Contribution Guidelines#

For general info about how to contribute to LinkML, please see FAQ: Contributing.

Development Environment Setup#

LinkML is developed with Python and uses Poetry (version >= 2.0) for dependency management and packaging. If you have not used Poetry before refer to the Poetry documentation for installation instructions.

Then clone the repository and install the development dependencies:

git clone https://github.com/linkml/linkml
cd linkml
poetry install --all-extras

LinkML Testing Framework#

The LinkML test suite can be found in the tests folder. The tests are written under specific subfolders within the main aggregated tests/ suite.
If you want to formalize your test case into a unit test, you should decide the scope of your test and find the subfolder that the test might best fit under. For example, if your test is related to one of the LinkML generators, then you would create it under the tests/test_generators/ subfolder.
There are a few model LinkML schemas within the test suite that you can use while creating your unit tests. You will ideally find them in an input directory within the scoped tests folder you are looking at. For example, the input directory under the tests/test_generators/ subfolder. One of the main test schemas that most LinkML generator related unit tests use is the kitchen_sink.yaml.
Outside of generators, users will want to create issue specific unit tests. For that, you need to turn your attention to the tests/test_issues/ folder.
- The current direction is to first make an issue with specific instructions to reproduce the bug under the issues tab on the main linkml GitHub repo.
- Then, create a test case for the above issue, with a minimal test LinkML schema in the input folder of the tests/test_issues/ suite. A minimal test case entails that you only use the elements you need in the schema for your unit test and not bloat the test schema with more elements that necessary.
- Ideally, it is also advised not to bloat the output folders under each sub test suites with large autogenerated outputs. You can use the Python standard tests/test_generators/test_excelgen.py.

Directions for creating a unit test#

There are two cases you need to consider while writing your unit test. The first one is when you are writing a test for bugs in the LinkML library and the second is for enhancements. For bugs, you would typically create your test case in the tests/test_issues/ folder, whereas for enhancements, you would typically create it in the tests/test_generators/, tests/test_utils/, subfolder.

While creating a test for a bug, you should make a file called test_linkml_issue_NNN.py. After that it is perhaps easiest to just copy over a test from an existing issue test case and modify it.
- As discussed above, ideally, all your test issues will have an accompanying minimal test schema. One pattern is that you include the schema in your Python unit test file as strings. Another pattern is to include them as separate YAML files in the input folder.
As for new functionality, as directed above, create a test case under any of the existing Python test files if there are tests already scoped with respect to your enhancement. Or if not, then create a new Python test file with your unit test.
The tests in this repo are a mix of Python unittest tests and pytest tests. See below for more details on the transition. Because of this mix, tests should always be run using the pytest CLI.
If you have already activated the poetry virtual environment, then there is no need to use the poetry run prefix.
If you want to isolate specific test functions from within your Python unit test file, then you can use the -k command line option.

To run a single test file using pytest:

poetry run pytest tests/test_issues/test_linkml_issue_NNN.py

You can run the full test suite in the following way:

poetry run pytest

or via a shortcut Makefile target:

make test

When you create a Pull Request with your unit test on the linkml repo, a GitHub Action run will be set off which runs the entire test suite with the new test case that you added, on that test case branch too.

Note: You will see a number of issues which are named test_issue_NNN.py. The numbers and convention for those issues are with reference to the old biolinkml issue numbering convention.

Marks#

Use marks to run or exclude groups of tests:

network - (currently unused in CI) - tests marked as requiring network access/making network requests in order to succeed.
slow - tests that are necessary to test technical correctness but are sufficiently long that it’s worth excluding them during routine development/when running tests locally. By default, tests marked slow are not run, and require the --with-slow flag to run. Slow tests are included in the CI testing action. Typical use is to do most development work without running slow tests, and then running the full test suite --with-slow before submitting or merging a pull request.
skip, xfail - see skip and xfail docs

General Tips#

Always make sure to use assert statements to compare the expected value with the actual value, rather than simply printing or logging the expected and actual values.
Avoid using print statements for logging purposes. Use the logging module natively provided by Python appropriately with its various logging levels like DEBUG, INFO, ERROR, etc.
You can create a config file by copying the test_config.ini.example to a test_config.ini file and making changes, for example, to the logging levels:

[test.settings]
DEFAULT_LOG_LEVEL: logging.ERROR
DEFAULT_LOG_LEVEL_TEXT: ERROR

Never hardcode any file paths. Always use import variables such as INPUT_DIR, OUTPUT_DIR and use os.path.join() to make file paths. This ensures that the tests will run independent of the OS they are running on.
Running the test suite may change many files in the output folders accompanying the sub test suites which overloads the git status lists with unnecessary file changes. This should be avoided by not adding the output files changed to git status, removing them if they got added automatically, or implement the temporary file solution discussed previously.
To ignore the changed files run the shell script hide_test_changes.sh.
To reset all test output files back to original state use the shell script checkout_outputs.sh.

