https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118346
Simon Martin <simartin at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |ASSIGNED
Assignee|unassigned at gcc dot gnu.org |simartin at gcc dot
gnu.org
--- Comment #2 from Simon Martin <simartin at gcc dot gnu.org> ---
Interesting one!
gcc diverges from clang / EDG / MSVC++ when using ((C*)0)->i)
=> https://godbolt.org/z/EoGxTs5dE
but does NOT when using __builtin_offsetof (C, i);
=> https://godbolt.org/z/7qGjMePMn
Looking at the history, the error has been added via r6-5625-g1a161cd7a75453
for bug #68711, and a follow-up (that is specific to __builtin_offsetof) was
done via r7-6354-g905d2812e7bd7e for bug #68727.
I think that we get exactly what we want (i.e. accept ((C*)0)->i while still
rejecting __builtin_offsetof (C, i)) by simply undoing the first commit.
This is what I'll be starting from.
