Testing¶

uv run pytest                 # the full suite
uv run pytest -k test_name    # one test

The suite runs in about two seconds. It targets 90%+ coverage, so a new code path usually needs a new test with it.

Two rules¶

The suite never reaches the network, and never runs Renovate. A test that needed a Renovate install and a GitHub token would not be a test. tests/helpers.py builds reports in the shape the real one has, and the tests swap them in for the scan.

The git tests use a real repository. Mocking git would test the mock. They build one in tmp_path instead, with GIT_CONFIG_GLOBAL pointed at a throwaway file so a developer's own commit.gpgsign or missing user.email cannot fail them.

What is where¶

File	Covers
`helpers.py`	Builders: sample files, updates, and reports.
`conftest.py`	Fixtures: the isolated git env and a test repository.
`test_main.py`	The command end to end, over a real repository.
`test_replace.py`	Templates and placing the edit.
`test_report.py`	Reading the report, holds, grouping.
`test_scan.py`	Which command runs, which token, what a failure says.
`test_files.py`	Reading and writing package files.
`test_git.py`	The clean check, staging, commits, hooks.
`test_lockfiles.py`	Finding and refreshing `uv.lock`.
`test_messages.py`	Commit subjects and bodies.

Settings that matter¶

These are in [tool.pytest.ini_options]:

--random-order runs tests in a shuffled order, so a test that depends on another one fails instead of passing by luck.
--numprocesses=logical runs them in parallel across cores.
filterwarnings = ["error"] makes warnings fail the suite. This tool exists to bump dependencies, so a new DeprecationWarning from a bump has to be seen and not scrolled past.
--cov-branch counts branch coverage, not just line coverage.

Writing one¶

Tests read as sentences and the docstring says why the behavior matters, not what the code does:

def test_split_applies_a_security_fix_even_while_it_is_pending() -> None:
    """A vulnerability alert bypasses Renovate's own hold, so it bypasses this one."""
    update = docker_update(pending=True, is_security=True)
    to_apply, held = split([update], with_major=False)
    assert to_apply == [update]
    assert held == []

Build updates with docker_update() or hook_update() from helpers.py and override the fields that matter to the test. They are the two shapes Renovate reports: one with a replaceString and a template, one with neither.

Tip

When you fix a bug, check the new test fails against the old code. A regression test that passes either way is not guarding anything.