[PATCH] D25697: Make the LTO comdat api more symbol table friendly

[Rafael Espíndola via llvm-commits]

I will update the FIXME, but it is basically a bug in the
representation. An alias can point any constant expression and we
might not be able to always evaluate it. For example, whether a llvm
"constant" is actually a constant depends on whether
-ffunction-section is used or not.

This was fixed in r209007 but reverted in r210062 and I don't have the
energy to fight it. With a bitcode symbol table we might be able to at
least say that the crazy aliases are not supported and error when
creating the table instead of when we try to use them.

Cheers,
Rafael


On 22 October 2016 at 14:30, Mehdi AMINI <mehdi.amini at apple.com> wrote:
> mehdi_amini accepted this revision.
> mehdi_amini added a comment.
> This revision is now accepted and ready to land.
>
> I didn't look at the state before, most of this is mechanically translated.
>
> The fact that llvm-lto2 is not testing this API is still an issue, but shouldn't block this.
>
> If you can improve the FIXME to better describe the error situation that'd be great.
>
> LGTM.
>
>
> https://reviews.llvm.org/D25697
>
>
>