https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106150
Jonathan Wakely <redi at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
See Also| |https://gcc.gnu.org/bugzill
| |a/show_bug.cgi?id=98423
--- Comment #7 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #3)
> Here is an example which is valid (after
> https://cplusplus.github.io/CWG/issues/2084.html):
> struct S1 {
> S1();
> };
> struct S {
> S();
> };
> union U {
> S s{};
> S1 s1;
> } u;
>
> The check in GCC for this seems to be off, if only the variant s is there,
> GCC (and clang) accepts it.
>
> So the full check for the defect report was never really done (and it was
> not even mentioned in the defect report commentary either).
Yes, all of gcc, clang, edg and msvc reject cases like this.
It should not matter that S1 does not have a trivial default ctor, because the
default member initializer should make this equivalent to:
union U {
S s;
S1 s1;
U() : s() { }
};
Having to write a user-provided constructor is annoying, because to do it
"right" in a generic std::lib template requires:
constexpr U() noexcept(is_nothrow_default_constructible_v<S>) requires
default_initializable<S> { }
But that should be approximately how the defaulted default ctor is defined
automatically, without all that verbosity.
This is a dup of PR 98423, where Jakub pointed out the code that needs to
change, and what needs to happen.
