Add support for _BitInt on clang

[Arghnews]

Issue #4007

Only enabled for clang >= 14 as that's all that supports _BitInt at the moment

We need a forwarding reference for map to be a "more specialized" match; const T&, T etc. are ambiguous.

I'm not too familiar, would there be a better way to do this by adding an entry into enum class type for _BitInts? Or this is fine?

---

Fyi this worked in version 9 because of this catch-all overload:

fmt/include/fmt/core.h

Lines 1506 to 1516 in a337011

	template <typename T, typename U = remove_cvref_t<T>,
	FMT_ENABLE_IF(!is_string<U>::value && !is_char<U>::value &&
	!std::is_array<U>::value &&
	!std::is_pointer<U>::value &&
	!has_format_as<U>::value &&
	(has_formatter<U, Context>::value ||
	has_fallback_formatter<U, char_type>::value))>
	FMT_CONSTEXPR FMT_INLINE auto map(T&& val)
	-> decltype(this->do_map(std::forward<T>(val))) {
	return do_map(std::forward<T>(val));
	}

Which was then subsequently narrowed down to extended FP types (which "broke" the accidental support for this):

fmt/include/fmt/base.h

Lines 1527 to 1537 in 6a192f8

	// is_fundamental is used to allow formatters for extended FP types.
	template <typename T, typename U = remove_const_t<T>,
	FMT_ENABLE_IF(
	(std::is_class<U>::value || std::is_enum<U>::value ||
	std::is_union<U>::value || std::is_fundamental<U>::value) &&
	!has_to_string_view<U>::value && !is_char<U>::value &&
	!is_named_arg<U>::value && !std::is_integral<U>::value &&
	!std::is_arithmetic<format_as_t<U>>::value)>
	FMT_MAP_API auto map(T& val) -> decltype(FMT_DECLTYPE_THIS do_map(val)) {
	return do_map(val);
	}

[vitaut]

Thanks for the PR!

[vitaut]

@Arghnews, do you plan to update this PR or shall we close it for now (can be reopened later)?

[Arghnews]

@Arghnews, do you plan to update this PR or shall we close it for now (can be reopened later)?

@vitaut Hi, sorry been busy and now I'm on holiday for 2 weeks, but will update once back and implement your feedback. I'd say leave it open and I'll update in 2-3 weeks, but it's up to you (can reopen of course). Cheers!

[vitaut]

No hurry. Let's keep it open and enjoy your holiday!

[Arghnews]

Hi,

Reworked this so that now fmtlib will format _BitInts up to 128 bits

Uses wider formatter ie. integer types are widened to 64bit if <= 64 bits wide (assume this is acceptable for performance)
If FMT_USE_INT128 then supports up to 128 bits

Can't write static_assert(fmt::is_formattable<signed_bitint<129>, char>{}, ""); as apple clang only allows up to 128 bits. BITINT_MAXWIDTH isn't defined, could check __apple_build_version__ but I just removed this test

Let me know any feedback @vitaut

[vitaut]

A few more comments.

[vitaut]

I think this can be moved to bitint_traits which would make number_of_bits and is_signed unnecessary.

[Arghnews]

I think this can be moved to bitint_traits which would make number_of_bits and is_signed unnecessary.

Nice suggestion, have done. Only thing to note, we must #include <cstdint> in base.h now, for the int64_t, if this is acceptable. This was already included in format.h

[vitaut]

Let's use (unsigned) long long which is guaranteed to be at least 64-bit and avoid additional include.

[Arghnews]

Let's use (unsigned) long long which is guaranteed to be at least 64-bit and avoid additional include.

Good shout, have made this change

[vitaut]

Merged, thanks!