| name | python-testing-strategies |
| description | Testing patterns for FastAPI with pytest-asyncio, httpx AsyncClient, fixtures, and test data factories. Use when writing tests, setting up test infrastructure, or improving coverage in a FastAPI project. |
| allowed_tools | ["Read","Write","Edit","Glob","Grep","Bash"] |
Python Testing Strategies — Quick Reference
Setup (CRITICAL)
Set asyncio_mode = "auto" in pyproject.toml to auto-detect async tests without markers. (The default "strict" mode also works if you mark tests with @pytest.mark.asyncio.)
See examples.md for the complete conftest.py and pyproject.toml setup.
Test Client
Use httpx.AsyncClient + ASGITransport — never TestClient for async tests.
See examples.md for client fixture setup.
Required Coverage Per Endpoint
- Happy path (create, read, update, delete)
- Not found (404)
- Validation error (422)
- Auth failure (401) — if protected
- Soft delete — deleted records not returned
Test Naming
async def test_create_item():
async def test_create_item_missing_name():
async def test_get_item_not_found():
async def test_get_item_soft_deleted():
Test Data Factories
Use simple factory functions with sensible defaults. Override only what the test cares about.
See examples.md for factory patterns.
Running Tests
pytest
pytest tests/test_items.py
pytest -k "test_create"
pytest --tb=short -q
Gotchas
-
Missing asyncio_mode = "auto" silently breaks everything. Tests appear to pass (0 collected) or hang indefinitely. Add [tool.pytest.ini_options] asyncio_mode = "auto" to pyproject.toml before writing any async test. This is the #1 cause of "my tests don't run."
-
TestClient and AsyncClient are not interchangeable. TestClient (from Starlette) is synchronous — it blocks. AsyncClient (from httpx) is async — it uses the event loop. If your app uses async def routes with await, you must use AsyncClient with ASGITransport. Using TestClient may hide async bugs because it runs synchronously.
-
Database state leaks between tests without proper cleanup. Each test needs a clean database. Use an autouse=True fixture that creates/drops tables, or truncate between tests. Without this, test order matters and random failures appear in CI.
-
response.json() returns snake_case from FastAPI by default. If your Pydantic models use alias_generator, the JSON keys may differ from your Python field names. Always assert against the actual JSON keys, not the Python attribute names.
-
Forgetting to await in async tests gives confusing errors. If you see <coroutine object ...> in test output instead of actual data, you forgot an await. Every client.get(), client.post(), etc. must be awaited.