Add fmt::print() overload to support compiled format (FMT_COMPILE)

[alexezeder]

Resolves #2280.

By adding a new exported function - detail::print_buffer(std::FILE*, const memory_buffer&). This function is used in both vprint() from format-inl.h and a new print() from compile.h.

[vitaut]

Thanks for the PR. In general looks good but I wonder if it's worth doing considering that format string processing time is likely small compared to I/O. Could you provide some benchmarks?

[alexezeder]

Okay, I have the results, and there are some improvements. It worth noting though that these results may be too specific for my PC.

Using static library:

Benchmark	Time
JustWrite42	12.2 ns
PrintTrivial	25.2 ns
PrintTrivialCompile	20.3 ns
PrintComplex	216 ns
PrintComplexCompile	135 ns

Using header-only library:

Benchmark	Time
JustWrite42	12.1 ns
PrintTrivial	25.2 ns
PrintTrivialCompile	16.6 ns
PrintComplex	216 ns
PrintComplexCompile	141 ns

Here a short notes about each benchmark:

• JustWrite42 - just write "42" to file using std::fwrite() (just to measure base I/O speed)
• PrintTrivial - use fmt::print("{}", 42) to write 42
• PrintTrivialCompile - same as PrintTrivial, but using compile-time API
• PrintComplex - same as trivial, but with "{:08x} {:0.2f} {:>7}" as format string
• PrintComplexCompile - same as PrintComplex, but using compile-time API

[vitaut]

I think it would be slightly better to pass the text as a string_view instead of a buffer. This also means that the name would have to change. Maybe detail::print?

[alexezeder]

Done, the most challenging part of this change was to name the new argument correctly 😄, but since you mentioned that it's the text, I used text as a name.

[vitaut]

Okay, I have the results, and there are some improvements.

Thanks for the benchmarks, they look convincing.