SpacePy uses GitHub Actions for continuous integration. Most of the relevant information is checked into the repository: the configuration file CI.yml manages the CI process, which ultimately runs the unit tests. However a few elements of the setup are not in the repository and are documented here. This may be useful if this ever has to be set up in the future, or if you want to run SpacePy CI tests on your fork before opening a pull request.
A workflow cannot be run until it has been run once against the
default branch (
master). This makes
it somewhat hard to test the workflow before merging; in SpacePy this was
handled by merging a tiny workflow first (PR 496).
Once this first run has been made, updated versions of the workflow
can be run from topic branches. It will show up under the
tab of a repository and the branch to use can be selected. It is also
possible to specify a branch by using the REST API; you will need an access token with
PRs require CI to pass before merging; this is managed with a branch
Settings from the tab at the top of a repository,
the left menu.) The relevant choices is “Require status checks to pass
before merging.” Every variant of the unit testing job (
ubuntu-18.04... etc.) will be in the list of checks; leave these alone and
select only the
All tests job. The name of this won’t change and it
will always depend on all the jobs in the workflow.
“Require branches to be up to date” should not be selected;
this encourages merging rather than our preferred rebase, and the tests
will run against a (temporary) merge regardless.
Rerunning CI on a PR¶
There is no way to manually trigger a workflow run on a pull request. SpacePy’s CI workflow is set up to trigger the workflow when a PR is marked ready for review, so one way to force a run is to mark the PR as a draft, and then as ready again.
Note that a PR will not trigger the CI if there is a merge conflict.
Dependencies for CI are stored in two caches: one for all pip dependencies, and one for the NASA CDF library. This minimizes CI time use for building dependencies.
Caches expire weekly (the week begins at 00 Monday, UTC). Caches can
also be force-expired by incrementing the versions in
the pip and/or CDF cache. Unfortunately this does require pushing a
SpacePy administrators can view the usage minutes, storage (for caches), etc. on our billing page.
- Doc generation date:
Sep 14, 2022