On Sat, 15 Nov 2025 at 10:27, Jakub Jelinek <[email protected]> wrote:
>
> On Sat, Nov 15, 2025 at 08:47:10AM +0000, Jonathan Wakely wrote:
> > On Fri, 14 Nov 2025 at 19:37 -0500, Marek Polacek wrote:
> > > This contains the libstdc++ bits (including <meta>).
> >
> > Excliting! A few small changes requested below ...
>
> Thanks, I've added changes to https://forge.sourceware.org/marek/gcc/pulls/109
>
> > > +#ifdef _GLIBCXX_SYSHDR
> > > +#pragma GCC system_header
> > > +#endif
> > > +
> > > +#include <array>
> > > +#include <initializer_list>
> > > +#include <optional>
> > > +#include <source_location>
> > > +#include <span>
> > > +#include <string>
> > > +#include <string_view>
> > > +#include <vector>
> >
> > We could move all these headers after the #if below, so that we don't
> > include anything if this is going to be an empty file. I don't think
> > we're consistent about how we do that elsewhere though, so it's not
> > important to change now.
>
> The bits/version.h part needs to stay before the #if because
> otherwise it would need to repeat the bits/version.h conditionals.
Sure, but that's fine.
That's the way it's done in e.g. <any>
include/std/any-#ifdef _GLIBCXX_SYSHDR
include/std/any-#pragma GCC system_header
include/std/any-#endif
include/std/any-
include/std/any-#define __glibcxx_want_any
include/std/any:#include <bits/version.h>
include/std/any-
include/std/any-#ifdef __cpp_lib_any // C++ >= 17
include/std/any-
include/std/any-#include <initializer_list>
include/std/any-#include <typeinfo>
It probably doesn't matter, because pre-C++26 code isn't going to
include <meta> anyway.
>
> > > +
> > > +#define __glibcxx_want_reflection
> > > +#include <bits/version.h>
> > > +
> > > +#if __glibcxx_reflection >= 202506L // C++ >= 26 && __cpp_impl_reflection
>
> > > + namespace meta
> > > + {
> > > + using std::meta::info;
> > > + using std::meta::exception;
> > > + using std::meta::operators;
> > > + using enum std::meta::operators;
> >
> > This is enough to export all the enumerators of this enum, right?
>
> Yes. Initially I had it commented out with TODO, but then tried
> it on a small testcase:
>
> module;
> namespace N { enum class A { A1, A2, A3 }; using enum A; }
> export module a1;
> export namespace N { using N::A; using enum N::A; }
>
> and
>
> import a1;
> int main ()
> {
> N::A a = N::A::A1;
> N::A b = N::A2;
> }
>
> and it worked, so then did the above and it worked too for
> import std;
> std::meta::operators a = std::meta::op_plus_plus;
> too. Not added to testsuite because I thought we don't have a single
> import std; test anywhere.
Jason's working on that in a branch.
> git grep shows now
> gcc/testsuite/g++.dg/modules/stdns_b.C:import std;
> gcc/testsuite/g++.dg/modules/using-22_b.C:import std;
> Though with reflection it is more complicated because normally built
> std module with just -std=c++26 won't export reflection stuff, one
> needs to build the std module with -std=c++26 -freflection.
>
> Jakub
>
