Re: C++20 and gcc12/10

[Greg Troxel <gdt%lexort.com@localhost>]

Jonathan Perkin <jperkin%pkgsrc.org@localhost> writes:

> * On 2025-04-06 at 13:29 BST, Greg Troxel wrote:
>
>>David Holland <dholland-pkgtech%netbsd.org@localhost> writes:
>>
>>> ok then, so what should the FEATURE be called? c++20-gcc10? c++20-lite?
>>
>>I think we first need to fix the problem that c++NN does not mean "full
>>compliance with the standard".   Then I think it makes sense to have
>>
>>  c++20-gcc10  # because that's what we actually care about
>>  c++20-foo    # for the foo features *required by c++20, not in gcc10
>>  c++20-bar    # for the bar features *required by c++20, not in gcc10
>
> No, just "foo" and "bar", like we already have for the other C++
> features we've added support for.  None of this should be specific to
> GCC.

Do you mean

  - formally change c++NN to mean really the entire standard?
  - have some set of features which is jointly c++20

  - [the part I don't get: how do you tag something with needs
    --std=c++NN and uses part of it]

>>One could add this for clang also, but it's interesting that's not an
>>issue.  Perhaps because we aren't trying to support systems with old
>>clang in base, and perhaps because clang's approach raises fewer
>>issues.
>
> Mostly, but we should still architect for OS, architecture, and
> compiler agnosticism.  Software releases and features are only going
> to keep accelerating, so it's only a matter of time until it becomes a
> problem (I'm now starting to run into some C++20 issues on my clang
> 13-based macOS build hosts so probably time I upgraded).

Sure, and I could see some way to split this so we have c++20-minimal,
where that contains everything supported by all compilers that credibly
claim to mostly support c++20.

To me the point is to have a label for a mild requirement for c++20,
rather than a 'really the whole language" requirement, and I don't see
how to get there from what I might guess you are thinking.