Reported to me by Attila Feher. https://godbolt.org/z/djh5KxjW6

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

int main() {
    std::cout << "Expected: [1.e-37]\n";
    std::cout << format("  Actual: [{:#6.0g}]\n", 1.234e-37);
}

libstdc++ and libc++ behave as expected, but MSVC prints:

Expected: [1.e-37]
  Actual: [ 1.e-37]

🕵️ Analysis

I debugged into this, and although I'm not absolutely certain of the root cause yet, this line appears to be involved:

_Zeroes_to_append is computed as -1 because _Extra_precision and _Precision are both 0, and _Digits is 1.

We've correctly computed _Width to be 6, but the negative _Zeroes_to_append damages it, which appears to be the proximate cause of the bug:

Later, we use _Zeroes_to_append in a clamped manner (i.e. negative values are treated as zero):

However, we need to completely understand the problem and all possible codepaths before developing a fix (i.e. we can't just clamp it to zero and call it a day). Note that the chars_format::general codepath further adjusts _Zeroes_to_append: