[Bug c++/87676] New: Presence of variadic constructor template breaks overload resolution for other constructors

[harald at gigawatt dot nl]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87676

            Bug ID: 87676
           Summary: Presence of variadic constructor template breaks
                    overload resolution for other constructors
           Product: gcc
           Version: 8.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: harald at gigawatt dot nl
  Target Milestone: ---

Reduced testcase:

  struct S1 { S1(int, int, int); };
  struct S2 {
    S2(const S2 &) = delete;
    S2(const S1 &);
    S2(S1 &&);
    template <typename...T> S2(T &&...);
  };
  S2 s({1, 2, 3});

I expect this to compile using the S2(S1 &&) constructor. Instead, GCC emits an
error:

  test.cc:8:15: error: call of overloaded ‘S2(<brace-enclosed initializer
list>)’ is ambiguous
   S2 s({1, 2, 3});
                 ^
  test.cc:5:3: note: candidate: ‘S2::S2(S1&&)’
     S2(S1 &&);
     ^~
  test.cc:4:3: note: candidate: ‘S2::S2(const S1&)’
     S2(const S1 &);
     ^~
  test.cc:3:3: note: candidate: ‘S2::S2(const S2&)’ <deleted>
     S2(const S2 &) = delete;
     ^~

That is, ambiguous between the three non-template constructors. But if the
constructor template is commented out, GCC no longer sees it as ambiguous, GCC
does prefer the S2(S1 &&) overload.

I do not understand why the presence of a constructor template would affect the
overload resolution between the other constructors. FWIW, clang behaves the
same way as GCC, which is a sign that there might be a good reason for the
current behaviour, but MSVC and Intel do accept the code.