https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120775
--- Comment #37 from Jakub Jelinek <jakub at gcc dot gnu.org> --- (In reply to Boris Staletic from comment #36) > I've gone through all of the examples in P2996 and found five errors > preventing Marek's fork from compiling all of them. > > See: https://godbolt.org/z/chYM1zPbh > > To elaborate: > > - The first error is not related to reflections, but has to do with > constexpr reference to local. > - The second error, I'm not sure, but could also be outside of the scope of > reflections and more to do with expansion statements. > - The third error is finicky. Changing the formatting can make it go away. > It prevents, sometimes, expressions like `obj.[:members[Is]:]...`, where > `Is` is a pack from an index sequence. > - The fourth says `obj.[:member_refl:]` is ambiguous when `obj` has multiple > members named `_`. According to P2996, name lookup shouldn't be performed. > - Finally, there's an ICE when trying to call `access_context::unchecked()` > from inside a template. > > Regarding "no name lookup", P2996 says this: > > > Note that a “member access splice” like s.[:member_number(1):] is a more > > direct > > member access mechanism than the traditional syntax. It doesn’t involve > > member > > name lookup, access checking, or — if the spliced reflection value > > represents a > > member function — overload resolution. > > I also have a library that relies on C++26 reflections, but I have not yet > tried to compile it with Marek's fork, as patterns like those that trigger > the second and third error above do show up in my library. > Either way, the library is available at: > https://codeberg.org/bstaletic/pymetabind The ICE in tsubst_expr is likely PR122658. Regarding some cases of expansion statements, there is https://gcc.gnu.org/pipermail/gcc-patches/2025-November/700380.html patch waiting for review, but more importantly expansion statements are known not to work with define_static_array directly in the current expansion statement definition, see https://github.com/cplusplus/CWG/issues/805 I have a patch in testing for that but I think the CWG needs to decide first what they want. A workaround that works for me is template for (constexpr auto m : (const std::span <const ...> define_static_array (...)). The "_" case is a known issue, see e.g. https://forge.sourceware.org/marek/gcc/src/branch/reflection/gcc/testsuite/g++.dg/reflect/member15.C mentioning it or the https://forge.sourceware.org/marek/gcc/commit/c9bc89c30bd7ade9e9c7d5d908c0d3272675ec08 commit message for details, I hope we'll work on that next week.
