[m-rev.] for review: [CTGC] bugfix, use typevarset from pred_info

[Julien Fischer]

On Mon, 19 Jun 2006, Nancy wrote:

> Julien Fischer wrote:
> > On Mon, 19 Jun 2006, Nancy Mazur wrote:
> >
> >> Hi,
> >>
> >> this is a bugfix. The error occurred when analysing TypeSpecOf__ predicates,
> >> but may as well occur in other predicates too. The problem was that the
> >> typevarset used to determine type substitutions was in sometimes simply set to
> >> an empty varset, instead of getting it from the pred_info structure as it
> >> should be.
> >
> > If this was caused by the XXX comment at the end of this diff (the on
> > that says to be checked) the that's fine, otherwise where was the other
> > typevarset coming from?
>
> Yes, that is indeed the clue of the fix...
>
> >> Question: I was wondering what the use is of analysing the TypeSpecOf__
> >> predicates, as the results are (explicitly) not recorded in the trans_opt
> >> files (Cf. the XXX remark in structure_sharing.analysis.m, and in analogy,
> >> in structure_reuse.analysis.m)? I might as well simply not analyse them?
> >
> > They're not recorded in the .trans_opt files because of limitations in
> > the way transitive intermodule optimization works.  It's certainly worth
> > analysing them, particularly if they are local to a module, it's just
> > that (at the moment) we won't be able to use specialized versions
> > created by structure reuse across module boundaries.
>
> Okay...
>
> >> Estimated hours taken: 5
> >> Branches: main
> >>
> >> Bugfix. Make sure to use the typevarset stored in the pred_info when
> >> determining the type substitution between formal and actual sharing
> >> information.
> >>
> >> compiler/structure_reuse.direct.detect_garbage.m:
> >> compiler/structure_reuse.direct.m:
> >> compiler/structure_reuse.indirect.m:
> >> compiler/structure_sharing.domain.m:
> >> 	Use the typevarset from the pred_info structure when looking up
> >> 	and renaming sharing information.
> >> 	Thread pred_info until the place where it is used.
> >
> > Is there a test case for this bug?
>
> Well, the list module wouldn't compile without the fix...
> Should I add that kind of testcase somewhere (and how?)?

You would need to extract a testcase from the list module and add it
to tests/valid (most probably).  I guess that since it's a standard
library module adding a separate testcase for it at the moment
isn't too important.

Julien.
--------------------------------------------------------------------------
mercury-reviews mailing list
post:  mercury-reviews at cs.mu.oz.au
administrative address: owner-mercury-reviews at cs.mu.oz.au
unsubscribe: Address: mercury-reviews-request at cs.mu.oz.au Message: unsubscribe
subscribe:   Address: mercury-reviews-request at cs.mu.oz.au Message: subscribe
--------------------------------------------------------------------------