Do not compile even trivial UTF16 and UTF32 tests with libc++

[Antony Polukhin]

In libc++ there are no specializations of std::numpunct<char16_t> and std::numpunct<char32_t> templates. So no test shall be run with char16_t and char32_t types on libc++.

[Marshall Clow]

Additional information:

The C++ standard (I looked at n3216) specifies (in 22.3.1.1.1) that numpunct<char> and numpunct<wchar_t> are required, and does not mention numpunct<char16_t> or numpunct<char32_t> at all.

gcc (I checked version 4.8) provides a completely templated implementation of std::numpunct, so it supports specializing on any character type. libc++ provides two implementations: one for char and the other for wchar_t.

[Antony Polukhin]

(In [83717]) Disable UTF16 and UTF32 tests for libc++ (refs #8369)

[Antony Polukhin]

Replying to marshall:

Additional information:

The C++ standard (I looked at n3216) specifies (in 22.3.1.1.1) that numpunct<char> and numpunct<wchar_t> are required, and does not mention numpunct<char16_t> or numpunct<char32_t> at all.

Yes, they are not required, which makes a lot of troubles when std::basic_istream and std::basic_ostream classes are used with those character types as template parameters (users miss this functionality, see ​1, ​2).

gcc (I checked version 4.8) provides a completely templated implementation of std::numpunct, so it supports specializing on any character type. libc++ provides two implementations: one for char and the other for wchar_t.

GCC-4.7 was also providing a fully templated version. Unfortunately, some support from glibc was required, so usage of std::numpunct<char16_t> was leading to segmentation faults (though have not checked it for a year, may be things changed).

[Antony Polukhin]

(In [83764]) Mrege from trunk:

• Disable UTF16 and UTF32 tests for libc++ (fixes #8369)