Should a UDL version of FMT_STRING be added?

[sth]

Usage of the FMT_STRING() macro could be replaced by a user defined literal operator that creates a fmt::compile_string.

The advantage would be a more concise syntax and avoiding the "ugly" use of a marco: fmt::print(FMT_STRING("{}: {}"), a, b) could be written as fmt::print("{}: {}"_fmt, a, b). The disadvantage would be increased API complexity by having this additional option to create a format string and possible compatibility questions for compilers without sufficient UDL support.

I know there is a ""_format() UDL, but as I understand it that replaces the whole fmt::format() function and can't be used together with other functions like fmt::print() or fmt::format_to(). A UDL that just creates a fmt::compile_string can be used more generically in all places where a format string is expected.

It can be implemented like this:

template <typename Char, Char... CHARS>
struct udl_compile_string : fmt::compile_string {
   using char_type = Char;

   static FMT_CONSTEXPR_DECL char_type s[] = {CHARS..., '\0'};

   FMT_MAYBE_UNUSED FMT_CONSTEXPR
   operator fmt::basic_string_view<char_type>() const {
      return fmt::detail::compile_string_to_view<char_type>(s);
   }
};

template <typename Char, Char... CHARS>
FMT_CONSTEXPR udl_compile_string<Char, CHARS...> operator""_fmt() {
   return {};
}

[vitaut]

Do you mean FMT_STRING?

[sth]

Yes that's what I mean... I edited it.

[vitaut]

Yes, a UDL would be better than a macro. Unfortunately it will only be implementable in a portable and efficient(ish) way in C++20.

[sarah-quinones]

here's a c++20 implementation https://godbolt.org/z/zYfErb

though this is currently only supported on gcc 10+, as far as i know

[vitaut]

Closing in favor of #2022. We can have compile-time checks without wrapping a format string in a UDL.