[libc++] Make std::allocator_arg and friends conforming in C++17

[ldionne]

Summary

This patch makes global tag variables like std::allocator_arg
conform to C++17 by defining them as inline constexpr variables.
This is possible without creating an ODR violation now that we don't
define strong definitions of those variables in the shared library
anymore.

[Mordante]

Thanks for working on this! This makes helps to enable modules in libc++.

LGTM, but I really would like to use a helper macro over #ifdefs.

libcxx/include/__functional/bind.h

67

I thought you had a marco _LIBCPP_INLINE_CONSTEXPR_VAR in a different patch, that could be sued here too.

[philnik]

In D145589#4178824, @Mordante wrote:

LGTM, but I really would like to use a helper macro over #ifdefs.

Another option would be to just always mark them inline. Both clang and GCC support this as an extension.

[Mordante]

In D145589#4178836, @philnik wrote:

In D145589#4178824, @Mordante wrote:

LGTM, but I really would like to use a helper macro over #ifdefs.

Another option would be to just always mark them inline. Both clang and GCC support this as an extension.

Really? I wasn't aware of that. In that case +1 for that solution.

[philnik]

In D145589#4178843, @Mordante wrote:

In D145589#4178836, @philnik wrote:

In D145589#4178824, @Mordante wrote:

LGTM, but I really would like to use a helper macro over #ifdefs.

Another option would be to just always mark them inline. Both clang and GCC support this as an extension.

Really? I wasn't aware of that. In that case +1 for that solution.

I think the main problem with that approach is that it will be hard to only use extensions in the libc++ headers and not in the tests. We could probably add some magic into _LIBCPP_BEGIN_NAMESPACE_STD and _LIBCPP_END_NAMESPACE_STD, since we have 99% of our code in there, but Clang and GCC support different extension in some cases, making it potentially quite annoying to find out which ones are supported.