[m-rev.] proposal for moving some options to new categories

Julien Fischer jfischer at opturion.com
Sat Jul 12 14:41:26 AEST 2025


On Sat, 12 Jul 2025 at 08:17, Zoltan Somogyi <zoltan.somogyi at runbox.com> wrote:
>
> Proposal for moving some options to new categories
>
> From oc_grade_mprof:
>     delete the following four options:
>     delete time_profiling  (NOT the .prof grade!)
>     delete profile_calls
>     delete profile_time
>     delete profile_memory  (NOT the .memprof grade!)
>
> Move the whole oc_verb_dev subsection
>     from the oc_verbosity section to the section containing oc_dev_ctrl
>
> Add new section, oc_intermod, for intermodule optimization
>     add a new oc_plainopt subsection as the first subsection of oc_intermod
>     make oc_transopt the second subsection of oc_intermod
>     add a new oc_latexmodel subsection as the third subsection of oc_intermod
>
> From the oc_opt_ctrl section
>     move to oc_plainopt
>         intermodule_optimization
>         read_opt_files_transitively
>         use_opt_files
>
>     move to oc_transopt
>         transitive_optimization
>         use_trans_opt_files
>
>     move to oc_latexmodel
>         intermodule_analysis
>         analysis_repeat
>         analysis_file_cache
>         analysis_file_cache_dir
>
> Does anyone have any objections to the above?

Not from me.

> For example, does anyone think that the four options whose deletion
> I am proposing will ever be useful again?

Possibly only to mprof developers.  (Time profiling is not supported on Windows,
so that's one place a developer working on that may want a finer level
of control
over the profiling than that offered by the grade components.)

> And on a related note, Julien, do you have any objection against
> making public the options for the second termination analyser, which is,
> or at least was, your baby?

I don't think there is any point in making them public.  For most users of
Mercury the benefit of termination2 will be small to non-existent.
Many years ago my plan was to have one overarching set of user-facing
options that controlled all of the termination analysers; that's probably still
the right way to go in the long run.

Julien.


More information about the reviews mailing list