When wchar_t is a real type, users shouldn't see unsigned short machinery

[fsb4000]

Fixes #216

[fsb4000]

I'm not quite sure if I need to do something with codecvt <unsigned short, char, mbstate_t>, need I?
And to which file do I need to move:

STL/stl/inc/iosfwd

Lines 254 to 258 in 78ff461

	#if defined(_CRTBLD)
	using ushistream = basic_istream<unsigned short, char_traits<unsigned short>>;
	using ushostream = basic_ostream<unsigned short, char_traits<unsigned short>>;
	using ushfilebuf = basic_filebuf<unsigned short, char_traits<unsigned short>>;
	#endif // defined(_CRTBLD)

?

[StephanTLavavej]

I'm not quite sure if I need to do something with codecvt <unsigned short, char, mbstate_t>, need I?

I suspect (70% sure) that it should be guarded by #ifdef _CRTBLD instead of #ifdef _NATIVE_WCHAR_T_DEFINED. That way, it will still be exported, but users with native wchar_t won't see it.

STL/stl/inc/xlocale

Lines 2106 to 2108 in 3136525

	#ifdef _NATIVE_WCHAR_T_DEFINED
	template <>
	class _CRTIMP2_PURE_IMPORT codecvt<unsigned short, char, mbstate_t> : public codecvt_base {

STL/stl/inc/xlocale

Line 2306 in 3136525

#endif // _NATIVE_WCHAR_T_DEFINED

---

And to which file do I need to move:

Only the 5 following source files mention ush* - I am 98% sure that these are the only files that need the typedefs.

STL/stl/src/ushcerr.cpp

Lines 10 to 12 in 3136525

	#define wistream ushistream
	#define wostream ushostream
	#define wfilebuf ushfilebuf

STL/stl/src/ushcin.cpp

Lines 10 to 12 in 3136525

	#define wistream ushistream
	#define wostream ushostream
	#define wfilebuf ushfilebuf

STL/stl/src/ushclog.cpp

Lines 10 to 12 in 3136525

	#define wistream ushistream
	#define wostream ushostream
	#define wfilebuf ushfilebuf

STL/stl/src/ushcout.cpp

Lines 10 to 12 in 3136525

	#define wistream ushistream
	#define wostream ushostream
	#define wfilebuf ushfilebuf

STL/stl/src/ushiostr.cpp

Lines 9 to 11 in 3136525

	#define wistream ushistream
	#define wostream ushostream
	#define wfilebuf ushfilebuf

[CaseyCarter]

I'm not quite sure if I need to do something with codecvt <unsigned short, char, mbstate_t>, need I?

I suspect (70% sure) that it should be guarded by #ifdef _CRTBLD instead of #ifdef _NATIVE_WCHAR_T_DEFINED. That way, it will still be exported, but users with native wchar_t won't see it.

Summarizing here a discussion I had with myself in Discord: explicit template specializations should be visible whenever it's possible to instantiate them implicitly from a primary template or partial specialization. Hiding explicit specializations is simply asking for ODR violation nightmares. Consequently, I think codecvt <unsigned short, char, mbstate_t> needs to be visible only when _ENFORCE_FACET_SPECIALIZATIONS is 0 (which should be the case when _CRTBLD is defined).

[CaseyCarter]

Thank you for helping to partially excise this historical abomination.

[CaseyCarter]

I've never given it any thought before, but seriously who put defined in the name of a macro so we have to query if MEOW_DEFINED is defined??!? This is even worse that putting not in the name of a boolean condition. (No change requested.)

[CaseyCarter]

🤮🤮🤮 (No change requested; looking in these old sources can be painful.)

[StephanTLavavej]

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

[StephanTLavavej]

Thanks again for eliminating these bizarre specializations as much as possible! 👻 🎉 😸