Re: Generating gdb index at link time

[Dodji Seketeli <dodji@redhat.com>]

Hello,

Cary Coutant <ccoutant@google.com> writes:

>>> Tom> This problem was that .debug_pub* are explicitly for public names, but
>>> Tom> for GDB we wanted to see all names (so that "break staticfunction" works
>>> Tom> and "break typo" doesn't cause reading all CUs).  This particular
>>> Tom> problem was resolved by you :) saying that we should just change GCC to
>>> Tom> emit all the names.
>>>
>>> Cary> No one has actually changed GCC to do that, yet, though -- right?
>>>
>>> I think either Dodji or I had a patch at one point.  It may have been
>>> part of Dodji's patch to change .debug_pub* to also include all the
>>> extra info we thought we needed at the time.
>>>
>>> I'm not sure if it is still around.
>>
>> If you can find it, we'd definitely be interested in it.

I stored the iterations of that patch as attachments to
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41130 before dropping the
ball on this issue.

[...]

>>
>>> Cary> I think the biggest problem is an issue I've seen where the DIE for a
>>> Cary> class isn't properly nested inside the right context (even after
>>> Cary> following DW_AT_specification). There's an ugly workaround for this in
>>> Cary> gdb (by looking for a member subprogram with a linkage name, and
>>> Cary> extracting the class name from that), which I copied into gold. As
>>> Cary> part of this work, I'd like to fix that bug in GCC, and any others
>>> Cary> like it.
>>>
>>> Yeah.  We've tried to file and/or fix problems like this as we've run
>>> across them.  There are still some open bugs around naming.
>>>

I'd be interested in looking into these nesting bugs when I have some
free cycles.  Please CC me on these.

Thanks.

-- 
		Dodji