Bug in 8.4.1 regarding dict

Frank_Schwidom · December 13, 2021, 6:11pm

Hi,

the following code shows a bug:

X.crash() := VALUE :- true
 , aggregate_all( sum( M.value ), member( M, X.values), VALUE)
 .

X.workaround() := VALUE :- true
 , findall( V, ( member( M, X.values), V=M.value), VALUES)
 , aggregate_all( sum( M ), member( M, VALUES), VALUE)
 .

test001 :- true
 , VALUE = x{values:[p{value:1},p{value:2}]}.crash()
 , writeln(VALUE)
 .

test002 :- true
 , VALUE = x{values:[p{value:1},p{value:2}]}.workaround()
 , writeln(VALUE)
 .

and the output is:



?- test002.
3
true.

?- test001.
ERROR: Arguments are not sufficiently instantiated
ERROR: In:
ERROR:   [16] throw(error(instantiation_error,_9338))
ERROR:   [14] '.'('<garbage_collected>',value,'<garbage_collected>') at /usr/local/lib/swipl/boot/dicts.pl:46
ERROR:   [13] crash(x{values:[...|...]},_9392) at /home/ox/dev-git/go/endgame-subgames-calculation/003_bug_report.pl:2
ERROR:   [12] eval_dict_function('<garbage_collected>',x,'<garbage_collected>','<garbage_collected>') at /usr/local/lib/swipl/boot/dicts.pl:102
ERROR:   [10] test001 at /home/ox/dev-git/go/endgame-subgames-calculation/003_bug_report.pl:12
ERROR:    [9] toplevel_call('<garbage_collected>') at /usr/local/lib/swipl/boot/toplevel.pl:1117
ERROR: 
ERROR: Note: some frames are missing due to last-call optimization.
ERROR: Re-run your program in debug mode (:- debug.) to get more detail.

Thanks in Advance
Frank Schwidom

Frank_Schwidom · December 13, 2021, 10:05pm

By playing around I found out that already the unification of a variable can cause the error when inside of a term a field dereference of a dict takes place.


X.crash() := VALUE :- true
 , aggregate_all( sum( M.value ), member( M, X.values), VALUE)
 .

X.workaround() := VALUE :- true
 , findall( V, ( member( M, X.values), V=M.value), VALUES)
 , aggregate_all( sum( M ), member( M, VALUES), VALUE)
 .

X.freezes() := VALUE :- true
 , C= aggregate_all( sum( M.value ), member( M, X.values), VALUE)
 , C
 .

X.crash2() := VALUE :- true 
 , TERM = ( member( M, X.values), V=M.value)

 % , findall( V, TERM, VALUES)
 % , VALUE=VALUES

 , VALUE=1

 % , aggregate_all( sum( M ), member( M, VALUES), VALUE)

 % , aggregate_all( sum( V ), TERM, VALUE)
 .

X.crash3() := VALUE :- true 
 , XVALUES= X.values
 , TERM = ( member( M, XVALUES), V=M.value)

 % , findall( V, TERM, VALUES)

 , VALUE=1

 % , aggregate_all( sum( M ), member( M, VALUES), VALUE)

 % , aggregate_all( sum( V ), TERM, VALUE)
 .

test001 :- true
 , VALUE = x{values:[p{value:1},p{value:2}]}.crash()
 , writeln(VALUE)
 .

test002 :- true
 , VALUE = x{values:[p{value:1},p{value:2}]}.workaround()
 , writeln(VALUE)
 .

test003 :- true
 , VALUE = x{values:[p{value:1},p{value:2}]}.freezes()
 , writeln(VALUE)
 .

test004 :- true
 , VALUE = x{values:[p{value:1},p{value:2}]}.crash2()
 , writeln(VALUE)
 .

test005 :- true
 , VALUE = x{values:[p{value:1},p{value:2}]}.crash3()
 , writeln(VALUE)
 .

output:

(ins)?- test003. % freezes for a moment and has no output

(ins)?- test004.
ERROR: Arguments are not sufficiently instantiated
ERROR: In:
ERROR:   [16] throw(error(instantiation_error,_1984))
ERROR:   [13] crash2(x{values:[...|...]},_2010) at /home/ox/dev-git/go/endgame-subgames-calculation/003_bug_report.pl:16
ERROR:   [12] eval_dict_function(crash2(),x,x{values:[...|...]},_2054) at /usr/local/lib/swipl/boot/dicts.pl:102
ERROR:   [10] test004 at /home/ox/dev-git/go/endgame-subgames-calculation/003_bug_report.pl:58
ERROR:    [9] toplevel_call(user:user:test004) at /usr/local/lib/swipl/boot/toplevel.pl:1117
ERROR: 
ERROR: Note: some frames are missing due to last-call optimization.
ERROR: Re-run your program in debug mode (:- debug.) to get more detail.

