Add basic <format> formatter (C++20)

[AbitTheGray]

TD;DR: Enable std::format for fpm::fixed with minimum code.

---

I have been using std::format from <format> (C++20) as it does not require you to specify the type, which I consider big upgrade from printf. While fpm has stream operators in ios.hpp, you need to specialize std::formatter<,> to use your own type. I've implemented very simple one just to allow me to print it anywhere - the only option there is +, as {0:+} for first argument, to always include sign character (using std::showpos). It uses std::stringstream to redirect the existing << operator.

I understand that the std::format probably has more overhead than printf which may not sit well with people who want to use fixed-point numbers due to them being faster than floating-point numbers. But you are not required to ever call std::format if that is a problem (or not supported by your platform) and I consider the std::format quality-of-life addition.
The specialized formatter is behind C++ feature test ( __cpp_lib_format without version check as the std::formatter is there from the beginning, there is also __cpp_lib_formatters which is for C++23 additions).

Possible enhancements:

• More formatting options (like precision)
• Optimized conversion to string for the formatter (related to Implement to_chars and from_chars #2 )

[AbitTheGray]

I forgot to include <version> so the test macro won't work correctly (will work only if you include it before the header). I'll fix it later today.

EDIT: Fixed, hopefully

[MikeLankamp]

Hi @AbitTheGray, sorry for the delay in the review!

I appreciate your contribution, it's certainly something I'd like to support in principle, but I did have a few comments that would require some significant effort to fix. I hope you might be willing and able to make this happen.

[MikeLankamp]

I think this #endif should be moved to the end, because as you said, __cpp_lib_format won't work if we don't include <version>.

[AbitTheGray]

In theory you could have the support by defining __cpp_lib_format but let's consider it non-standard. So yes, I'll use this #if as a wrapper around the whole code block.

[MikeLankamp]

Since this doesn't assume that std::format exists (as per the version check), should this assume that std::ranges exists? Perhaps a more standard std::copy is preferable here.

[AbitTheGray]

This is inside #ifdef __cpp_lib_format but you are right about std::ranges, I'll just use standard copy 👍

[MikeLankamp]

I think if this now exposes std::format capabilities, it should be done completely and fully. Would you be able to add support for the full float formatting specification? https://en.cppreference.com/w/cpp/utility/format/spec

[AbitTheGray]

I will do my best but I've never wrote this complex formatting. I just hope I'll be able to put the output together with existing operator<< (that is a complex one).

[MikeLankamp]

Could you please add sufficient tests for this new functionality?

[AbitTheGray]

Sure. Now that I know you are OK with including it (and want full "Standard format specification"), it is definitely good idea to add tests. With the first implementation, I just tried few variants. (I know, it is not the way to test it)

[AbitTheGray]

I think that I've implemented everything I can.

Sadly there are 2 missing features:

• Nested width and precision - std::format("{:2.{}f}", 3.14, 2) does not work for fpm::fixed because I don't know how to implement it
• Your std::hexfloat implementation differs from the one of double and I did not intend to re-implement it so that test is disabled

I was hoping the nested width and precision is auto-solved before being handed to the std::formatter<> but that is not true, std::format("{:{}}", fpm::fixed_16_16{0}, 2) gives you "{}}" in the "width detection and processing" code.

For the hex one, value of 2.015625 with format "{:10.6a}" results in "1.020000p+1" for double but fpm::fixed_16_16 gives me " 0x1.02p+1".

---

It is possible there are still some bugs as I've written only a small subset of possible tests. I've focused only on correct, well defined behavior.