[Bug c++/126310] [16/17 regression] Miscompile due to erroneous value range propagation since r16-5887
"jason at gcc dot gnu.org via Gcc-bugs" <[email protected]>
| Newsgroups | gmane.comp.gcc.bugs |
|---|---|
| Message-ID | <[email protected]/bugzilla/> |
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126310
Jason Merrill <jason at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jason at gcc dot gnu.org
--- Comment #5 from Jason Merrill <jason at gcc dot gnu.org> ---
(In reply to Martin Jambor from comment #4)
> The record_type contexts may be bit telling, 0x7ffff75ea498 has
> 0x7ffff75e55e8 as its canonical-type but differs in alias set and,
Hmm, dow can it differ in alias set? It looks like get_alias_set goes through
TYPE_CANONICAL.
> well, has its own chain of field_decls.
Yes, build_ptrmemfunc_type builds a different RECORD_TYPE for each variant.
> I'm not sure what this means, though. I suppose the code should then
> (at least for non-unions?) rely on offsets rather than on pointer
> comparison?
Or names.