Where to start
We 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.
Parsl code should adhere to Python pep-8. Install flake8 and run the following code to identify non-compliant code:
$ flake8 parsl/
Note: the continuous integration environment will validate all pull requests using this command.
The following convention should be followed: ClassName, ExceptionName, GLOBAL_CONSTANT_NAME, and lowercase_with_underscores for everything else.
Parsl follows the
major.minor[.maintenance[.build]] numbering scheme for versions. Once major features
for a specific milestone (minor version) are met, the minor version is incremented and released via PyPI and Conda.
Fixes to minor releases are made via maintenance releases. Packaging instructions are included in the
pytest to run most tests. All tests should be placed in
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 run with a specific configuration live under the
They can be launched with a pytest parameter of
--config local and each test file should initialise a DFK
Tests which should run with with any configuration live under
parsl/tests/test*/ and should be named
They can be run with any configuration, by specifying
where CONFIGPATH is a path to a
.py file exporting a parsl configuration
config. The parsl-specific test fixtures will ensure
a suitable DFK is loaded with that configuration for each test.
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). Travis does not provide a sufficiently clean
network and so .travis.yml runs all tests with
-k "not cleannet" to
disable those tests.
A pytest marker of
issue363 can be used to select or deselect tests
that will fail because of issue 363 when running without a shared file
Some other markers are available but unused in Travis testing;
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
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 .travis.yml, 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 which includes the associated issue number, 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 flake8 parsl nosetests -vx parsl/tests/test_threads parsl/tests/test_data parsl/tests/test_checkpointing
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.
Credit and Contributions
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.
Discussion and Support
The best way to discuss development activities is via Git issues.
To get involved in community discussion please join the Parsl Slack channel.