Update isfinite declaration to be constexpr for c++23 only #3746
Conversation
hchataing
force-pushed
the
constexpr-isfinite
branch
2 times, most recently
from
f009126
to
04338ce
Compare
|
Question about the environment: |
|
Granted, I might have been too quick in diagnosing the issue |
The implementation failed to be constexpr in the situation where: __cplusplus == 202002L && !defined(__cpp_lib_is_constant_evaluated) Fix fmtlib#3745
hchataing
force-pushed
the
constexpr-isfinite
branch
from
04338ce
to
f68bff6
Compare
| @@ -2752,7 +2752,7 @@ template <typename T, FMT_ENABLE_IF(std::is_floating_point<T>::value&& | |||
| has_isfinite<T>::value)> | |||
| FMT_CONSTEXPR20 bool isfinite(T value) { | |||
| constexpr T inf = T(std::numeric_limits<double>::infinity()); | |||
| if (is_constant_evaluated()) | |||
| if (is_constant_evaluated(true)) | |||
There was a problem hiding this comment.
One drawback is that it switches to a slightly worse implementation on older systems: https://www.godbolt.org/z/WWzjEsrxj. But I guess we can optimize it later.
There was a problem hiding this comment.
I don't see an alternative other than testing FMT_CONSTEXPR20 in the body to replace is_constant_evaluated :(
There was a problem hiding this comment.
We can use __builtin_is_constant_evaluated instead of std::is_constant_evaluated.
There was a problem hiding this comment.
Or we can improve
Lines 108 to 110 in 89860eb
condition by adding a stdlib version check.
@hchataing, which version of libc++ (_LIBCPP_VERSION) or libstdc++ (_GLIBCXX_RELEASE) you are using?
There was a problem hiding this comment.
@hchataing check PR #3754 please
There was a problem hiding this comment.
Neither macro is defined by the compiler... but maybe I am checking wrong (clang++ -std=c++20 -E -dM foo.cc)
|
Closing in favor of #3754 but thanks for the effort. |
Fix #3745