<xlocnum>: Finish overhauling num_get

[StephanTLavavej]

This file was previously overhauled to call strtod() and strtof(), fixing long-standing correctness and performance issues:

STL/stl/inc/xlocnum

Lines 48 to 62 in 58c5c4d

	inline double _Stodx_v2(const char* _Str, char** _Endptr, int _Pten, int* _Perr) { // convert string to double
	int& _Errno_ref = errno; // Nonzero cost, pay it once
	const int _Orig = _Errno_ref;
	_Errno_ref = 0;
	double _Val = _CSTD strtod(_Str, _Endptr);
	*_Perr = _Errno_ref;
	_Errno_ref = _Orig;
	if (_Pten != 0) {
	_Val *= _CSTD pow(10.0, static_cast<double>(_Pten));
	}
	return _Val;
	}

STL/stl/inc/xlocnum

Lines 65 to 79 in 58c5c4d

	inline float _Stofx_v2(const char* _Str, char** _Endptr, int _Pten, int* _Perr) { // convert string to float
	int& _Errno_ref = errno; // Nonzero cost, pay it once
	const int _Orig = _Errno_ref;
	_Errno_ref = 0;
	float _Val = _CSTD strtof(_Str, _Endptr);
	*_Perr = _Errno_ref;
	_Errno_ref = _Orig;
	if (_Pten != 0) {
	_Val *= _CSTD powf(10.0f, static_cast<float>(_Pten));
	}
	return _Val;
	}

For binary compatibility, the new-ish machinery is activated by a sentinel value:

STL/stl/inc/xlocnum

Lines 563 to 565 in 58c5c4d

	// TRANSITION, ABI: Sentinel value used by num_get::do_get()
	// to enable correct "V2" behavior in _Getffld() and _Getffldx()
	#define _ENABLE_V2_BEHAVIOR 1000000000

In vNext, we need to:

1. Eliminate the _ENABLE_V2_BEHAVIOR sentinel value and make it unconditional.
2. Eliminate lots of dead code (whole functions and source files).
3. Eliminate other dead logic; e.g. _Val *= _CSTD pow(10.0, static_cast<double>(_Pten)); is almost certainly dead (which is good, because it would be incorrect - anything that wants to multiply by a power of ten is highly suspicious) incorrect, see <xlocnum>: "multiply by power of 10" logic is imprecise #1582.

It may be possible to go further. Using charconv for iostreams floating-point would result in dramatic performance wins, but we'd have to solve issues like:

• Locale sensitivity
• Differing formats (e.g. 0x prefixes)
• Runtime rounding modes

vNext note: Resolving this issue will require breaking binary compatibility. We won't be able to accept pull requests for this issue until the vNext branch is available. See #169 for more information.

[StephanTLavavej]

num_get isn't fantastic, but I believe this is done. #3364 eliminated my old _ENABLE_V2_BEHAVIOR weirdness, and #5836 purged lots of dead code.