Core

Core is a fork of Apollo cyber, a publish-subscribe system used as middleware for autonomous driving.

Benchmark

Latency of publish and subscribe messages.

How to build

1. Deploy build env.

sudo bash scripts/deploy/build.sh

2. Run the build script.

bash scripts/build.sh

For a narrower pycyber-focused build/test loop, use the existing Bazel targets directly:

bazel test --color=no --curses=no \
  //cyber/python/internal:py_cyber_test \
  //cyber/python/internal:py_record_test \
  //cyber/python/cyber_py3/test:init_test \
  //cyber/python/cyber_py3/test:cyber_test \
  //cyber/python/cyber_py3/test:cyber_time_test \
  //cyber/python/cyber_py3/test:record_test

Example

1. Set Environment Variable

   source scripts/env/runtime.bash

2. Run the Publisher and Subscriber

   • In the first terminal, run ./bazel-bin/cyber/examples/listener to start the subscriber program.
   • In the second terminal, run ./bazel-bin/cyber/examples/talker to start the publisher program.

3. Check the log in data/log

Build the pycyber wheel

Recommended: use the included one-step packaging script which stages, builds, repairs and validates the wheel. Always run packaging inside an isolated Python environment (venv) or a clean container to avoid dependency conflicts.

Quick start (recommended)

1. Create and activate a virtual environment (venv):

python3 -m venv .packaging-venv
source .packaging-venv/bin/activate
python -m pip install --upgrade pip
pip install build auditwheel twine setuptools_scm
# On Linux, also install patchelf
sudo apt-get update && sudo apt-get install -y patchelf

2. Run the one-step packaging script. The script will use PYCYBER_VERSION if set, otherwise it reads the most recent tag matching pycyber-v*:

# Preferred: use tag-derived version (make and push a tag first)
./scripts/release/build_and_package_pycyber.sh

# Or override version explicitly:
PYCYBER_VERSION=0.0.7 ./scripts/release/build_and_package_pycyber.sh

Artifacts and checks:

• Artifacts are written to packaging/pycyber/wheelhouse/.
• The script runs twine check and the repository's validate_pycyber.py smoke test.
• A SHA256SUMS file is created in the wheelhouse.

Script flags:

• --skip-auditwheel : Skip auditwheel repair (useful on non-Linux machines).
• --skip-validate : Skip running the smoke validation.
• --skip-sdist : Only build wheel, skip sdist.
• --skip-build : Skip staging (build_pycyber.py), useful if staging already exists.
• --outdir DIR : Write artifacts to a custom directory.

Manual (step-by-step) packaging

If you prefer to run the steps manually, follow these commands (run inside a clean venv / container):

1. Stage the package from Bazel outputs:

python3 scripts/release/build_pycyber.py

2. Build sdist + wheel:

python3 -m build --sdist --wheel --outdir packaging/pycyber/dist packaging/pycyber

3. Repair Linux wheels (auditwheel):

auditwheel repair packaging/pycyber/dist/*.whl -w packaging/pycyber/wheelhouse

4. Check metadata and run validation:

python3 -m twine check packaging/pycyber/wheelhouse/*
python3 scripts/release/validate_pycyber.py --wheel packaging/pycyber/wheelhouse/*.whl

5. Upload to TestPyPI (recommended first):

TWINE_USERNAME=__token__ TWINE_PASSWORD=$TEST_PYPI_TOKEN \
  python -m twine upload --repository testpypi packaging/pycyber/wheelhouse/*

# Test install from TestPyPI in a fresh venv:
pip install --index-url https://test.pypi.org/simple/ --no-deps pycyber==0.0.7

6. Final upload to PyPI:

TWINE_USERNAME=__token__ TWINE_PASSWORD=$PYPI_TOKEN \
  python -m twine upload packaging/pycyber/wheelhouse/*

Versioning

• pyproject.toml is configured to use setuptools_scm with the tag pattern ^pycyber-v(?P<version>.+)$.
• Preferred release flow: create an annotated tag on the release commit and push it:

git tag -a pycyber-v1.2.3 -m "Release 1.2.3"
git push origin --tags

• The one-step script will extract the version from the tag automatically. If needed, override with PYCYBER_VERSION.

Troubleshooting

• pip dependency conflicts (for example: streamlit vs protobuf) occur when build tools are installed into the global Python environment. Always use a clean virtualenv (venv) or container for packaging.
• For reproducible manylinux wheels and multi-Python builds, use cibuildwheel or the official manylinux Docker images: https://github.com/pypa/manylinux
• If auditwheel cannot patch RPATH because patchelf is missing, install patchelf (Debian/Ubuntu: apt-get install -y patchelf) or run the packaging inside an appropriate Linux build environment.

Security and best practices

• Use PyPI API tokens (username __token__) with minimal scope; store tokens as secrets in CI.
• Upload to TestPyPI first to verify installation before publishing to the real PyPI.
• Tag releases and attach built artifacts and SHA256SUMS to GitHub Releases.
• Optionally GPG-sign artifacts for extra authenticity.

CI

• The repository includes .github/workflows/release-pycyber.yml which is triggered by tags pycyber-v*. The workflow checks out the full history (fetch-depth: 0) so setuptools_scm can compute versions correctly.