[Bug tree-optimization/58794] [4.8/4.9 Regression] ICE in set_lattice_value, at tree-ssa-ccp.c:455 on x86_64-linux-gnu (at -O1, -O2, and -O3)

["ebotcazou at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58794

--- Comment #5 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
> I'm sure that we can build such FIELD_DECL only with Ada though, so, Eric,
> can you provide a testcase where that happens - thus, that shows that
> side-effects cannot be ignored here by for example comparing
> &f.x and &f.x for a case where that is not equal?  I think we need to
> concern ourselves only with mutating side-effects, not a volatile load.

I don't think so, we do not rely on expressions appearing in offsets or sizes
to implement the semantics of the language, that would be too error-prone;
instead these expressions are computed once for all when the type is
elaborated.

> The question is whether the patch is ok as-is or if I have to make
> behavior dependent on is_gimple_form (ugh).  A testcase that breaks
> if not guarding it that way would be nice to have (I'll check if anything
> existing triggers).

Go ahead I'd say.