We eagerly welcome contributions of any type (e.g., bug fixes, new features, reporting issues, documentation, etc). If you're looking for a good place to get started you might like to peruse our current Git issues (those marked with help wanted are a good place to start).
Please be aware of Parsl's Code of Conduct.
If you are not familiar with GitHub pull requests, the main mechanism to contribute changes to our code, there is documentation available.
The best places to ask questions or discuss development activities are:
- in our Slack's #parsl-hackers channel. You can join our Slack here.
- using GitHub issues.
Parsl code should adhere to Python PEP-8. This is enforced in CI (with some exceptions). You can also run this test yourself using make flake8.
The following convention should be followed: ClassName, ExceptionName, GLOBAL_CONSTANT_NAME, and lowercase_with_underscores for everything else.
Parsl follows the calendar versioning scheme with YYYY.MM.DD numbering scheme for versions.
This scheme was chosen following a switch from ad-hoc versioning and manual release processes to an automated weekly process.
Releases are pushed from github actions to PyPI and will be picked up automatically by Conda.
Manual packaging instructions are included in the
packaging docs
Classes should be documented following the NumPy/SciPy style. A concise summary is available here. User and developer documentation is auto-generated and made available on ReadTheDocs.
Parsl uses pytest to run most tests. All tests should be placed in
the parsl/tests directory.
There are two broad groups of tests: those which must run with a specific configuration, and those which should work with any configuration.
Tests which should run with with any configuration live under
themed directories parsl/tests/test*/ and should be named test*.py.
They can be run with any configuration, by specifying --config CONFIGPATH
where CONFIGPATH is a path to a .py file exporting a parsl configuration
object named config. The parsl-specific test fixtures will ensure
a suitable DFK is loaded with that configuration for each test.
Tests which require their own specially configured DFK, or no DFK at all,
should be labelled with @pytest.mark.local and can be run with
--config local.
Provide the special configuration creating a local_config function
that returns the required configuration in that test file.
Or, provide both a local_setup function that loads the proper configuration
and local_teardown that stops parsl.
There is more fine-grained enabling and disabling of tests within the above categories:
A pytest marker of cleannet (for clean network) can be used to select
or deselect tests which need a very clean network (for example, for tests
making FTP transfers). When the test environment (github actions) does not
provide a sufficiently clean network, run all tests with -k "not cleannet" to
disable those tests.
Some other markers are available but unused in testing;
see pytest --markers parsl/tests/ for more details.
A specific test in a specific file can be run like this::
$ pytest test_python_apps/test_basic.py::test_simple
A timeout can be added to test runs using a pytest parameter such as
--timeout=60
Many tests are marked with @pytest.mark.skip for reasons usually
specified directly in the annotation - generally because they are broken
in one way or another.
There is also some coverage testing available. The CI by default records coverage for most of the tests that it runs and outputs a brief report at the end of each CI run. This is purely informational and a Lack of coverage won't produce a CI failure.
It is possible to produce a more detailed coverage report on your development machine: make sure you have no .coverage file, run the test commands as shown in .github/workflows/ci.yaml, and then run coverage report to produce the summary as seen in CI, or run coverage html to produce annotated source code in the htmlcov/ subdirectory. This will show, line by line, if each line of parsl source code was executed during the coverage test.
If you are a contributor to Parsl at large, we recommend forking the repository and submitting pull requests from your fork. The Parsl development team has the additional privilege of creating development branches on the repository. Parsl development follows a common pull request-based workflow similar to GitHub flow. That is:
- every development activity (except very minor changes, which can be discussed in the PR) should have a related GitHub issue
- all development occurs in branches (named with a short descriptive name, for example, add-globus-transfer-#1)
- the master branch is always stable
- development branches should include tests for added features
- development branches should be tested after being brought up-to-date with the master (in this way, what is being tested is what is actually going into the code; otherwise unexpected issues from merging may come up)
- branches what have been successfully tested are merged via pull requests (PRs)
- PRs should be used for review and discussion
- PRs should be reviewed in a timely manner, to reduce effort keeping them synced with other changes happening on the master branch
Git commit messages should include a single summary sentence followed by a more explanatory paragraph. Note: all commit messages should reference the GitHub issue to which they relate. A nice discussion on the topic can be found here.
Implemented Globus data staging support Added the ability to reference and automatically transfer Globus-accessible files. References are represented using the Parsl file format “globus://endpoint/path/file.” If Globus endpoints are known for source and destination Parsl will use the Globus transfer service to move data to the compute host. Fixes #-1.
Developers may find it useful to setup a pre-commit git hook to automatically lint and run tests. This is a script which is run before each commit. For example:
$ cat ~/parsl/.git/hooks/pre-commit #!/bin/sh make lint flake8 mypy local_thread_test
All project documentation is written in reStructuredText. Sphinx is used to generate the HTML documentation from the rst documentation and structured docstrings in Parsl code. Project documentation is built automatically and added to the Parsl documentation.
Parsl wants to make sure that all contributors get credit for their contributions. When you make your first contribution, it should include updating the codemeta.json file to include yourself as a contributor to the project.