https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104218
Bug ID: 104218
Summary: 23_containers/vector/ext_pointer/types tests rely on
GCC overload-resolution bug
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: libstdc++
Assignee: unassigned at gcc dot gnu.org
Reporter: brooks at gcc dot gnu.org
Target Milestone: ---
In cleaning up some old internal bugs, I came across this one from years ago
when I was running the 4.9.4 libstdc++ testsuite through Clang, and it looks
like it's still relevant.
The 23_containers/vector/ext_pointer/types/{1,2}.cc programs compile
successfully with GCC, of course, but with Clang we get "use of overloaded
operator '-' is ambiguous" errors in std::vector<N::X,
__gnu_cxx::_ExtPtr_allocator<N::X> >.
This reduced testcase shows the difference in the two compilers' behavior:
----
template<typename T> struct X {
template<typename U> void operator-(const X<U>&);
};
template<typename T> void operator-(T, T);
void f(X<int> x) { x - x; }
----
Compiler explorer link: https://godbolt.org/z/nnb98W5nW
Richard Smith believes this is a GCC bug, not a Clang bug, per this analysis:
> GCC appears to believe the member function is more specialized, and calls it.
> Clang and I think that's wrong: deduction fails in both directions in partial
> ordering, so the call is ambiguous.
>
> Indeed, C++ DR 532 gives a very similar example. Following its rule, the above
> would be equivalent to:
>
> template<typename T> struct X {};
> template<typename T, typename U> void operator-(X<T> &, const X<U>&);
> template<typename T> void operator-(T, T);
>
> void f(X<int> x) { x - x; }
>
> ... which GCC and Clang agree is ambiguous. So: this test is relying on a GCC
> bug.
----
