https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79009
Martin Dorey <martindorey at gmail dot com> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |martindorey at gmail dot com --- Comment #1 from Martin Dorey <martindorey at gmail dot com> --- In my testing today, gcc 4.7.2, gcc 6.2.1 and Debian experimental's gcc 7.0.0 20161230, among other versions, reject the one-argument foo and bar instantiations from the OP's testcase, as I think they should. They accept the two-argument foo, I think incorrectly, an accepts-invalid bug. (clang rejects all three.) But the Subject line's use of "nondependent" suggests the OP is reporting an accepts-invalid error on the one-argument instantiations. Could it be that's a regression between 20161230 and 2017-01-05? I don't feel the two-argument foo - the bug I can reproduce - is covered by the Subject line. That's OK, as: Bug #59960 - accepts ill-formed 'auto a1 = t1, a2 = t2;' where t1 and t2 have different template types ... looks to have reported much the same thing, some three years earlier. It's unloved by any cc:s and, unlike this PR, has no Depends-on.