A repository hosting prebuilds of Python packages suitable for installing on an Opentrons robot.
This repository is under construction and the code here has not yet been released.
The OT-2 uses a custom Buildroot-created Linux system running on a Raspberry Pi. Because it is a custom system, it cannot use binary python packages made with Raspbian in mind; because it is an arm7hf architecture, it cannot use more general manylinux wheels made with x86-64 or arm64 in mind. Since the OT-2 itself does not have a compiler, we use this repository to provide prebuilt python wheels that are installable on an OT-2 by using a PEP-503 simple package index.
Any package identified here can be installed on an OT-2 through pip.
Please open an issue if there is a Python package that you want that is not present, or if there is a specific version of a Python package that the index does not have. Please also feel free to open a pull request to add it! Before requesting a package or new version, please try installing it on an OT-2 first; only packages with native code components need to be here, since pure-python packages are installable from PyPI.
Be aware that the OT-2 system can't always handle the addition of a specific package. For instance, the package might be a thin wrapper around a system library that is not available on the OT-2; or it might have a toolchain that does not allow cross-compilation with the tools we have available.
This repo is intended to work in production on github CI, but you can test builds locally (and should, since some packages can have surprising build system requirements). The entry point for building is ./build-packages. It's a python script that should work on anything above python 3.7. You can get usage by running ./build-packages --help.
You'll also need to have Docker installed and running and accessible with how you call ./build-packages (i.e., if you don't want to use sudo then the docker socket should be user-accessible).
This is tested and maintained for x64 docker. It should work if you run it from
- An x86_64 linux host
- Including WSL
- An x86_64 OSX host
It might work on a native Windows x86_64 host.
It definitely will not work on any machine architecture that presents the docker container with an arm64 environment (e.g. M1/M2 macs, phones, embedded devices). The buildroot SDK that powers the package builds needs to run in an x86_64 architecture. If that's all you have, you'll have to run it in a cross-architecture vm.
The build tools are a separate package that is developed in this repo in ./tools - see that directory for more.
To add a package to the repo, make a new set of directories in the package/ directory for it. A best practice is to make that directory named package_name/version - for instance, package/pandas/1.5.0. For instance, if you wanted to add another version of pandas, you could add package/pandas/2.0.0.
In that directory, make a python script called build.py. Its job is to call build.build_package with correct arguments. A good example is package/pandas/1.5.0/build.py. build.build_package lives in tools/builder/package_build/orchestrate.py if you want to look at the source.
The source argument takes an object describing where to get the package. There is a helper function to build it: builder.package_build.github_source, which lets you specify
- the github org (i.e.
pandas-devfor pandas) - the github repo (i.e.
pandasfor pandas) - a git tag to find
- some other options about how the package is distributed and what kind of source it is. Check out
tools/builder/package_build/__init__.pyto see more.
Then, you set the setup_commands (typically these will be build_ext and bdist_wheel, but it depends on the package) and any build dependencies. Build dependencies are probably listed in the package metadata; they may be there as setup_depends or pyproject toml build system requirements. They may also just be assumed to be present. You can figure out what's required by reading the package code, or by trying to build it in an empty venv.
Finally, try a build with ./build-packages.
See the README.md in tools/!