(ins)?- test005.
ERROR: Arguments are not sufficiently instantiated
ERROR: In:
ERROR:   [16] throw(error(instantiation_error,_10796))
ERROR:   [13] crash3(x{values:[...|...]},_10822) at /home/ox/dev-git/go/endgame-subgames-calculation/003_bug_report.pl:29
ERROR:   [12] eval_dict_function(crash3(),x,x{values:[...|...]},_10866) at /usr/local/lib/swipl/boot/dicts.pl:102
ERROR:   [10] test005 at /home/ox/dev-git/go/endgame-subgames-calculation/003_bug_report.pl:63
ERROR:    [9] toplevel_call(user:user:test005) at /usr/local/lib/swipl/boot/toplevel.pl:1117
ERROR: 
ERROR: Note: some frames are missing due to last-call optimization.
ERROR: Re-run your program in debug mode (:- debug.) to get more detail.

Regards,
Frank

jan · December 14, 2021, 8:01am

By the simple rule of goal expansion this becomes

.(M, value, VM),
.(X, values, VX),
aggregate_all(sum(VM), member(M, VX), Value).

Which obviously gives the wrong result. I do not see a sensible way to tell the system to translate this into the intended

.(X, values, VX)
aggregate_all(sum(VM), (member(M, VX), .(M,value, VM)), Values).

Of course we could try additional rules. We’d need to know the 2nd argument is a goal. We know this time due to the meta_predicate/1 declaration for aggregate_all. We also need to know that for this case member/2 uses the second argument as input and the first as output. In general it is a relation though, so we cannot know.

I think it is better to keep the rewriting simple (and predictable) and let the user figure out to use a helper predicate to get the expansion order right.

Frank_Schwidom · December 15, 2021, 5:15pm

Ok I believe I understand. So I have to write it out.

X.nocrash() := VALUE :- true
 , aggregate_all( sum( X1 ), ( X3 = X.values, member( M, X3), X1 = M.value ), VALUE)
 .


test001 :- true
 , VALUE = x{values:[p{value:1},p{value:2}]}.nocrash()
 , writeln(VALUE)
 .

output :

(ins)?- test001.
3
true.

But there is still a problem:

when I load the code via ‘-s’ and run it I get the result:

(ins)$ swipl -s 004_expansion_problem.pl 
reading ~/.swiplrc (FS)
Welcome to SWI-Prolog (threaded, 64 bits, version 8.4.1)
SWI-Prolog comes with ABSOLUTELY NO WARRANTY. This is free software.
Please run ?- license. for legal details.

For online help and background, visit https://www.swi-prolog.org
For built-in help, use ?- help(Topic). or ?- apropos(Word).

(ins)?- test001.
ERROR: Arguments are not sufficiently instantiated
ERROR: In:
ERROR:   [16] throw(error(instantiation_error,_20638))
ERROR:   [14] '.'(_20662,value,_20666) at /usr/local/lib/swipl/boot/dicts.pl:46
ERROR:   [13] nocrash(x{values:[...|...]},_20692) at /home/ox/dev-git/swi-prolog/mapping_the_dot/001/004_expansion_problem.pl:2
ERROR:   [12] eval_dict_function(nocrash(),x,x{values:[...|...]},_20736) at /usr/local/lib/swipl/boot/dicts.pl:102
ERROR:   [10] test001 at /home/ox/dev-git/swi-prolog/mapping_the_dot/001/004_expansion_problem.pl:8
ERROR:    [9] toplevel_call(user:user:test001) at /usr/local/lib/swipl/boot/toplevel.pl:1117
ERROR: 
ERROR: Note: some frames are missing due to last-call optimization.
ERROR: Re-run your program in debug mode (:- debug.) to get more detail.

After just renaming X3 to X4 I get

(ins)?- make.
% /home/ox/dev-git/swi-prolog/mapping_the_dot/001/004_expansion_problem compiled 0.00 sec, 0 clauses
true.

