The correct fix would be to set __cpp_lib_chrono == 201803 when all the types are defined.
On Wed, 30 Sept 2026 at 14:41, Jonathan Wakely <[email protected]> wrote: > > P.S. patches for libstdc++ need to be sent to the gcc-patches list, as > for all patches to GCC code. I've CC'd that list now. > > On Wed, 30 Sept 2026 at 14:34, Jonathan Wakely <[email protected]> wrote: > > > > On Wed, 30 Sept 2026 at 14:28, Jonathan Wakely <[email protected]> wrote: > > > > > > On Wed, 30 Sept 2026 at 14:21, <[email protected]> wrote: > > > > > > > > From: Hugo Beauzée-Luyssen <[email protected]> > > > > > > > > <chrono> gates its <string>/<vector>/<sstream>/shared_ptr/unique_ptr > > > > includes on __cpp_lib_chrono >= 201803L, which bits/version.h only > > > > defines when _GLIBCXX_USE_CXX11_ABI is set. But the actual > > > > chrono::tzdb/time_zone declarations further down the header are > > > > gated on the broader _GLIBCXX_HOSTED && (_GLIBCXX_USE_CXX11_ABI || > > > > !_GLIBCXX_USE_DUAL_ABI), deliberately kept available for old-ABI-only > > > > builds. > > > > > > > > libstdc++-v3/ChangeLog: > > > > > > > > * include/std/chrono: Fix missing headers when building for the > > > > old ABI > > > > > > > > Assisted-by: Claude (claude-sonnet-5) > > > > Signed-off-by: Hugo Beauzée-Luyssen <[email protected]> > > > > --- > > > > libstdc++-v3/include/std/chrono | 3 ++- > > > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > > > > > diff --git a/libstdc++-v3/include/std/chrono > > > > b/libstdc++-v3/include/std/chrono > > > > index 4116f1c815e..82402045e6f 100644 > > > > --- a/libstdc++-v3/include/std/chrono > > > > +++ b/libstdc++-v3/include/std/chrono > > > > @@ -57,7 +57,8 @@ > > > > # include <bits/stl_algo.h> // upper_bound > > > > # include <bits/range_access.h> // begin/end for arrays > > > > #endif > > > > -#if __cpp_lib_chrono >= 201803L // C++20 && HOSTED && USE_CXX11_ABI > > > > +#if __cpp_lib_chrono >= 201803L \ > > > > + || (defined __glibcxx_chrono_cxx20 && _GLIBCXX_HOSTED && > > > > !_GLIBCXX_USE_DUAL_ABI) > > > > > > This seems unnecessarily complicated. > > > > > > Why not just: > > > > > > #if defined __glibcxx_chrono_cxx20 && _GLIBCXX_HOSTED > > > > > > Doesn't that cover both cases? > > > > Worst case, we include some unnecessary headers for the old-ABI case > > in a dual-ABI build. > > > > > > > > > > > > # include <sstream> > > > > # include <string> > > > > # include <vector> > > > > -- > > > > 2.43.0 > > > > > > > >
