurufu]

This is surprising bug, I think there is a problem with attributed variables unification. For example if you rewrite last 4 cbor_minor_value//2 rules in the following way:

cbor_minor_value(24, val(x1, V), B, C) :- numbytes_number(1,V,B,C).
cbor_minor_value(25, val(x2, V), B, C) :- numbytes_number(2,V,B,C).
cbor_minor_value(26, val(x4, V), B, C) :- numbytes_number(4,V,B,C).
cbor_minor_value(27, val(x8, V), B, C) :- numbytes_number(8,V,B,C).

Them test fails, but if you rewrite them as:

cbor_minor_value(24, val(x1, V), B, C) :- B=D, numbytes_number(1,V,D,C).
cbor_minor_value(25, val(x2, V), B, C) :- B=D, numbytes_number(2,V,D,C).
cbor_minor_value(26, val(x4, V), B, C) :- B=D, numbytes_number(4,V,D,C).
cbor_minor_value(27, val(x8, V), B, C) :- B=D, numbytes_number(8,V,D,C).

Then suddenly all tests pass. 🤔

[UWN]

Please do narrow down the problem to the very smallest program possible. Here is what I obtained by stepwise reducing your program, removing everything unnecessary:

:- use_module(library(freeze)).

main1 :-
  freeze(Minor,true),
  cbor_minor_value1(Minor, []).

main2 :-
  freeze(Minor,true),
  cbor_minor_value2(Minor, []).

cbor_minor_value1(24, S0) :- numbytes_number(1, S0).

cbor_minor_value2(24, S0) :- S0=S1, numbytes_number(1, S1).

numbytes_number(_, []).

?- main1.
   false, unexpected.
   true. % expected, but not found
?- main2.
   true. % expected

So it is unrelated to clpz.

[Kiyoshi364]

Thanks!

I'm adding 3 tests more.
I hope they help.

main3 :-
  freeze(Minor,true),
  cbor_minor_value3(Minor, []).

main4 :-
  freeze(Minor,true),
  cbor_minor_value4(Minor, []).

main5 :-
  freeze(Minor,true),
  cbor_minor_value5(Minor, []).

cbor_minor_value3(24, S0) :- numbytes_number(1, S1), S0=S1.

cbor_minor_value4(24, S0) :- true, numbytes_number(1, S0).

cbor_minor_value5(24, S0) :- numbytes_number(1, S0), true.

numbytes_number(_, []).

/*
?- main3.
   true. % expected
?- main4.
   true. % expected
?- main5.
   false, unexpected.
   true. % expected, but not found
*/

[UWN]

Another version that shows that the module name of the attribute is involved:

:- use_module(library(freeze)).
   
main :-
  freeze(Minor,true),
  cbor_minor_value(Minor, 1, []).

cbor_minor_value(24, One, S) :- numbytes_number(1, One, S).

numbytes_number(_, One, _) :- arg(One,f(200),200).

?- main.
   error(type_error(integer,freeze),arg/3), unexpected.
   type_error(integer,freeze), unexpected. % no matter what the 2nd arg looks like
   true. % expected, but not found

[bakaq]

I just confirmed that this is the same problem as #2706.