Numba Integration Testing

As part of the continuous quality assurance of the Numba project we test the compatibility of Numba with the latests releases of specific dependents, i.e. libraries that make heavy use of Numba. The target Numba version to test with, will be the latest version of the git master branch where the test suite passed. Configurations for both CircleCI and Travis CI are provided but currently (April 2020) only the CircleCI configuration is active.

CircleCI:

This setup uses TexasBBQ under the hood.

Tested Projects

• umap
• librosa
• clifford
• awkward
• pydata/sparse
• fastparquet
• pygbm
• datashader
• tardis
• poliastro
• numba-dppy

Adding a new Target

In order to add a new target, you need to:

• Implement the logic for the target, by subclassing texasbbq.CondaTarget or texasbbq.GitTarget in the file switchboard.py and overloading necessary methods.
• Add an appropriate stanza in the CI configuration file .circleci/config.yml.
• Updating the README.rst to reflect the current list of projects being tested.
• Submit a pull-request on Github.

Testing a Branch or Pull-Request

Normally, Numba is installed as a conda packge from https://anaconda.org using a CondaSource configuration. However, sometimes it can be useful to run the integration-testing from a branch or a pull-request. The following configuration demonstrates how to obtain the branch refactor_it from the Github fork at github.com/esc/numba.

Please be advised that you must mirror the tags of the upstream Numba repository at github.com/numba/numba to the desired fork in such cases. This is because the Numba version is determined from the closest reachable tag in the Git history so recent tags must be present for the build system to accurately determine the Numba version. Otherwise you may end up with a nonsensical version number that is likely to cause confusion.

from texasbbq import GitSource


class NumbaSource(GitSource):

    module = __name__

    @property
    def name(self):
        return "numba"

    @property
    def clone_url(self):
        return "git://github.com/esc/numba"

    @property
    def git_ref(self):
        return "refactor_it"

    @property
    def conda_dependencies(self):
        return ["-c numba/label/dev llvmlite",
                "numpy pyyaml colorama scipy jinja2 cffi ipython ",
                "gcc_linux-64 gxx_linux-64",
                ]

    @property
    def install_command(self):
        return ("python setup.py build_ext -i && "
                "python setup.py develop --no-deps")

License

Copyright 2019 Anaconda Inc. Licensed under the terms of the BSD two-clause license. See the file LICENSE for details.