[Bug middle-end/83460] [8 regression] FAIL: g++.dg/pr79095-4.C
law at redhat dot com
gcc-bugzilla@gcc.gnu.org
Mon Dec 18 18:11:00 GMT 2017
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83460
--- Comment #6 from Jeffrey A. Law <law at redhat dot com> ---
Actually the 79095-4 test is testing that we do loop distribution and
propagation of constants into the the memset call and that after propagation we
realize there's a bogus path that we actually can't remove given the state of
the IL.
Martin's changes to <vector> compromise that test in that it provides enough
information to the compiler and the compiler should eliminate the problematic
call to memset.
If we collapse things down to C it's actually not going to be a terribly
interesting test. And I'd prefer to avoid the CPP output as it's likely to
have host/target dependencies that are going to cause long term headaches.
I think we should just declare this test compromised. Martin's test from
r255753 has reasonable tests and thus I don't think it's even worth turning
this into a verification that Martin's work has the intended effects.
More information about the Gcc-bugs
mailing list