On Thursday, 20 March 2025 14:49:33 Pacific Daylight Time Sebastian Wittmeier
via Std-Proposals wrote:
> Or is it for optional features of the standard library?
>
>
> Then the standard library would have to mark its option, but not implemented
> features as [[unimplemented]].

An optional Standard Library feature would be marked by the presence or
absence of a __cpp_lib_macro. That's already standard practice for even
features that are mandatory, but which we all realise take time to become
reality so may not be available yet.

There has been no need to declare things with an attribute. There are no
warnings caused from this, because you either have a hard error or a proper
compilation.

The only prior art we have for availability attributes are Apple's headers.
All the functions are declared and can be called, but those are implemented as
weak functions, which allows you to verify the presence of at runtime. Those
are also a source of annoyance because they mismatch with the C++ __cpp_lib
macros: for example, you'll find out that std::filesystem classes are present,
but you can't use them because the functions resolve to nullptr on some older
systems. You'd have to make a choice to raise your minimum OS version or not
rely at all on the C++ macros.

As a result, the impedance mismatch of the two sets of availability
methodologies has not been a smooth success. In fact, I recommend against
there being two procedures in the first place.
 
>
> But to get rid of the warnings, the cpp file would have to include another
> header.
 
>
> Then the standard library could instead not declare those functions.
>
>
> And the other header could declare them.
>

-- 
Thiago Macieira - thiago (AT) macieira.info - thiago (AT) kde.org
  Principal Engineer - Intel DCAI Platform & System Engineering