Fix num_get's float-parsing problem

[achabense]

Fixes #3980 / DevCom-10450662 / VSO-1876474 / AB#1876474

Also fixes DevCom-10445082 / VSO-1873604 / AB#1873604.

Likely fixes VSO-1874745 / AB#1874745.

Likely fixes VSO-1876601 / AB#1876601.

[achabense]

I haven't checked the standard carefully, but please check whether this is just another problem:
If the function sees exponent character and finds no following digits, it will consider the whole section as invalid:
(maybe not a problem, as gcc and clang have the same output)

#include <iostream>
#include <sstream>
using namespace std;

int main()
{
    istringstream istr("1.3e");
    double x = 0;
    istr >> x;

    double y = stod("1.3e");
    cout << x << ' ' << y; // "0 1.3"
}

Relevant logic:
https://github.com/microsoft/STL/blob/6c69a73911b33892919ec628c0ea5bbf0caf8a6a/stl/inc/xlocnum#L1003-L1006r
https://github.com/microsoft/STL/blob/6c69a73911b33892919ec628c0ea5bbf0caf8a6a/stl/inc/xlocnum#L1042-L1044r

[StephanTLavavej]

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

[statementreply]

I haven't checked the standard carefully, but please check whether this is just another problem: If the function sees exponent character and finds no following digits, it will consider the whole section as invalid: (maybe not a problem, as gcc and clang have the same output)

I think this is the correct behavior according to [facet.num.get.virtuals]/3.3.6:

The numeric value to be stored can be one of:

• zero, if the conversion function does not convert the entire field.
• ...

The resultant numeric value is stored in val. If the conversion function does not convert the entire field, or if the field represents a value outside the range of representable values, ios_base​::​failbit is assigned to err.

[achabense]

For the tests introduced in this pr, clang accepts all and gcc is (wrongly?) rejecting those with 0x prefixes.

[StephanTLavavej]

#3364 implemented LWG-2381 which taught the STL to recognize p and P exponents, thus accepting hexfloats. Also see the Annex C wording added by the LWG issue resolution that explains this change.

[CaseyCarter]

Just style nitpicks; ignore me.

[CaseyCarter]

Not worth resetting testing: there are a log of strs floating around here, it might be nice to call this parameter "input" or something.

[CaseyCarter]

Not worth resetting testing: vwls pls.
Flt isn't that much shorter than Float. If you want a descriptive name, be descriptive. If you want a short name, T would work fine here and honestly std::floating_point T (or static_assert(std::is_floating_point_v<T>) in the function body) would be even better.

[StephanTLavavej]

Float specifically is dangerous due to the potential confusion with float. Floating or any other name avoids that hazard.

[StephanTLavavej]

Thanks for fixing this major regression! 🛠️ ✨ 🎉