On Mon, 7 Sep 2020, 10:34 Jakub Jelinek, <ja...@redhat.com> wrote: > On Mon, Sep 07, 2020 at 10:27:13AM +0100, Jonathan Wakely via Gcc wrote: > > On Mon, 7 Sep 2020 at 09:18, Iain Sandoe wrote: > > > > > > Perhaps the PR should be reopened with “accepts invalid”? > > > > My impression from the PR is that the reporter was using a different > > ABI, where the name isn't reserved. Maybe the testcase should only be > > accepted with -fno-threadsafe-statics or -ffreestanding or something > > to say "I'm doing things differently". > > > > Or we could just say that G++ reserves the Itanium ABI names > > unconditionally, even if it doesn't need to use them, in which case it > > would be accepts-invalid. > > All identifiers starting with two underscores are reserved for the > implementation already. >
Doh, of course. So they'd have to be using some other unsupported ABI which uses that name for a different meaning, which seems like a badly designed ABI given that the "__cxa_" prefix is already claimed by the Itanium ABI. So we shouldn't ICE but any other outcome for that testcase would be ok, including rejecting it.