[Bug middle-end/54945] Too strong non-aliasing analysis?

[matz at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54945

--- Comment #13 from Michael Matz <matz at gcc dot gnu.org> 2012-10-19 13:12:35 UTC ---
(In reply to comment #12)
> What do you mean by invalid? It is certainly not undefined behavior.

No, but the expectation implicitely coded into the return code is invalid.
Sorry for sloppy use of words.  The outcome depends on GCC putting both
objects next to each other which noone guarantees.  The testcase can yield 0
or 1 depending on the moon phase.