http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47858
--- Comment #2 from Alexandre Oliva <aoliva at gcc dot gnu.org> 2011-02-24 13:50:18 UTC --- I think it would make sense to model this call-site SRA as partial inlining. It wouldn't get us perfect debug info, but it would at least give us something sensible at the inlined entry point, without duplicating all the infrastructure of partial inlining. Any reason why it shouldn't be modeled this way?