Fix formatting std::chrono::duration types to wide strings

[ghost]

I was getting several compiler errors when trying to format a std::chrono::duration to a wide string, and found that various parts of chrono.h had a hard dependency on narrow chars. I tried to make it agnostic as possible, and it seems to work fine for me now, although I haven't tested it with anything besides std::chrono::duration<double, std::milli>.

If anything needs to be changed or fixed to get this merged just let me know.

I agree that my contributions are licensed under the {fmt} license, and agree to future changes to the licensing.

[vitaut]

Thanks for the PR! Looks good but could you add a test case checking that format compiles when passing a wide format string an a chrono object?

[ghost]

I have committed a wide version of the FormatDefault case, will that work?

[ghost]

Hmm, seems to be failing with the 42µs case. I might need to look at this more.

  Actual: L"42\xFFC2\xFFB5s"
Expected: L"42µs"
Which is: L"42\xC2\xB5s"

Edit: Yep, it looks like I missed a couple things with format_duration_unit/get_units. I'll let you know when it is fixed.

[ghost]

After some trial and error, I believe the µ issue was from the encoding of the source file itself, so I specialized all of the get_units for wchar_t and in the wide version I used \u00B5 instead of the µ symbol, to make sure it is encoded correctly.

I also added Char template overloads everywhere necessary to make sure everything is being done with wide characters. I think this should pass all the tests.

[vitaut]

Good idea.

[vitaut]

I suggest making unit a string_view and removing s.

[ghost]

I'll need to rework some things to do this properly since get_units can return a nullptr, which fmt's string_view doesn't handle properly as far as I know. Since it passes the pointer to std::char_traits<T>::length() which will dereference it without checking.

So it either needs to stay as-is, or I could revise get_units to return string_views, and specifically construct a null-data 0-length string_view from there, which wouldn't invoke the length call.

Personally I think it would be cleaner to leave this as-is.

[vitaut]

Let's leave as is this then.

[vitaut]

Merged, thanks!