Re: [RFA/commit/Win64] Remove new extra leading underscore in symbol name

[Kai Tietz <ktietz70@googlemail.com>]

2010/6/24 Pedro Alves <pedro@codesourcery.com>:
> On Thursday 24 June 2010 20:00:21, Joel Brobecker wrote:
>> No, you're probably right. I was slowly realizing this while I was
>> updating the comment I wrote in the previous patch. The problem is:
>> what's the right way to detect how the binary was built?
>
> I don't think there's a "right way".  At least, I can't think of
> one.  Maybe there's some heuristic way, like looking for some well
> known symbol in the implementation namespace for two or
> three underscores, say.  Always likely to fail, for several reasons,
> one of them that you still misfix asm defined underscored symbols.
> And you'd probably want to consider handling the reverse -- missing
> underscores, in gdb/bfd's perspective.  Very fragile, if you ask me.

Right, to ask bfd for toolchain's target default underscoring is for
sure not bad, but doesn't address the issue complete. Probing of
symbols could be a way, but it is indeed fragile.

>> Right now,
>> the bfd change is a major incompatibility nightmare since minimal
>> symbols and symbols no longer have the same name.
>
> It was an ABI change.  Incompatibilities are sort of expected by
> design, when all the tools don't agree on the ABI.  :-)

:)

>> GDB needs to be able to support both (IMO).
>
> Does the --enable-leading-mingw64-underscores switch affect
> bfd as well, and fix this?  While supporting both sounds ideal,
> in practice, I'd think it to be enough for vendors to support
> gdb builds that match the ABI of their compiler.  Or two builds,
> if they care, until the old ABI is phased out.

Yes, by --enable-leading-mingw64-underscores just the default in bfd
is changed. So it could be a work-a-round for supporting binaries
having non-ABI name-decoration.

> --
> Pedro Alves
>

Kai

-- 
|  (\_/) This is Bunny. Copy and paste
| (='.'=) Bunny into your signature to help
| (")_(") him gain world domination