This project uses pytest and enforces 100% coverage on every pull request.
pip install -r requirements-dev.txt -r requirements-mcp.txt
pytest --cov=app --cov-report=term-missingThe MCP dependency lives in requirements-mcp.txt so the GitHub Action does
not have to install it, but the test suite covers app/mcp_server.py, so you
need both files to reach 100%.
To check the coverage gate the same way CI does:
coverage report --fail-under=100.github/workflows/tests.yml runs on every pull request and on every push to
main, against Python 3.12 and 3.13.
It fails if:
- any test fails, or
- total coverage drops below 100%
A pull request that lowers coverage will not pass CI. If you add code, add the tests that cover it in the same PR.
Grouped in classes, one per component:
| Class | Covers |
|---|---|
TestRepoManager |
repository count and listing, with pagination and error paths |
TestIssueManager |
issue extraction and field parsing, plus their error paths |
TestUtils |
pagination maths and list flattening |
TestAPIClient |
HTTP requests, HTTP errors and unexpected errors |
TestAPIError |
the custom exception's string representation |
TestTemplateManager |
grouping by language, Jinja2 rendering, and CSV/JSON output |
TestConfig |
template path resolution |
Grouped in classes, one per component:
| Class | Covers |
|---|---|
TestParseLabels |
reading the labels column, including unreadable values |
TestLoadIssues |
dataset normalization, and the missing dataset error path |
TestFilterIssues |
every filter, the sort order and the limit |
TestCountByLanguage |
language counts and their ordering |
TestCountByRepo |
repository counts, with and without a language |
TestDatasetError |
the custom exception's string representation |
TestConfig |
dataset path resolution, default and ISSUES_CSV |
The MCP tools, each against a dataset written to tmp_path and pointed at with
ISSUES_CSV: search_issues with and without filters, list_languages,
list_repositories, dataset_info, that all four tools are registered and
described for the model, and the __main__ block. load_dataset is also
covered without a local
file: the published CSV is fetched through a mocked requests.get (never a
real request), including cache hits, expiry, stale-cache fallback, and fetch
errors.
The end-to-end flow of the script: the normal run, the empty cases (no repos, no
issues), the --output flag, and the __main__ blocks of update_issues and
render_readme.
In app/tests/conftest.py:
mock_session— a mock ofrequests.Sessionmock_env_vars— patchesHEADERSandUSERNAMESinapp.update_issues
Local to test_update_issues.py:
mock_args— a mock of the parsed CLI arguments
Mock the API client rather than making real requests:
@patch('app.core.api_handler.APIClient')
def test_something(self, mock_api_client):
mock_api_instance = MagicMock()
mock_api_client.return_value = mock_api_instanceUse tmp_path for anything that touches the filesystem, and cover the error
path as well as the happy one — most classes here have both.