[ryanmrichard]

Empty commit to trigger a republish. First of 11 — this is the canary for the ecosystem-wide ABI rebuild.

Why

The published wheels were built on manylinux2014, which cannot emit the new libstdc++ ABI — it defines _GLIBCXX_USE_DUAL_ABI 0, so -D_GLIBCXX_USE_CXX11_ABI=1 is silently discarded. C++ consumers on gcc-14/clang-18 then fail to link the wheels' SDK:

undefined reference to `pluginplay::ModuleManager::at(std::__cxx11::basic_string<...> const&) const'
undefined reference to `pluginplay::utility::generate_uuid[abi:cxx11]()'
undefined reference to `utilities::printing::Demangler::demangle[abi:cxx11](char const*)'

No PR has ever built a manylinux wheel — build_pypi_dist runs only from merge.yaml, and test_pip_build doesn't use cibuildwheel — which is how this survived.

What this picks up

• bumps the wheel container for ABI .github#194 — manylinux_2_28 image + gcc-toolset-14
• NWChemEx/.github@620e081 — stops delocate vendoring a second OpenMPI into the macOS wheels (two MPI images in one process segfaulted at MPI_Init)
• nwxcmake 0.2.6 — ABI guard and stamp

Verified locally before opening this

• All 11 packages build on manylinux_2_28.
• Full-stack consumer (links nwx::scf, nwx::integrals, nwx::chemcache, nwx::nux) links and runs against the rebuilt wheelhouse on gcc-14 and clang-18.
• Control against the currently published wheels fails with the exact symbols above — so the check can actually detect the defect.
• macOS MPI fix verified A/B: default repair → segfault with CI's signature; patched repair → clean exit.

🤖 Generated with Claude Code

[github-actions]

🚀 [bumpr] Bumped!
New version:v0.1.47
Changes:v0.1.46...v0.1.47