fix: use cibuildwheel for building PyPI wheel

.github/workflows/build-wheel-release-upload.yml

@@ -23,12 +23,16 @@ jobs:
   #
   # Choose this for open-source packages distributed via PyPI and/or
   # conda-forge where broad public availability is the goal.
+  #
+  # build_c_ext_with_cibw builds the wheels with cibuildwheel using
+  # [tool.cibuildwheel] in pyproject.toml.
   build-release:
     uses: scikit-package/release-scripts/.github/workflows/_build-wheel-release-upload.yml@v0
     with:
       project: pyobjcryst
       submodules: "true"
       c_extension: true
+      build_c_ext_with_cibw: true
       maintainer_github_username: sbillinge
     secrets:
       PYPI_TOKEN: ${{ secrets.PYPI_TOKEN }}

CMakeLists.txt

@@ -152,10 +152,21 @@ target_compile_definitions(ObjCryst PUBLIC
 target_link_libraries(ObjCryst PUBLIC cctbx newmat)
 
 # Put VFNDebug.cpp first so debug-level constants are defined before use
-set_source_files_properties(
-    "${OBJCRYST_SRC}/Quirks/VFNDebug.cpp"
-    PROPERTIES COMPILE_FLAGS "-O0"
-)
+if(NOT MSVC)
+    set_source_files_properties(
+        "${OBJCRYST_SRC}/Quirks/VFNDebug.cpp"
+        PROPERTIES COMPILE_FLAGS "-O0"
+    )
+endif()
+
+# MSVC: same settings as the libobjcryst Windows build (/EHsc, M_PI etc.),
+# plus /bigobj for the large binding translation units.
+if(MSVC)
+    foreach(_tgt newmat cctbx ObjCryst)
+        target_compile_definitions(${_tgt} PUBLIC _USE_MATH_DEFINES NOMINMAX)
+        target_compile_options(${_tgt} PUBLIC /EHsc /bigobj)
+    endforeach()
+endif()
 
 # ---------------------------------------------------------------------------
 # nanobind extension: _pyobjcryst
@@ -178,3 +189,11 @@ target_link_libraries(_pyobjcryst PRIVATE ObjCryst)
 
 # Install into pyobjcryst package
 install(TARGETS _pyobjcryst LIBRARY DESTINATION pyobjcryst)
+
+# Licenses of the statically linked third-party code ship with the wheel.
+install(FILES "${CMAKE_CURRENT_SOURCE_DIR}/extern/objcryst/LICENSE.txt"
+        DESTINATION pyobjcryst/licenses/objcryst)
+install(FILES "${CCTBX_SRC}/LICENSE_2_0.txt" "${CCTBX_SRC}/COPYRIGHT_2_0.txt"
+        DESTINATION pyobjcryst/licenses/cctbx)
+install(FILES "${NEWMAT_SRC}/readme.txt" "${NEWMAT_SRC}/nm11.htm"
+        DESTINATION pyobjcryst/licenses/newmat)

nanobind_migration_notes.md

@@ -40,15 +40,18 @@ Known, accepted gaps — real, not oversights, but still open:
   The `Molecule`↔`MolAtom` cycle is fixed. `Crystal`↔`Scatterer`/`Molecule`/`ScatteringPower` cycles
   remain (56 leaked instances in the current full-suite run) — not safe to fix the same way, see
   [keep_alive Reference Cycles](#keep_alive-reference-cycles). Not blocking functionality.
-- **CI has not been exercised against this build.** `.github/workflows/` is inherited from `main`
-  unmodified (it calls a shared reusable workflow); it has never actually run against the
-  nanobind/CMake build because nothing on this branch has been pushed yet.
-- **Cross-platform**: everything above has been built and tested on Linux only. macOS/Windows are
-  unverified.
+- **Wheel builds**: PyPI wheels are built with `cibuildwheel` by the shared release workflow,
+  configured in `[tool.cibuildwheel]` in `pyproject.toml`. The old shared workflow stripped the
+  libraries `auditwheel` had bundled, so those wheels only worked inside a conda environment.
+  The cibuildwheel version of the shared workflow is not released yet:
+  `build-wheel-release-upload.yml` points at `SYMSCAE/release-scripts@cibuildwheel` for now and
+  must be switched back to `scikit-package/release-scripts@v0` once it is.
+- **Cross-platform**: Linux, macOS (x86_64/arm64) and Windows wheels are built and tested by
+  cibuildwheel in the release workflow.
 - **Performance**: not benchmarked against Boost.Python.
-- **Versioning**: `pyproject.toml` uses a static version (`2026.2.0rc0`) instead of
-  `main`'s dynamic tag-based versioning (`setuptools-git-versioning`) — scikit-build-core needs a
-  different mechanism for that; flagged as a `TODO` in `pyproject.toml` itself, not yet resolved.
+- **Versioning**: the version comes from the git tag via `setuptools-scm`
+  (`metadata.version.provider = "scikit_build_core.metadata.setuptools_scm"`), replacing
+  `main`'s `setuptools-git-versioning`. Builds need the tags (`fetch-depth: 0` in CI).
 
