Use permissive wide-to-narrow transcoding in filesystem_error

[Chronial]

Fixes #927: std::filesystem_error constructor used to raise a std::system_error when given paths that are not representable in the current codepage. This replaces unrepresentable characters with a replacement character instead of raising system_error.

[ghost]

All CLA requirements met.

[BillyONeal]

This looks reasonable but needs tests.

[BillyONeal]

I don't know why the CI is unhappy.

This looks reasonable but needs tests.

For example:

#include <assert.h>
#include <filesystem>
#include <string_view>
#include <system_error>

using namespace std;
namespace fs = std::filesystem;

void test_filesystem_error_with_bad_codepage_characters() {
    fs::path problem_path{L"\U0001F4A3"}; // an emoji that goes boom
    fs::filesystem_error err{"something bad happened", problem_path, error_code{}};
    assert(string_view{err.what()}.find('?') != string_view::npos);
}

int main() { test_filesystem_error_with_bad_codepage_characters(); }

[Chronial]

Thanks, I'll add a test. Is it ok to assume that the tests won't run with a utf-8 system codepage? Not that I have any idea what else to do about that :).

[BillyONeal]

Thanks, I'll add a test. Is it ok to assume that the tests won't run with a utf-8 system codepage? Not that I have any idea what else to do about that :).

If I understand correctly getting that requires a manifest we aren't going to be adding to our bits. If you're concerned you can replace my 'emoji that goes boom' with an incorrect surrogate pair or something like that.

[Chronial]

Added a test. I am not sure whether WideCharToMultiByte will always use a ? as replacement character – according to the docs, that's up to the codepage to specify. I think that's fine, so I rather checked to make sure that the rest of the path is still there.

I also added a check to make sure the test is actually testing something.

[StephanTLavavej]

This looks good to me - I have some superficial nitpicks, but all of the logic appears to be correct and following our various conventions (thanks for imitating the existing machinery closely).

[StephanTLavavej]

Yay! Thanks for clang-formatting - looks like my brain can't perfectly simulate it yet.

[Chronial]

Thanks for your quick feedback 👍

[CaseyCarter]

Congratulations on your first STL contribution!