Python Testing

Python testing conventions

pytest is the whole toolkit — plain assert, fixtures, parametrization, and Hypothesis for property tests. No unittest.TestCase boilerplate, no assertion DSLs. Tests live in tests/ and run with uv run pytest.

1. Parametrize instead of repeating

One test, a row per case — the table-driven pattern. Adding a behavior is adding a row:

import pytest

from myapp.kv import parse

@pytest.mark.parametrize(
    ("line", "want_key", "raises"),
    [
        ("a=b", "a", False),
        ("ab", None, True),       # missing separator
        ("=b", None, True),       # empty key
        ("a=", "a", False),       # empty value is fine
    ],
)
def test_parse(line: str, want_key: str | None, raises: bool) -> None:
    if raises:
        with pytest.raises(ValueError):
            parse(line)
    else:
        assert parse(line)[0] == want_key

Name each test for the behavior it pins down; assert directly — pytest rewrites it into a rich failure message.

2. Fixtures for setup, built-ins first

Shared setup is a @pytest.fixture; teardown goes after a yield. Reach for the built-in fixtures before inventing your own — tmp_path (a real temp directory), monkeypatch (patch env/attrs, auto-reverted), capsys (capture stdout/stderr). Put cross-file fixtures in tests/conftest.py.

3. Fakes over mocks

Prefer a small hand-written fake — a class with canned returns satisfying the consumer's Protocol (see python-typing) — to unittest.mock.MagicMock. Assert on observable behavior (return values, recorded state), not on which methods were called. Use monkeypatch to swap a dependency at a boundary; reserve mock for third-party seams you do not own, and avoid asserting call counts — they couple tests to implementation.

4. Property tests with Hypothesis

Every parser, encoder, or validator handling untrusted input gets a Hypothesis test — the analogue of Go's native fuzzing. Generate inputs and assert the invariants that must hold for any input (no crash, round-trips, never emits an invalid result):

from hypothesis import given, strategies as st

from myapp.kv import parse

@given(st.text())
def test_parse_never_crashes(s: str) -> None:
    try:
        key, _ = parse(s)
    except ValueError:
        return  # rejecting bad input is fine
    assert key != ""  # but a parsed key is never empty

Hypothesis shrinks any failing case to a minimal example and records it, so the regression replays on every run. Use @given with explicit strategies; seed known tricky cases with @example.

5. Invocations

uv run pytest                 # the suite (unit + property tests)
uv run pytest -q              # quiet, for the make/CI gate
uv run pytest -k name         # run tests matching an expression
uv run pytest --cov=myapp     # coverage (needs the pytest-cov dev dependency)

make test runs uv run pytest (see python-project); CI runs the same (see python-release). For the interfaces that make code testable see python-typing; for house style see python-style.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: bitwise-media-group
• Source: bitwise-media-group/skills
• License: MIT

Install and usage instructions live in the source repository linked above.