https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119916
Jason Merrill <jason at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2025-04-28
Ever confirmed|0 |1
--- Comment #14 from Jason Merrill <jason at gcc dot gnu.org> ---
The new GCC behavior seems to disagree with the EWG direction on CWG2563 from
https://wiki.edg.com/bin/view/Wg21issaquah2023/NotesEWGCWG2563 "We want to
allow for the split between get_return_value and the conversion.".
Apparently EWG supported Lewis' position from
https://lists.isocpp.org/core/2023/01/13715.php
It's frustrating that apparently EWG got to see an example but it isn't
preserved anywhere.
It looks like Clang 15 introduced and then fixed this same bug
(https://github.com/llvm/llvm-project/issues/56532); the fix is described as
"- Fix coroutines issue where ``get_return_object()`` result was always
eargerly
converted to the return type. Eager initialization (allowing RVO) is now
only
perfomed when these types match, otherwise deferred initialization is used,
enabling short-circuiting coroutines use cases."
I guess it would make sense for us to take the same approach.
I'm surprised to see that I'm assigned to drafting 2563, guess I'd better get
on that.
