[m-rev.] for post-commit review: improve debug code in polymorphism_type_class_info.m

[Peter Wang]

On Thu, 10 Sep 2026 12:30:36 +0200 "Zoltan Somogyi" <zoltan.somogyi at runbox.com> wrote:
> This helped me to track down the source of one of the problems
> in the tci_liveness_abort test case: the creation of a stray base typeclass
> info. As the attached trace output file (for a cutdown version of that
> test case) shows,
> 
> - we generate the same value as the base typeclass info for both
>   foo(list(T)) and foo(list(list(T))), but
> 
> - we do not reuse the variable holding the first base typeclass info
>   when we need the second, because we look for them in different
>   "slots" in the typeclass info map.
> 
> This diff makes no algorithmic changes in non-debug code; that is
> for later diffs.
> 
> For review by anyone.

> diff --git a/compiler/const_struct.m b/compiler/const_struct.m
> index ba56d51ae..e89b2ad1a 100644
> --- a/compiler/const_struct.m
> +++ b/compiler/const_struct.m
> @@ -70,7 +70,18 @@
>                  % The constraint.
>                  prog_constraint,
>  
> -                % The already seen constraints.
> +                % The already seen constraints, in reverse order (meaming

typo: meaning

> diff --git a/compiler/polymorphism.m b/compiler/polymorphism.m
> index c328cae66..ba48708d4 100644
> --- a/compiler/polymorphism.m
> +++ b/compiler/polymorphism.m
> @@ -271,7 +271,7 @@ polymorphism_process_generated_pred(PredId, !ModuleInfo) :-
>  polymorphism_process_pred(PredId, SafeToContinue, !Specs, !ModuleInfo) :-
>      trace [compiletime(flag("debug_poly_caches")), io(!IO)] (
>          % Replace 99999 with the id of the predicate you want to debug.
> -        ( if pred_id_to_int(PredId) = 99999 then
> +        ( if pred_id_to_int(PredId) = 0 then
>              poly_info_set_selected_pred(is_selected_pred, !IO)
>          else
>              true

I assume you didn't mean to commit this.

Peter