 None of the above blocks day-to-day use of the library from Python — every method that was
 reachable under Boost.Python is reachable under nanobind (modulo the override caveat above, which

news/cibuildwheel-wheels.rst

@@ -0,0 +1,25 @@
+**Added:**
+
+* <news item>
+
+**Changed:**
+
+* Build the PyPI wheels with ``cibuildwheel`` (manylinux, macOS x86_64/arm64 and Windows AMD64), configured in ``[tool.cibuildwheel]`` in ``pyproject.toml``.
+* The package version is now derived from the git tag via ``setuptools-scm``.
+
+**Deprecated:**
+
+* <news item>
+
+**Removed:**
+
+* <news item>
+
+**Fixed:**
+
+* PyPI wheels are now self-contained and no longer need ``libObjCryst`` or ``libboost_python`` from a conda environment; previously ``import pyobjcryst`` failed in a plain virtual environment.
+* Declare ``packaging`` as a runtime dependency (used by ``pyobjcryst.powderpattern``).
+
+**Security:**
+
+* <news item>

news/nanobind-packaging.rst

@@ -5,7 +5,6 @@
 **Changed:**
 
 * Boost headers are now taken from the PyPI ``libboost-headers`` package when building with pip, falling back to system/conda Boost headers (or ``-DBOOST_INCLUDE_DIR=<path>``).
-* Set package version to ``2026.2.0rc0``.
 
 **Deprecated:**
 

pyproject.toml

@@ -1,14 +1,16 @@
 [build-system]
-requires = ["scikit-build-core>=0.9", "nanobind", "numpy", "libboost-headers"]
+requires = [
+    "scikit-build-core>=0.10",
+    "setuptools-scm>=8",
+    "nanobind",
+    "numpy",
+    "libboost-headers",
+]
 build-backend = "scikit_build_core.build"
 
 [project]
 name = "pyobjcryst"
-# TODO: main uses setuptools-git-versioning (dynamic, tag-based) with
-# setuptools.build_meta; scikit-build-core needs a different dynamic-version
-# provider to match that exactly. Pinned as a static placeholder for now --
-# revisit before release, see nanobind_migration_notes.md.
-version = "2026.2.0rc0"
+dynamic = ["version"]
 authors = [
   { name="Christopher Farrow" },
   { name="Pavol Juhás" },
@@ -22,7 +24,7 @@ description = "Python bindings to the ObjCryst++ library."
 keywords = ['objcryst', 'atom structure crystallography', 'powder diffraction']
 readme = "README.rst"
 requires-python = ">=3.12, <3.16"
-dependencies = ["numpy"]
+dependencies = ["numpy", "packaging"]
 classifiers = [
         'Development Status :: 5 - Production/Stable',
         'Environment :: Console',
@@ -52,8 +54,46 @@ Issues = "https://github.com/diffpy/pyobjcryst/issues/"
 pyobjcryst = "pyobjcryst.pyobjcryst_app:main"
 
 [tool.scikit-build]
+minimum-version = "build-system.requires"
 cmake.build-type = "Release"
 wheel.packages = ["src/pyobjcryst"]
+sdist.cmake = true
+sdist.exclude = [
+    "build",
+    "dist",
+    ".pytest_cache",
+    # Parts of the ObjCryst++ submodule not needed to build the extension.
+    "extern/objcryst/Fox",
+    "extern/objcryst/test",
+    "extern/objcryst/ObjCryst/doc",
+    "extern/objcryst/ObjCryst/example",
+]
+metadata.version.provider = "scikit_build_core.metadata.setuptools_scm"
+
+[tool.setuptools_scm]
+version_scheme = "no-guess-dev"
+
+# PyPI wheels: ObjCryst++, cctbx and newmat are built from extern/ and linked
+# statically, Boost is header-only and nanobind is linked statically, so the
+# wheels install into any environment without conda or system libraries.
+[tool.cibuildwheel]
+build = ["cp312-*", "cp313-*", "cp314-*", "cp315-*"]
+skip = ["*-musllinux_*"]
+test-requires = ["pytest"]
+# The tests open tests/testdata/* relative to the working directory.
+test-sources = ["tests"]
+test-command = "pytest tests"
+
+[tool.cibuildwheel.linux]
+archs = ["x86_64"]
+
+[tool.cibuildwheel.macos]
+archs = ["native"]
+environment = { MACOSX_DEPLOYMENT_TARGET = "11.0" }
+
+# No Windows ARM64 wheels.
+[tool.cibuildwheel.windows]
+archs = ["AMD64"]
 
 [tool.codespell]
 exclude-file = ".codespell/ignore_lines.txt"

requirements/pip.txt

@@ -1,3 +1,4 @@
 setuptools
 numpy
+packaging
 libboost-headers