http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48065
--- Comment #6 from Jan Hubicka <hubicka at gcc dot gnu.org> 2013-01-03 16:37:00 UTC --- Hmm, rejecting inlines late at ltrans stage really is problem as discussed earlier - we make assumptions on the optimizations of constant parameters and we won't export offline copy in partitioning so it may lead to undefined reference. How do you expect this to work? Where we do the refusal? I do not see any code there. I would really preffer inlining to work (and produce wrong code in undefined cases perhaps with warning) rather than giving up on the inline plan. If this is not possible to implement, we can materialize a clone with the expected constant propagated and redirect call there, but it is tricky to implement, too.
