Re: [C++ PATCH] ctor vtable vcall offsets

[Mark Mitchell <mark at codesourcery dot com>]

>>>>> "Jason" == Jason Merrill <jason@redhat.com> writes:

    Jason> so you can understand why I read it the way I do.  :)

Indeed.

    Jason> The main problem with third-party thunks is the time
    Jason> penalty for the indirect branch; that's why we've tried so
    Jason> hard to avoid requiring any.

That's true on IA64 -- which is after all the platform we were
supposed to be worrying about.  It's less true on other chips, but
it's still a problem, I agree.

At one point, I suggested trying to make the linker generate these
thunks.  That was too wacky, but would have made all this simpler.
You could have had your cake and eaten it too, so to speak...

Your arguments are persuasive.  Your suggestion is fine with me.

    Jason> Yep.  Indeed, I suggested a "choose me" flag for the ELF
    Jason> COMDAT groups, but I don't remember whether or not it was
    Jason> accepted; I think some people wanted to be able to just
    Jason> choose the first copy encountered (i.e. the SCO behavior).
    Jason> Anyone know where to find a copy of the evolving ELF spec?

Not I...  Jim Dehnert probably knows though.

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com