Re: new patches using -fopt-info (issue5294043)

[Richard Guenther <richard dot guenther at gmail dot com>]

On Thu, Oct 20, 2011 at 10:36 AM, Basile Starynkevitch
<basile@starynkevitch.net> wrote:
> On Thu, Oct 20, 2011 at 10:21:27AM +0200, Richard Guenther wrote:
>
>> I'd rather have a way to make dump-files more structured (so, following
>> some standard reporting scheme) than introducing yet another way
>> of output. ?[after making dump-files more consistent it will be easy
>> to revisit patches like this, there would be a natural general central
>> way to implement it]
>
> I'm not sure to understand what more structured dump-files mean. Are you
> thinking of making them in XML or in JSON format (both requires some minimal
> structure).

No, with structured I mean passes should dump things in a common format.
Which in the end is achieved most easily by providing some helpers that
take care of the formatting details.  Consider the various ways passes dump
that they simplified a statement.  Or the various ways passes dump a
kind of lattice they computed for all SSA names, or how they dump information
on some loop.

>
>> So, please fix dump-files instead.
>
> I'm not sure to understand what that means, because I am not sure to grasp
> the intended meaning and roles of dump files (in particular, from a plugin
> point of view). And I'm not sure to understand what fixing them means.

I don't see how plugins come into the picture here or how they should be
any different from the core compiler.

>
> My incomplete understanding is that dump files are for any kind of
> GCC-debugging (or GCC-plugin-debuggging) output which any pass might feel
> useful, but that is probably too weak as a definition. Or are dump files
> something more precise that that?

No, that's exactly what they are about.  They are already structured
by means of the dump-file modifiers (the TDF_ flags), so you can already
filter the output.  It should be simple to extend this, and at the end,
select a part of the output to stdout as well.

Richard.