<yvals_core.h>: _EMIT_STL_ERROR leads to uninformative error logs

[StephanTLavavej]

#2897 introduced:

STL/stl/inc/yvals_core.h

Lines 512 to 514 in 6c69a73

	#define _EMIT_STL_ERROR(NUMBER, MESSAGE) \
	_EMIT_STL_MESSAGE("error " #NUMBER ": " MESSAGE) \
	static_assert(false, "Error in C++ Standard Library usage.")

This emits a #pragma message which is captured in full build logs but not error-only logs, followed by a static_assert which does appear in error-only logs but with a message that doesn't say what's actually wrong.

This has the potential to confuse users who are looking at error-only logs; e.g. in twistedfall/opencv-rust#452 the problem was very simple (Clang version was too old), but the static_assert lost the reason.

For _EMIT_STL_ERROR, I think we should use static_assert alone (with the given NUMBER and MESSAGE); I don't see any reason to use #pragma message here.

[frederick-vs-ja]

STL/stl/inc/yvals_core.h

Lines 509 to 511 in 6c69a73

	#define _EMIT_STL_WARNING(NUMBER, MESSAGE) \
	_EMIT_STL_MESSAGE("warning " #NUMBER ": " MESSAGE) \
	static_assert(true, "")

Maybe related: can/should we remove the seemingly useless static_assert from _EMIT_STL_WARNING?

[frederick-vs-ja]

I found that #pragma message has its own benefit:

• it doesn't emit unhelpful error codes (C1189 for #error or C2338 for static_assert);
• it can cause Visual Studio to generate a link in Error List - however, the generated links are currently invalid.

But it can't be used for rejecting an invalid program.

Error diagnostics have some drawbacks:

• currently, static_assert doesn't prevent MSVC emitting following error messages;
• #error prevents MSVC emitting following error messages, but stringization doesn't take place within it; and
• both of them emit unhelpful error codes.

Ideally, there should be something behaving like #pragma message except that it terminates compilation after emitting the error message.