Build System for OCP

Recipe

The github action can be found in .github/workflows/build-ocp.yml

Wheels

The action creates two different types of delocated wheels:

1. cadquery_ocp which is build against pypi's VTK 9.2.6
2. cadquery_ocp_novtk which comes without VTK support

The wheels encapsulate the native OCP module into a folder OCP.

OCP
├── Adaptor2d
│   └── __init__.py
├── Adaptor3d
│   └── __init__.py
...
├── IVtk
│   └── __init__.py
├── IVtkOCC
│   └── __init__.py
├── IVtkTools
│   └── __init__.py
├── IVtkVTK
│   └── __init__.py
...
├── OCP.cpython-311-x86_64-linux-gnu.so
...

with __init__.py e.g. for Adaptor2d looking like

from ..OCP.Adaptor2d import *

OCP behaves exactly like the conda version, i.e. you can just ìmport OCP.BRep as usual. However, error traces can involve OCP.OCP.BRep ... prefixing the module with the OCP folder

Supported Operation Systems

The wheels are created for

• Windows (Intel)
• MacOS (Intel): running from macOS 11.11 or newer
• MacOS (arm64): running from macOS 11.11 or newer
• Linux (Intel): running Ubuntu 20.04 or newer (GLIBC_2.29 and GLIBCXX_3.4.26)

Supported Python Versions

• The vtk version can only be built for Python 3.11 and older, since this is the limitation of pypi VTK 9.2.6.
• The novtk version can be built for Python3.10 and newer, up to 3.13. However, currently only 3.11 is built.

Tests

• The vtk wheels are tested against build123d and `cadquery``
• The novtk wheels are tested against a patched version of build123d only (vtk support removed)

Known issues

• For macOS (Intel), nlopt 2.9 is not on pypi. The test installs nlopt from conda.
• For Windows, casadi and nlopt create a segmentation fault on exit (even when OCP and VTK are not installed). The test installs nlopt and casadi from conda.
• ocpsvg has cadquery_ocp as a dependency and will install parallel to the novtk wheel. The test uninstalls cadquery_ocp after installation of build123d.

Development

The action heavily caches artifacts since some steps can take 0.5 - 1.5 h.

For vtk wheels:

• VTK-9.2.6-py3.11-<os>-: The generated VTK SDK
• OCCT-7.7.2-VTK-ON-<os>-: The compiled OCCT SDK with VTK support
• OCP-source-7.7.2-VTK-ON-<os>-: The generated OCP source with VTK support
• OCP-7.7.2-VTK-ON-py<python version>-<os>-: The compiled OCP Python module with VTK support

For novtk wheels:

• OCCT-7.7.2-VTK-OFF-<os>-: The compiled OCCT SDK without VTK support
• OCP-source-7.7.2-VTK-OFF-<os>-: The generated OCP source without VTK support
• OCP-7.7.2-VTK-OFF-py<python version>-<os>-: The compiled OCP Python module without VTK support

<os> being "ubuntu-22.04", "macos-13", "macos-14", and "windows-2019".

To recompile, delete the respective cached elements first.

Note: Run this strategy first to create the caches:

    matrix:
    os: ["ubuntu-22.04", "macos-13", "macos-14", "windows-2019"]
    python-version: ["3.11"]
    # python-version: ["3.10", "3.11", "3.12", "3.13"]
    use-vtk: ["OFF", "ON"]

    # exclude:
    #   - use-vtk: ON
    #     python-version: 3.12
    #   - use-vtk: ON
    #     python-version: 3.13

Otherwise a lot of duplicate long lasting jobs will run. Then run:

    matrix:
    os: ["ubuntu-22.04", "macos-13", "macos-14", "windows-2019"]
    # python-version: ["3.11"]
    python-version: ["3.10", "3.11", "3.12", "3.13"]
    use-vtk: ["OFF", "ON"]

    exclude:
      - use-vtk: ON
        python-version: 3.12
      - use-vtk: ON
        python-version: 3.13