Questions about FMT_COMPILE

[alexezeder]

I'm experimenting with FMT_COMPILE lately and here are a few points about it that are interesting for me. Note that my assumptions below can be wrong.

1. FMT_COMPILE (or I should probably say compile-time API) doesn't support arg_id. There are no tests for that and when I try to use arg_id in this API then a compiler is trying to tell me that it's unable to do this. I don't think it's impossible to implement but it just takes some time.

2. The compilation error with arg_id inside of FMT_COMPILE, as well as other non-valid for compile-time API format string I suppose, is non-intuitive. Here is an example: https://godbolt.org/z/ha38a6
   At least for me, the error that there is no type is non-intuitive. Especially when there are implemented handling inside parsing code:
   

   fmt/include/fmt/compile.h

   Line 544 in 4a6eadb

   return unknown_format();

   
   and strange check that allows the compiler to continue its journey to that no ::type error:
   

   fmt/include/fmt/compile.h

   Lines 573 to 575 in 4a6eadb

   	if constexpr (std::is_same<remove_cvref_t<decltype(result)>,
   	detail::unknown_format>()) {
   	return detail::compiled_format<S, Args...>(to_string_view(format_str));

   
   Of course, I can miss some case where it should go that path, so let me know.

3. The situation with a format string that has fewer replacement fields than the number of arguments that passed to the format function is valid. For example here are non-failing calls:

   fmt::format(FMT_COMPILE("{}"), 42, 42);
   fmt::format(FMT_COMPILE(""), 42);

   The same usage on Compiler Explorer.
   But the runtime API also sticks with this behaviour, even with FMT_STRING checks, so maybe there is some explanation for that.

[vitaut]

1. Right, some features including positional arguments are not yet implemented in format string compilation.

2. The idea is to fallback on runtime formatting for unsupported features but it still needs more work.

3. It is intentional and consistent with runtime format, printf and std::format.

[roman-orekhov]

Right, some features including positional arguments are not yet implemented in format string compilation.

Why isn't it documented for both FMT_STRING and FMT_COMPILE?

[vitaut]

Positional arguments are supported in both FMT_STRING and FMT_COMPILE.