(ins)?- test001.
3
true.

Regards,
Frank

Frank_Schwidom · December 15, 2021, 5:40pm

The last 2 attempts I had to rename the variable 2 times and call make to get the effect.

j4n_bur53 · December 15, 2021, 6:01pm

There is an auto-loading dependency. At least in SWI-Prolog 8.5.2 and
likely before. The auto-loading of aggregate_all results in loading of meta-
declaration, and then the expansion is different.

Without pre-loading aggregate_all I get:

?- [user].
X.crash() := VALUE :-  aggregate_all( sum( M.value ), member( M, X.values), VALUE).
X.crash2() := VALUE :-  aggregate_all( sum( H), (member( M, X.values), H = M.value), VALUE).

?- listing(crash).
crash(A, B) :-
    '.'(C, value, D),
    '.'(A, values, E),
    aggregate_all(sum(D), member(C, E), B),
    true.

true.

?- listing(crash2).
crash2(A, B) :-
    '.'(A, values, C),
    '.'(D, value, E),
    aggregate_all(sum(F),
                  ( member(D, C),
                    F=E
                  ),
                  B),
    true.

true.

With pre-loading aggregate all I get:

/* This causes the auto load */
?- aggregate_all(sum(X), (X=1;X=2), S).
S = 3.

?- [user].
X.crash() := VALUE :-  aggregate_all( sum( M.value ), member( M, X.values), VALUE).
X.crash2() := VALUE :-  aggregate_all( sum( H), (member( M, X.values), H = M.value), VALUE).

?- listing(crash).
crash(A, B) :-
    '.'(C, value, D),
    aggregate_all(sum(D),
                  ( '.'(A, values, E),
                    member(C, E)
                  ),
                  B),
    true.

true.

?- listing(crash2).
crash2(A, B) :-
    aggregate_all(sum(C),
                  ( '.'(A, values, D),
                    member(E, D),
                    '.'(E, value, F),
                    C=F
                  ),
                  B),
    true.

true.

Hope this helps!

Frank_Schwidom · December 15, 2021, 6:42pm

Ok, but now I have the next problem.

How can I avoid the need to export the nocrash predicate ?

% problem : nocrash has to be exported
:- module( test, [test001/0, nocrash/2 ]).

% :- aggregate_all( sum(X), member(X, [1,2]), _). % loads module aggregate

:- use_module(library(aggregate)).

X.nocrash() := VALUE :- true
 , aggregate_all( sum( X1 ), ( X5 = X.values, member( M, X5), X1 = M.value ), VALUE)
 .

test001 :- true
 , VALUE = x{values:[p{value:1},p{value:2}]}.nocrash()
 , writeln(VALUE)
 .

Frank_Schwidom · December 16, 2021, 12:16pm

Thanks for your answers so far. The topic is for me almost solved in the way that I have to load a specific library to avoid the wrong term rewriting order.

But I am not sure to keep the issue open because lt keeps opaque to me in which order which libraries are loaded and how the bug appears and how to avoid it in the future because the errormessage itself - if it happens - don’t really clarify the origin of the misbehaviour.

Maybe there should be library dependency description anywhere so the bug cannot happen anymore undependendly whether I have loaded a lib or it gets autoloaded.

Regards,
Frank

jan · December 17, 2021, 8:09am

The binding of functions on dicts is provided by the dict tag that is mapped to the module.

Possibly. Interaction between term/goal rewriting is typically hard. Several systems have partial solutions, but all pretty different. I’ve been working on a proper solution, but this project stalled due to lack of urgency and time.

Typically autoloading is problematic for meta-predicates and libraries involving term/goal rewriting. We could easily detect the first. The latter is a bit harder to do reliably

Topic		Replies	Views
Get_dict_ex/3 does not exist? Help!	4	623	July 17, 2020
Getting all key-value pairs out of a dict using bagof/3 fails, but succeeds with findall/3 Help!	9	434	July 20, 2020
Luxembourg National Day Code Drop: A replacement for must_be/2, a dict prettyprinter and assorted stuff General	1	428	July 3, 2021
Unit test failure on strings, inconsistent match fails?! General	15	348	June 8, 2020
PL-Unit problem with JSON terms Help!	4	640	October 1, 2020

Bug in 8.4.1 regarding dict

Related Topics