https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94988
--- Comment #11 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Alexandre Oliva from comment #9)
> ISTM that the test invokes undefined behavior because the assignment and the
> increment in the loop both modify the same storage without an intervening
> sequence point. ISTM that the dynamic type of that storage is thus
> uncertain, and accessing it afterwards, without an intervening store that
> resolves its type either way, would also invoke undefined behavior.
I think the source can be rewritten to
*b = x;
b++;
and still show the original issue on the GIMPLE side.
