is_nothrow_move_constructible does not compile on GCC 4.6

[akrzemi1@…]

Tests is_nothrow_move_constructible_test and is_nothrow_move_constructible_test_no_intrinsics fail in branch Develop on tester igaztanaga-gcc-4.6c++11 (and also igaztanaga-gcc-4.7c++11).

By "fail" I mean that the trait returns neither true nor false, but simply fails the compilation. This in turn causes failures in Boost.Optional.

[John Maddock]

I don't know how to fix this - those GCC versions fail inside the noexcept expression when the class being tested has a private assignment operator or copy constructor. There appear to be no compiler intrinsics available for those GCC versions either. So I'm stuck.

BTW I don't believe this ever worked, but I could be wrong?

[akrzemi1]

Replying to johnmaddock:

I don't know how to fix this - those GCC versions fail inside the noexcept expression when the class being tested has a private assignment operator or copy constructor. There appear to be no compiler intrinsics available for those GCC versions either. So I'm stuck.

I understand the difficulty. Is it possible to isolate the offending compiler versions and provide a trivial definition of the trait for them? I can see it is not easy as the traits seam to work for other testers with GCC 4.6.

BTW I don't believe this ever worked, but I could be wrong?

I cannot be sure. But I thing these errors weren't there in Boost.Optional. Perhaps these testers are new.

[John Maddock]

OK, I've basically disabled these two traits for older gcc versions - see ​https://github.com/boostorg/type_traits/commit/1f3a3a1553dc316ed002354dce3ba3f081b2e102 - note that this introduces new failures as all UDT's now report false with these traits for gcc < 4.8, this is safe but obviously sub-optimal.

[John Maddock]

Update: in response to ​https://github.com/boostorg/variant/issues/47 I've tentatively re-enabled support for is_nothrow_move_constructible in gcc-4.7. Tests all pass locally, so we'll see.