Re: FYI: fix latent bug in dw2_find_symbol_file

[Cary Coutant <ccoutant@google.com>]

>>> Doug> You need to call init_cutu_and_read_dies in both debug-types and
>>> Doug> non-debug-types cases.  When DWO files are in use DW_AT_name lives
>>> Doug> in the DWO file.  OTOH TUs typically don't have DW_AT_name (you
>>> Doug> need to look at DW_AT_decl_file).  OTOOH this function is only
>>> Doug> called by find_main_filename.
>>>
>>> What gcc branch should I use to test this?
>>> And do I need gold?
>>>
>>> Doug> I'm tempted to rename the routine {,*_}find_function_file (or some
>>> Doug> such) and have dw2_find_symbol_file ignore TUs.
>>>
>>> Fine by me, but I don't plan to do it.
>>
>> Well, there's only 5 or so places you need to change for the renaming. :-)
>>
>> What I'm saying is the patch is wrong for TUs but TUs are irrelevant
>> here anyway.
>> What did you want to test?
>
> Heh.  Monday blues.
> [The patch is also wrong for CUs.]
>
> To test with DWO files you'll need a fission gcc+binutils.
> I will let Sterling or Cary suggest what's the best to use.

Since only the first of the fission patches has been checked in
upstream so far, I'd suggest that the best thing to try is the
google/gcc-4_6 branch. There, the fission option is spelled
"-gfission" (it'll be "-gsplit-dwarf" when checked in on trunk). I
have a branch in my own git repo that's a lot closer to top of trunk,
though, and I could push that to the google/debugfission branch in the
git mirror if you'd like.

-cary