On Thu, 29 Jun 2023, 08:21 Jonathan Wakely, <cxx_at_[hidden]> wrote:

>
>
> On Wed, 28 Jun 2023 at 23:37, Frederick Virchanza Gotham via Std-Proposals
> <std-proposals_at_[hidden]> wrote:
>
>> A few months ago I was thinking of submitting this to Boost, but now
>> I'm actually thinking maybe it should be in the Standard Library.
>>
>
> It won't be approved by the Parallelism & Concurrency study group, for the
> reasons already explained previously.
>
> Maybe you should propose it for Boost though. The review might be useful.
>
>
>
>>
>>
>> #if (__cplusplus >= 201700L) || (defined(__has_cpp_attribute) &&
>> __has_cpp_attribute(nodiscard))
>>
>
> This is not how you test for attributes. For a start, you're assuming that
> if __cplusplus >= 201703 is true then the attribute is supported, which is
> not correct. The point of __has_cpp_attribute is to check if the attribute
> is supported, so that's what you should use. More importantly, if
> __has_cpp_attribute is not defined then this line will give an error. Using
> an undefined function-like macro is ill-formed.
>
> You need to do:
> #ifdef __has_cpp_attribute
> # if __has_cpp_attribute(nodiscard)
> # define attrib_nodiscard [[nodiscard]]
> # endif
> #endif
> #ifndef attrib_nodiscard
> # define attrib_nodiscard [[nodiscard]]
>

Oops, copy & paste error. Obviously this fallback definition should be
empty:
#define attrib_nodiscard

#endif
>
>
>
>
>> # define attrib_nodiscard [[nodiscard]]
>> #else
>> # define attrib_nodiscard /* nothing */
>> #endif
>>
>> #if (__cplusplus >= 201907L) && defined(__cpp_concepts)
>>
>
> This is not how you use feature test macros. You're supposed to test the
> value of the macro itself, not __cplusplus, so:
>
> #if __cpp_concepts >= 201907L
>
> Your version would give the wrong answer for a compiler with partial C++20
> support, that only supports an older implementation of concepts.
>
>
>
>