[PATCH] Bug 12843: put ELF reloc section against section with template function instantiation in the same COMDAT group

Tim Northover t.p.northover at gmail.com
Sun Jul 7 12:10:27 PDT 2013


> Oh, I missed the point that Solaris CC creates
> .group%__1cDbar4Ci_6FTA_i_, not .group.
> Is it what you mean?

Not really; the name of group sections is irrelevant to the
specification. I mean why do we want to add the relocation sections to
the COMDAT group in the first place? Just because what we (and GCC) do
is different from Solaris doesn't necessarily make it wrong.

To me, it seems like they're already bound to the relocated section:
there's no group for .text and .rela.text, or for .data and
.rela.data, so why is one needed for these other cases?

Also, because the new behaviour is different from GCC, there's a
greater risk of it breaking other linkers, in my view. Solaris is a
niche compiler whose behaviour won't necessarily be tested against
each and every linker.

Does our current behaviour cause the Solaris linker to break, or is it
just someone's aesthetic preference?

Cheers.

Tim.



More information about the llvm-commits mailing list