SICStus Prolog: statistics/2 values affected by JIT

Question

Update: As reported by @PerMildner, the below issue (present with SICStus Prolog 4.3.2) has vanished with the new SICStus Prolog 4.3.3! Way to go!

---

I'm using SICStus Prolog 4.3.2 and I'm getting to know some of the statistics/2 keys that I haven't used before... Here's my code:

:- use_module(library(lists)).

a_is_b_minus_c(A,B,C) :-
   A is B-C.

call_keys_deltas(Goal_0, Keys, Deltas) :-
   maplist(statistics, Keys, Values0),
   call(Goal_0),
   maplist(statistics, Keys, Values1),
   maplist(a_is_b_minus_c, Deltas, Values1, Values0).

call_keys_deltas/3 gathers some information about the execution of predicates like:

boolsA([]).                        
boolsA([0|Bs]) :- boolsA(Bs).
boolsA([1|Bs]) :- boolsA(Bs).

%%

bool(0).
bool(1).

boolsB([]).
boolsB([B|Bs]) :- bool(B), boolsB(Bs).

I'm expecting two choicepoints for ?- boolsA([0,0]).—but none for ?- boolsB([0,0]).

The numbers I get immediately after the JIT ran differ from the ones I get once it is done:

• With JIT

  $ export SP_JIT=enabled && sicstus
  % ...
  | ?- call_keys_deltas(boolsA([0,0]), [choice_used,trail_used], Ds).
  Ds = [112,904] ? ;
  no
  | ?- call_keys_deltas(boolsA([0,0]), [choice_used,trail_used], Ds).
  Ds = [112,16] ? ;
  no
  | ?- call_keys_deltas(boolsB([0,0]), [choice_used,trail_used], Ds).
  Ds = [0,1264] ? ;
  no
  | ?- call_keys_deltas(boolsB([0,0]), [choice_used,trail_used], Ds).
  Ds = [0,8] ? ;
  no
  

• Without JIT

  $ export SP_JIT=disabled && sicstus
  % ...
  | ?- call_keys_deltas(boolsA([0,0]), [choice_used,trail_used], Ds).
  Ds = [112,16] ? ;
  no
  | ?- call_keys_deltas(boolsA([0,0]), [choice_used,trail_used], Ds).
  Ds = [112,16] ? ;
  no
  | ?- call_keys_deltas(boolsB([0,0]), [choice_used,trail_used], Ds).
  Ds = [0,8] ? ;
  no
  | ?- call_keys_deltas(boolsB([0,0]), [choice_used,trail_used], Ds).
  Ds = [0,8] ? ;
  no
  

I wonder if above discrepancy is intentional... Thank for your help!

Answer 1

The difference was not intentional. It should be gone in the latest release of SICStus Prolog (4.3.3).

Note that the choice_used did not differ, even before SICStus 4.3.3, and that statistic is probably what you should use in order to "quantify non-determinacy".