[m-rev.] for post-commit review: append to .opt files in intermod.m

[Julien Fischer]

Hi Zoltan,

On Mon, 14 Sep 2015, Zoltan Somogyi wrote:

> This diff mostly just moves code, specifically, it moves the code
> that writes the results of most of the intermodule analysis passes
> to one place, intermod.m. However, two things would benefit
> from review.
>
> (1) The name of the new module, term_constr_main_types.m.
> If anyone has a better name, please tell me.
>
> (2) The documentation of the term_constr_*.m modules
> in compiler_design.html.

The new documentation is fine, although compiler/polyhedron.m (part of
the libs package) is missing from the document.

> I also noticed that while several of the analyses record their
> results for a procedure in the procedure's proc_info, some
> store it separately. The unused_args, exception, trailing and
> mm_tabling analyses all store their results in the module_info
> as a map from pred_proc_ids to the proc-specific analysis results.
>
> Would anyone object if I moved these analysis results to the
> proc_infos of the procedures involved? It seems cleaner, and
> would simplify some code I am planning to write, but maybe
> there is a gotcha I don't know about that would make it a bad idea.

I wrote latter three and don't recall there being a specific reason.

> I also have another question: would anyone object if I removed
> trans_opt.m, moving what little is left in there to intermod.m?
> It would significantly improve intermod.m's cohesion. (Before this
> diff, trans_opt.m invoked each of several analysis modules individually;
> now all the invoked predicates are in intermod.m.)

No, go ahead.

Julien.