https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101829
--- Comment #2 from Jan Hubicka <hubicka at ucw dot cz> --- > It might be possible to inline such functions by creating a "stub" call either > after or before the inlined function body where the "stub" would just be there > to represent the attributes. > > Say, inline a 'deallocator' fn > > my_free (x); > > as > > <inlined impl> > .IFN_STUB_DEALLOCATE (x); > > or sth like that where those stubs are removed after any diagnostics with > regard to pairing of malloc/free have been issued (not sure where that > currently > happens). We have similar issue with functions calling alloca (that we may inline if was able to insert stack restore). I was always lazy to implement this since we need to also insert it on the EH cleanup edges, but perhaps it is time to learn inliner about such tricks... Honza
