STDEXEC_HAS_FEATURE(__cpp_lib_execution) fails with recent libstdc++.

[jehelset]

I think 0242ad9 broke NVIDIA/stdexec for at least gcc-15 and gcc-trunk (https://godbolt.org/z/9ez66G7vv).

it checks __has_feature(__cpp_lib_execution), but in libstdc++ is a macro defined at https://github.com/gcc-mirror/gcc/blob/master/libstdc%2B%2B-v3/include/bits/version.h#L672.

should perhaps guard with

#if defined(__cpp_lib_execution)
#  define STDEXEC_HAS_EXECUTION_POLICY() 0
#elif __has_feature(__cpp_lib_execution)
#  define STDEXEC_HAS_EXECUTION_POLICY() 1
#else
#  define STDEXEC_HAS_EXECUTION_POLICY() 0
#endif

@lucteo

[prlw1]

As it stands, I think the logic in the above is backwards and the first 0 should be a 1, and the resulting build failure is a separate bug.

However, this is on the right lines. A little digging in case it helps:

STDEXEC_HAS_FEATURE is used twice in stdexec: 1) to detect thread_sanitizer, and 2) to detect __cpp_lib_execution.

In 1), that is precisely to detect a compiler feature. In 2), it is used to check whether the compiler has implemented a C++ library feature.
The use in 2) was standardised in c++20 with the advent of the <version> header, in P0754R2.

Using STDEXEC_HAS_FEATURE for case 2) has been replaced by including <version> and testing for __cpp_lib... macros directly.

The nit is that execution policies first arrived in c++17, so predates this checking system. On the other hand, I think both execution and version landed in gcc 9, so I think testing for __cpp_lib_execution should work no matter what.

I don't know the clang situation, and the 1) use has to remain, as I think it is clang specific, and gcc's compatibility __has_feature is intended to work with it.

[lucteo]

Appologize, I missed this. I can take a look when I get some spare time (these days are rather busy for me)