Am Freitag, dem 25.07.2025 um 20:12 -0700 schrieb Bill Wendling:
> On Thu, Jul 24, 2025 at 3:20 PM Martin Uecker <ma.uec...@gmail.com> wrote:
> > Am Donnerstag, dem 24.07.2025 um 15:06 -0700 schrieb Bill Wendling:
> > > > constexpr size_t size = 4;
> > > > struct foo {
> > > > char (*buf)[size] __counted_by(size); // two different "size"!
> > > > int size;
> > > > };
> > >
> > >
It is since C23 and the example is
accepted as valid ISO C 23 by both GCC and Clang:
https://godbolt.org/z/ejKMYbqee
You could also repace it with an enumeration constant, it
does not make any difference to the name lookup rules in C.
> So I suppose what you wrote is
> supposed to be C++, which is beyond the scope of this RFC, because C++
> has scoping rules that are far more complex than C's, and also doesn't
> allow for VLA's in structs/classes. Without the 'constexpr' keyword,
> the above becomes a struct with a VLA in it. But all of this is
> besides the point. If we're worried about two different name
> resolutions because of a GNU-only feature, then we're getting off
> topic.
I am not, the example above is valid C.
Although GNU extensions is also something GCC needs to care about,
so not necessarily offtopic.
And even without the "constexpr" it would not be a structure
with a VLA, but a structure with a variably-modified type in it,
Even though such types are not allowed in structures, WG14 made support
for variably-modified types mandatory in ISO C23 (previously support
was optional) exactly to be able to improve bounds safety in C. As a
future language evolution is also relevant for this discussion, it
would also not be off topic.
Martin