IDE Tips#

PyCharm, IntelliJ:

To run a single test:

Open to Run/Debug
+ to add test
Choose Python “tests > unittests”
In dialog:
- Target: Select Script path, browse to test script
- Choose Python interpreter
- Click ‘OK’ to save
Run or debug the test from the configurations menu

To run all tests:

Open to Run/Debug
+ to add test
Choose “Python tests > unittests”

Unittest to pytest conversion#

As of August 2023 this project has started converting its test suite from being based on the native unittest module to being based on pytest. Because of the presence of both styles of test in the codebase, it is recommended that you always use pytest to run tests.

New tests in any directory should be written using pytest.

Custom pytest fixtures#

input_path This fixture provides a factory function to get the path to a file within the input directory adjacent to the current test file. So for example with this file structure:
```
tests
└── test_foo
    ├── input
    │   └── schema_a.yaml
    └── test_foo_feature_a.py
```
You would access schema_a.yaml within your test like:
```
# test_foo_feature_a.py
def test_feature_a(input_path):
  path_to_schema = input_path("schema_a.yaml")
```
snapshot This fixture provides a way of comparing a string, temporary file, or temporary directory to a known, good artifact. The artifacts are stored in a directory called __snapshots__ alongside the test.

Warning: snapshot testing can be very powerful, but it can also lead to brittle tests. You should seriously consider alternatives before writing this type of test.

If you make a change that intentionally causes some output to not match the saved snapshot file(s), you should update the snapshots by running pytest with the --generate-snapshots flag. You should try to run only a single or small group of tests with this flag (as opposed to the entire test suite). An exception to this rule is when preparing a new minor version of linkml after the metamodel changes, changes to the metamodel can have many (inconsequential) changes to multiple snapshots. The updated snapshot files should be checked in to Git alongside your other code changes.

Debugging tip: sometimes a snapshot-based test may fail on GitHub actions, but may appear to pass locally. This can happen if the test is marked as a slow test, in which case you may need to use --generate-snapshots in combination with --with-slow (see below).

Testing linkml PRs with development versions of linkml-runtime#

linkml is tightly coupled to upstream linkml-runtime.

In some circumstances, paired changes need to be made against both linkml and linkml-runtime. Then testing with the last release of linkml-runtime is insufficient.

In such cases, you can specify that your PR needs to be tested with a specific linkml-runtime branch and repository. Specifying this information in the first two lines of your pull request´s opening message like this:

upstream_repo: user-or-org-name/linkml-runtime
upstream_branch: some-complicated-feature

Hey everyone … (PR continues)

The order of the lines with upstream_repo and upstream_branch tags does not matter, but they must be the first two lines of the pull request comment.

Maintainers can also specify upstream branches to test against when dispatching the test_with_unreleased_runtime workflow manually via the GUI prompt.

Testing against an unverified upstream branch is not necessarily dangerous. The input is stored as a variable first and not executed as untrusted code. But maintainers should take care to verify that the upstream branch and repo are correct and expected given the context of the PR.

Testing linkml-runtime PRs against any upstream linkml repository/branch#

For linkml-runtime a similar action as above is available allowing you to select a linkml repository to test against. For more see linkml-runtime/CONRBUTING.

Code formatting and linting#

This repository is configured to use Black and Ruff to ensure good formatting and code quality standards.

Each of these tools can be run on-demand via tox. To check the code for issues run:

poetry run tox -e lint

Any issues should be fixed before committing or pushing changes. This command is automatically run against pull requests and the testing workflow will fail if issues are found.

Formatting and some code quality issues can be fixed automatically by running:

poetry run tox -e format

You can configure these tools to run automatically before each commit by using pre-commit. To set this up, first ensure that you have the pre-commit package installed globally. This can be done via pipx:

pipx install pre-commit

Or one of their other supported installation options.

Then to enable the pre-commit hooks in this repository run:

pre-commit install

Now each git commit will trigger a check of the files staged in that commit and block the commit if issues were found. In an emergency you can bypass the pre-commit hooks by using the --no-verify flag on git commit.

Release to PyPI#

A GitHub action is set up to automatically release the package to PyPI. When it is ready for a new release, create a GitHub release. The version should be in the vX.X.X format following the semantic versioning specification.

Release notes should always be autogenerated (just click the “generate release notes” button) - these can be tweaked but the emphasis should be on making sure PR titles and githu issue titles are clear and concise. Note we don’t keep a separate CHANGELOG file, it all goes in the release notes.

After the release is created, the GitHub action will be triggered to publish to PyPI. The release version will be used to create the PyPI package.

If the PyPI release failed, make fixes, delete the GitHub release, and recreate a release with the same version again.

GitHub Best Practices#

PRs#

PRs should be in a DRAFT state until they are ready for review and tests are passing.
PRs should be reviewed by at least one other person.
- All automated tests should be passing via GitHub actions before a code review is requested.
- Reviews can be requested of any contributing member of the LinkML organization.
Make a DRAFT PR for your branch even if you’ve just started working on something. This gives other developers insight into your work and allows them to provide feedback early on.
In general, each PR should be associated with a ticket.

Ticket/Issue Creation#

Search existing issues before creating a new one.
Give your issue a short but actionable title
Describe the problem and the context and include a repeatable example.
Clearly state what needs to be done to close the ticket.