Re: [RFC] Support for comdat-based dwarf debug info reduction for types.

[Doug Evans <dje@google.com>]

On Fri, Jan 16, 2009 at 7:31 PM, Joel Brobecker <brobecker@adacore.com> wrote:
> Doug,
>
>> Hi.  Here is my current patch to support comdat-based dwarf debug info
>> reduction for types.
>>
>> http://wiki.dwarfstd.org/index.php?title=COMDAT_Type_Sections
>
> I don't remember anyone answering your email. That's too bad, because
> I think that any attempt at reducing the size of debugging info is
> interesting.
>
>> I'm hoping someone can find some to give me a review.
>> The patch can't go in today as the corresponding GCC support isn't
>> in the main trunk yet.
>
> Regarding timing, I don't think we have to wait for GCC. As far as I can
> tell, this would add a functionality without disturbing the current
> behavior, right? Or perhaps the problem is actual testing of your code?

Correct, it adds functionality without disturbing current behaviour.
[if there is a disturbance, that's a bug]

The patch has been well tested with and without the corresponding gcc
patch of Cary's.  No claim is made that it couldn't use more testing
of course. :-)  IIRC, there is one known bug, tracking of the source
file a type came from; I plan to fix that before the gdb release with
the patch goes out of course.

Depending on the relative timings of gdb/gcc tree status and releases,
we may want to get the gdb support in sooner rather than later (so gdb
releases support the feature when it is in gcc).

One reason for waiting might be that some other compiler might
implement the current proposal, and then the dwarf spec changes and
gdb has to support the old version.  The proposal only has a few
remaining minor issues, and I think the probability of anything like
this happening is remote; I offer it for completeness' sake.

> Speaking of GCC, do you know the status of the enhancement?

We're waiting on gcc to enter stage1.  The patch is ready, modulo AIUI
it hasn't been reviewed yet, and modulo I don't know how much gcc has
changed since the patch was last tested.

>> This patch is incomplete in that the proposal has moved the ODR signature
>> from the CU header to an attribute.
>> This is a trivial change, relative to the entire patch.
>
> Do you have an updated version of your patch. I've mostyl been hacking
> the dwarf2 reader, so hopefully someone more veteran can review the
> code eventually, but my comments will (hopefully!) be better than
> nothing :).

I will resubmit a patch against current CVS head next week.
The patch is currently in the archer-dje-dwarf-sig8 archer branch.

Thanks.