rsmith added a comment. In D147722#4249612 <https://reviews.llvm.org/D147722#4249612>, @erichkeane wrote:
> Hmmm... the examples I thought would cause a problem don't seem to when > switching that to Done. So I'm going to run regression tests and give that a > shot. > > I think I now get what you mean about 'Identity' arguments, we basically need > to have arguments to keep something like the above example (where 'foo' is a > template, and the partial specialization is done correctly), and when trying > to substitute into it, we would end up having the arguments to `Outer` and > `Foo`, but substitution would think of the `Foo` args as the `Inner` args for > depth. I'm giving the testing a shot, and will see if that fixes the problem > without causing a regression in any of the concerns I have. I suspect I > don't have a problem with the 'identity' for the same reason 'Done' works: we > never hit that situation. If we know that all further levels will be "identity" levels (mapping template-param-i-j to template-param-i-j), then we can use `MultiLevelTemplateArgumentList::addOuterRetainedLevels` to represent that and leave those levels alone. That will mean the `getNumSubstitutedLevels() == 0` check in `SubstituteConstraintExpression` will fire a lot more often, which seems like a nice benefit. CHANGES SINCE LAST ACTION https://reviews.llvm.org/D147722/new/ https://reviews.llvm.org/D147722 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
