<ostream>: With setprecision(0) showpoint fixed, a bogus '.' is emitted for infinity and NaN

[StephanTLavavej]

C:\Temp>type meow.cpp

#include <iomanip>
#include <iostream>
#include <limits>
using namespace std;

int main() {
    constexpr double inf_dbl = numeric_limits<double>::infinity();
    constexpr double nan_dbl = numeric_limits<double>::quiet_NaN();

    cout << "setprecision(0) showpoint fixed:\n";
    cout << setprecision(0) << showpoint << fixed;

    cout << inf_dbl << "\n";
    cout << -inf_dbl << "\n";
    cout << nan_dbl << "\n";
    cout << -nan_dbl << "\n";

    cout << "\nAnd showpos:\n";
    cout << showpos;

    cout << inf_dbl << "\n";
    cout << -inf_dbl << "\n";
    cout << nan_dbl << "\n";
    cout << -nan_dbl << "\n";
}

C:\Temp>cl /EHsc /nologo /W4 /MTd /Od meow.cpp && meow
meow.cpp
setprecision(0) showpoint fixed:
i.nf
-.inf
n.an
-.nan(ind)

And showpos:
+i.nf
-.inf
+n.an
-.nan(ind)

libstdc++ and libc++ behave correctly (note that MSVC's STL has an intentional behavior difference regarding negative NaNs, though): https://godbolt.org/z/MYsPxc98h

Originally reported as DevCom-10512845 / VSO-1917556 / AB#1917556 .

[MattStephanson]

The STL appears to be sending the correct format flags to printf. The UCRT seems to be adding the point regardless of whether the value is finite or not. Perhaps we could work around it by omitting the # flag for non-finite values.

[frederick-vs-ja]

The STL appears to be sending the correct format flags to printf. The UCRT seems to be adding the point regardless of whether the value is finite or not. Perhaps we could work around it by omitting the # flag for non-finite values.

Thanks! I've verified that this is essentially a bug of UCRT and reported DevCom-10519861.

I think we should add some workaround at this moment.