[m-rev.] for review: fix mantis bug 384

Julien Fischer jfischer at opturion.com
Thu Apr 2 11:29:52 AEDT 2015

Hi Zoltan,

On Wed, 1 Apr 2015, Zoltan Somogyi wrote:

> For review by Julien. Sorry I didn't get it done yesterday, but
> it was a bit more involved than I originally thought it would be.
> Fix a bug that put $typed_insts in .opt files.
> compiler/mercury_to_mercury.m:
>     Make the predicates that output insts take a parameter that says whether
>     we are trying to print valid Mercury code (in which case we strip any
>     $typed_inst wrappers around other insts), or as-informative-as-possible
>     representations of insts for debugging, in which case we do NOT strip
>     those wrappers. When invoked on insts that CANNOT be printed as valid
>     Mercury even though valid Mercury is being requested, abort.
>     Rationalize the argument order of the involved predicates.
>     Move the output_lang type here from hlds_out_util.m to make the above
>     possible. Change the flag that governs whether item names are qualified
>     or not from a bool to a bespoke enum type. Fix a bug involving that bool.
> compiler/hlds_out_mode.m:
>     This module also has a predicate that output insts, though it already had
>     a valid mercury/debug output flag. If it is asked to print valid Mercury,
>     make it strip $typed_inst wrappers and abort on insts that cannot be
>     printed as valid Mercury.
>     A better fix would be to eliminate this predicate, and other predicates
>     in this module that also duplicate predicates in mercury_to_mercury.m,
>     after generalizing the latter as needed, but that would take more time.


> compiler/mode_constraint_robdd.m:
> compiler/mode_constraints.m:
>     Fix an unrelated problem that ALSO prevented a bootcheck with -O5
>     --intermodule-optimization, which is that with these options, several
>     predicates in both these modules accumulated enough type ambiguities
>     for the type checker to throw up its hands in despair.
>     The type checker should NOT be influenced by the additional types, data
>     constructors, field names, functions and predicates added by these options
>     to their respective tables, but it is. Fixing that bug is future work.
>     For now, just add reduce the ambiguities to the allowable level, mostly
>     by adding module qualifiers, and by converting some function calls to
>     predicate calls. (Function calls introduce ambiguities, since the
>     typechecker thinks they may be partially applied predicate calls.)

That looks fine.


More information about the reviews mailing list