Skip to content

Build.md

Latest commit

 

History

History
94 lines (68 loc) · 2.73 KB

Build.md

File metadata and controls

94 lines (68 loc) · 2.73 KB
Raw

Building LillyMol Python

The public LillyMol Python modules are built from the nanobind bindings. During local development, build them with Bazel and stage the shared objects into the repository lib directory used by run_python.sh.

Required Python Packages

Use a recent Python 3. Development has been exercised with Python 3.13, and the bindings are expected to work with other supported Python 3 versions.

python -m pip install nanobind absl-py protobuf numpy

At system level, make sure the Python headers and BLAS development libraries are installed. Package names vary by distribution; on Debian/Ubuntu systems this is typically:

sudo apt install python3-dev libblas-dev

numpy is needed for the array-returning fingerprint, descriptor, and coordinate helpers. protobuf is needed for proto-backed configuration objects and tests.

Optional Python Packages

If you use the xgboost QSAR model building tools in LillyMol:

python -m pip install xgboost scikit-learn matplotlib pandas

If you use the molecular property profile tool:

python -m pip install scipy numpy

Virtual environments and uv both work well. For example:

uv venv /path/to/lillymol_venv
uv pip install --python /path/to/lillymol_venv/bin/python numpy protobuf

Local Development

From the source checkout:

cd ${LILLYMOL_HOME}/src
bazel build -c opt //nanobind:all
./copy_shared_libraries.sh ../lib

Then run Python through the top-level wrapper, which sets PYTHONPATH and LD_LIBRARY_PATH for the staged extension modules and LillyMol shared libraries:

${LILLYMOL_HOME}/run_python.sh my_script.py

Use -c opt for normal testing and timing. Bazel's default fastbuild is much slower and can produce tiny floating point differences from the optimised command line tools.

Wheel Packaging

The wheel infrastructure lives in ${LILLYMOL_HOME}/python. It packages Bazel-built extension modules and private LillyMol shared libraries; setuptools does not compile the C++ code.

cd ${LILLYMOL_HOME}/src
bazel build -c opt //nanobind:all
./copy_shared_libraries.sh ../lib

cd ${LILLYMOL_HOME}/python
./scripts/stage_wheel_files.sh
python -m pip install build wheel setuptools
python -m build --wheel

The wheel is written to ${LILLYMOL_HOME}/python/dist. The generated python/prebuilt, python/build, and python/dist directories are build artifacts. Commit only the wheel-building infrastructure, not staged binaries or generated wheels.

The wheel should install the nanobind-backed lillymol module, the lillymol_bdb BerkeleyDB helper module when BDB support is built, and small compatibility modules such as lillymol_io and lillymol_tools that re-export from lillymol.