Build fails on i386 (with FMT_USE_INT128=0)

[jwt27]

The library currently does not compile on i386 due to:

In file included from /home/jw/fmt/src/format.cc:8:
/home/jw/fmt/include/fmt/format-inl.h: In instantiation of 'bool fmt::v8::detail::basic_fp<F>::assign(Float) [with Float = long double; F = fmt::v8::detail::uint128_fallback]':
/home/jw/fmt/include/fmt/format-inl.h:216:69:   required from 'fmt::v8::detail::basic_fp<F>::basic_fp(Float) [with Float = long double; F = fmt::v8::detail::uint128_fallback]'
/home/jw/fmt/include/fmt/format-inl.h:2212:20:   required from 'int fmt::v8::detail::format_float(T, int, fmt::v8::detail::float_specs, fmt::v8::detail::buffer<char>&) [with T = long double]'
/home/jw/fmt/src/format.cc:92:64:   required from here
/home/jw/fmt/include/fmt/format-inl.h:228:51: error: call to non-'constexpr' function 'fmt::v8::detail::uint128_fallback fmt::v8::detail::uint128_fallback::operator<<(int) const'
  228 |     constexpr auto implicit_bit = carrier_uint(1) << num_float_significand_bits;
      |                                   ~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from /home/jw/fmt/include/fmt/format-inl.h:29,
                 from /home/jw/fmt/src/format.cc:8:
/home/jw/fmt/include/fmt/format.h:360:22: note: 'fmt::v8::detail::uint128_fallback fmt::v8::detail::uint128_fallback::operator<<(int) const' declared here
  360 |   FMT_CONSTEXPR auto operator<<(int shift) const -> uint128_fallback {
      |                      ^~~~~~~~
In file included from /home/jw/fmt/src/format.cc:8:
/home/jw/fmt/include/fmt/format-inl.h:229:52: error: call to non-'constexpr' function 'fmt::v8::detail::uint128_fallback fmt::v8::detail::operator-(const fmt::v8::detail::uint128_fallback&, uint64_t)'
  229 |     constexpr auto significand_mask = implicit_bit - 1;
      |                                       ~~~~~~~~~~~~~^~~
In file included from /home/jw/fmt/include/fmt/format-inl.h:29,
                 from /home/jw/fmt/src/format.cc:8:
/home/jw/fmt/include/fmt/format.h:351:15: note: 'fmt::v8::detail::uint128_fallback fmt::v8::detail::operator-(const fmt::v8::detail::uint128_fallback&, uint64_t)' declared here
  351 |   friend auto operator-(const uint128_fallback& lhs, uint64_t rhs)
      |               ^~~~~~~~
In file included from /home/jw/fmt/src/format.cc:8:
/home/jw/fmt/include/fmt/format-inl.h:238:9: error: no match for 'operator+=' (operand types are 'fmt::v8::detail::uint128_fallback' and 'fmt::v8::detail::uint128_fallback')
  238 |       f += static_cast<F>(implicit_bit);
      |       ~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from /home/jw/fmt/include/fmt/format-inl.h:29,
                 from /home/jw/fmt/src/format.cc:8:
/home/jw/fmt/include/fmt/format.h:367:22: note: candidate: 'void fmt::v8::detail::uint128_fallback::operator+=(uint64_t)'
  367 |   FMT_CONSTEXPR void operator+=(uint64_t n) {
      |                      ^~~~~~~~
/home/jw/fmt/include/fmt/format.h:367:42: note:   no known conversion for argument 1 from 'fmt::v8::detail::uint128_fallback' to 'uint64_t' {aka 'long long unsigned int'}
  367 |   FMT_CONSTEXPR void operator+=(uint64_t n) {
      |                                 ~~~~~~~~~^

This is caused by commit db74598.

[vitaut]

Should be fixed in 17dda58. Thanks for reporting.