[Bug lto/85248] [6/7/8 Regression] Miscompilation when using error function that returns

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85248

--- Comment #7 from Martin Liška <marxin at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #6)
> I think we want to merge the cgraph nodes but _not_ merge decls (thus not
> call lto_symtab_prevail_decl).  Iff that is supported somehow...?  Probably
> we expect the decl <-> cgraph node be a 1:1 relation, not a n:1 one.

In case of MPX, there's some weirdness. I don't want to be familiar with that.

> 
> I think the issue you run into can happen generally, no?  Or we indeed do
> not reclaim the mpx stuff properly.

I guess it's MPX related.

> 
> Anyhow, I guess you need to debug the specific ICE further or reduce the
> scope
> of "flags" you bail out for now to that of the error() case (noreturn).

I'll do that. In GCC 9 we can remove MPX and utilize flags_from_decl_or_type.