LWG-2381 Inconsistency in parsing floating point numbers

[StephanTLavavej]

LWG-2381 Inconsistency in parsing floating point numbers

C:\Temp>type meow.cpp

#include <cassert>
#include <cstdlib>
#include <sstream>

bool test(const char* const str, const double correct) {
    const double dbl1 = std::strtod(str, nullptr);
    assert(dbl1 == correct);

    std::istringstream iss(str);
    double dbl2 = 0.0;
    iss >> dbl2;
    return dbl2 == correct;
}

int main() {
    assert(test("1.0123456789e+5", 0x1.8b7291613d31cp+16));
    assert(test("0x1.01234567p+5", 0x1.01234567p+5));
    assert(test("0x1.89abcdefp-5", 0x1.89abcdefp-5));
}

C:\Temp>cl /EHsc /nologo /W4 /MTd meow.cpp && meow
meow.cpp
Assertion failed: test("0x1.01234567p+5", 0x1.01234567p+5), file meow.cpp, line 17

This was voted into the Working Paper in October 2021, see #2236.

[frederick-vs-ja]

It seems the implementation in MSVC STL is non-conforming even before LWG-2381. Floating-point numbers should be parsed in the same way despite whether ios_base::hexfloat is set in the flags.
We should merge _Getffld and _Getffldx functions in <xlocnum> as the required functionality is the union of theirs.