rjmccall added a comment. In D92361#2432578 <https://reviews.llvm.org/D92361#2432578>, @ahatanak wrote:
> It seems like you are discussing the case where a class/struct annotated with > `trivial_abi` contains a member that isn't destructively movable. In that > case, clang correctly diagnoses it today. For example, if you remove the > attribute from `S2` in the above example and add it to `S3` instead, it warns > that `trivial_abi` cannot be applied to `S3` because `S2` is a non-trivial > class type. > > What I wasn't sure was whether `S1` (which isn't annotated with `trivial_abi` > in the original code I posted) should be treated as a destructively movable > type despite having all its copy/move constructors deleted when its only > member (`s0`) is destructively movable. > > Based on the discussion we had a few years ago > (http://lists.llvm.org/pipermail/cfe-dev/2017-November/055966.html), I think > the answer is yes, but I just wanted to confirm. Hmm. My first instinct is to say that a type that "adds new information" about its special members — i.e. that explicitly declares its destructor or copy/move constructors for some reason other than to immediately explicitly default them — should always be treated conservatively in the absence of a `trivial_abi` attribute. So a type that explicitly deletes all its copy/move constructors should never be treated as destructively movable without the attribute. A type with implicitly-deleted copy/move constructors due solely to subobjects that are still destructively movable should still be destructively movable, I think. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D92361/new/ https://reviews.llvm.org/D92361 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
