https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82632
Patrick Palka <ppalka at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |ppalka at gcc dot gnu.org
--- Comment #3 from Patrick Palka <ppalka at gcc dot gnu.org> ---
(In reply to Luke Dalessandro from comment #2)
> Is this demonstrating the same behavior, and is clang wrong to accept?
>
> ```
> template <class T, int = 1>
> struct Foo
> {
> Foo() = default;
> Foo(Foo const&) = delete;
>
> template <int I>
> Foo(Foo<T, I> const&) {}
> };
>
> template <class T, int I>
> Foo(Foo<T,I>) -> Foo<T,I+1>;
>
> Foo<int> a;
> Foo b = a;
> ```
>
> https://godbolt.org/z/njqPEKMM1
It seems clang is correct to accept, since a user defined deduction guide is
preferred over an equally specialized copy deduction candidate according to
[over.match.best]. I think this is instead a symptom of the recently fixed
PR86439; GCC trunk accepts this testcase after r12-1744.
Before closing this PR, we ought to add these two testcases to the testsuite.
