The explicit instantiation declaration suppresses the definition of A<int>::A() in defaulted.o, but the explicit instantiation definition doesn't cause that symbol to be emitted in impl.o, so when that constructor is not inlined there is no definition.

As a single file:

template<typename T>
struct A
{
    T x;
    A() = default;
    A(const A &other) = delete;
};

extern template class A<int>;

int main()
{
    A<int> a;
}

This compiles with clang but not G++ because Clang doesn't create a reference to A<int>::A() from main(), so it doesn't matter that the explicit instantiation is not defined in the program.

(In reply to Jonathan Wakely from comment #1)
> The explicit instantiation declaration suppresses the definition of
> A<int>::A() in defaulted.o, but the explicit instantiation definition
> doesn't cause that symbol to be emitted in impl.o, so when that constructor
> is not inlined there is no definition.

That's more or less what I figured was happening. Can you clarify whether you think this a GCC bug or just me misunderstanding the language? Thanks.

It's a bug

*** Bug 58078 has been marked as a duplicate of this bug. ***

*** Bug 76521 has been marked as a duplicate of this bug. ***

*** Bug 51629 has been marked as a duplicate of this bug. ***

Related to (and maybe a dup of) PR 60796

Also https://llvm.org/bugs/show_bug.cgi?id=22763

*** Bug 60796 has been marked as a duplicate of this bug. ***

Author: jason
Date: Fri Aug 26 15:10:51 2016
New Revision: 239782

URL: https://gcc.gnu.org/viewcvs?rev=239782&root=gcc&view=rev
Log:
	PR c++/57728 - explicit instantiation and defaulted functions

	* pt.c (do_type_instantiation): Don't mess with non-user-provided
	member functions.

Added:
    trunk/gcc/testsuite/g++.dg/cpp0x/explicit11.C
    trunk/gcc/testsuite/g++.dg/cpp0x/explicit12.C
Modified:
    trunk/gcc/cp/ChangeLog
    trunk/gcc/cp/class.c
    trunk/gcc/cp/pt.c

For aarch64 and arm, the new test g++.dg/cpp0x/explicit12.C shows as UNRESOLVED and the log complain about output files not existing.

The test has { dg-do link .. } and { dg-final { scan-assembler-not .. } } but dg-do link doesn't seem to preserve the assembler files. Was the test intended to have a { dg-options "-save-temps" } set?

Matthew

Author: jason
Date: Thu Sep  1 01:55:47 2016
New Revision: 239913

URL: https://gcc.gnu.org/viewcvs?rev=239913&root=gcc&view=rev
Log:
	PR c++/57728 - adjust testcase

	* g++.dg/cpp0x/explicit12.C: Add -save-temps.

Modified:
    trunk/gcc/testsuite/g++.dg/cpp0x/explicit12.C

Author: jason
Date: Tue Dec  6 18:51:37 2016
New Revision: 243311

URL: https://gcc.gnu.org/viewcvs?rev=243311&root=gcc&view=rev
Log:
	PR c++/57728 - explicit instantiation and defaulted functions

	* pt.c (do_type_instantiation): Don't mess with non-user-provided
	member functions.

Added:
    branches/gcc-6-branch/gcc/testsuite/g++.dg/cpp0x/explicit12.C
Modified:
    branches/gcc-6-branch/gcc/cp/ChangeLog
    branches/gcc-6-branch/gcc/cp/class.c
    branches/gcc-6-branch/gcc/cp/pt.c

This bug saw its last activity nearly a year ago, with patches from Jason applied, but the issue does not seem to be fixed in its entirety (using gcc 7.2).

Specifically, this translation unit:

template<class T>
struct C {
  C() = default;
  void f() { }
  int i = 5;
};
template class C<int>;

causes C<int>::f() to be defined, but the defaulted constructor of C<int> is not defined. (Replacing "= default" with "{}" causes definition of the C<int> constructor.)

$ nm -C x.o
0000000000000000 W C<int>::f()

The current behavior is obviously not what we want if we choose to use explicit instantiations.

*** Bug 67632 has been marked as a duplicate of this bug. ***

Jens, see the discussion in https://bugs.llvm.org/show_bug.cgi?id=22763

Jason, should this be closed as FIXED for 6.3 and up? I believe what Jens observes is the intended behaviour of GCC and Clang now.

Yes.