testing-python（测试 Python）

Testing Python

Tests prove features work. Coverage is secondary. E2e tests beat unit tests. Real beats mocked.

Philosophy

• Trustworthiness > coverage. A test that mocks away the tested thing proves nothing.
• 5 good e2e tests > 100 unit tests with heavy mocking.
• Pareto principle. Write the fewest tests that cover 80% of what matters. E2e tests naturally do this.
• Unit tests for pure logic only. Functions that transform data honestly.
• Real over mocked. Real HTTP servers (pytest-httpserver), real tmp dirs, real processes.

Test Planning

When writing new tests, plan before coding:

1. List all potential test cases for the feature
2. Categorize each as critical, medium, or small importance
3. Discard small-importance cases — not worth the maintenance cost
4. Write remaining cases in plain text to docs/plans/test-cases-<feature>.md
5. Only then write test code

Test Priority

1. CLI / e2e tests — run actual commands, check output + exit codes
2. Integration tests — component interaction through public API
3. Unit tests — pure data transformation functions
4. Skip — framework glue, UI layout, trivial getters

Project Setup

Directory Structure

tests/
├── unit/              # Pure function tests
├── integration/       # CLI tests, component interaction
├── fixtures/          # Shared test data and helpers
└── conftest.py        # Shared fixtures

pyproject.toml Configuration

[tool.pytest.ini_options]
testpaths = ["tests"]
python_files = ["test_*.py"]
python_classes = ["Test*"]
python_functions = ["test_*"]
asyncio_mode = "auto"
addopts = ["-n", "auto", "--dist", "worksteal"]

markers = [
    "unit: unit tests",
    "integration: integration tests",
]

Dependencies

[dependency-groups]
dev = [
    "pytest>=9.0.1",
    "pytest-xdist>=3.5.0",
    "pytest-cov>=7.0.0",
    "pytest-asyncio>=1.3.0",
    # "pytest-qt>=4.5.0",          # For Qt apps
    # "pytest-httpserver>=1.1.0",  # For HTTP mocking
]

Test Examples

CLI Test Example

import subprocess

def test_list_profiles_empty() -> None:
    result = subprocess.run(
        ["uv", "run", "poe", "app", "list"],
        capture_output=True, text=True,
    )
    assert result.returncode == 0
    assert "No profiles found" in result.stdout

def test_create_and_list_profile(tmp_path: Path) -> None:
    env = {**os.environ, "APP_DATA_DIR": str(tmp_path)}
    subprocess.run(
        ["uv", "run", "poe", "app", "create", "test-profile"],
        env=env, check=True,
    )
    result = subprocess.run(
        ["uv", "run", "poe", "app", "list"],
        capture_output=True, text=True, env=env,
    )
    assert "test-profile" in result.stdout

Result Pattern Test Example

def test_load_config_missing_file() -> None:
    result = load_config(Path("/nonexistent"))
    assert result.is_err
    assert "not found" in result.unwrap_err()

def test_load_config_valid() -> None:
    result = load_config(Path("tests/fixtures/valid_config.yaml"))
    assert result.is_ok
    config = result.unwrap()
    assert config.name == "test"

Async Test Example

@pytest.mark.asyncio
async def test_fetch_data() -> None:
    result = await fetch_data("https://httpbin.org/get")
    assert result.is_ok

Fixtures

Temporary Directories

@pytest.fixture
def app_data_dir(tmp_path: Path) -> Path:
    data_dir = tmp_path / "data"
    data_dir.mkdir()
    return data_dir

Environment Override

@pytest.fixture
def isolated_env(monkeypatch: pytest.MonkeyPatch, tmp_path: Path) -> tuple[Path, Path]:
    config = tmp_path / "config"
    data = tmp_path / "data"
    config.mkdir()
    data.mkdir()
    monkeypatch.setenv("XDG_CONFIG_HOME", str(config))
    monkeypatch.setenv("XDG_DATA_HOME", str(data))
    return config, data

Sample Data

@pytest.fixture
def sample_audio_16khz() -> np.ndarray:
    return np.zeros(16000, dtype=np.float32)  # 1 second of silence

HTTP Server Mock (Real Server, Not Patched)

@pytest.fixture
def mock_api(httpserver: HTTPServer) -> HTTPServer:
    httpserver.expect_request("/api/data").respond_with_json({"status": "ok"})
    return httpserver

def test_fetch_from_api(mock_api: HTTPServer) -> None:
    result = fetch_data(mock_api.url_for("/api/data"))
    assert result.is_ok

Running Tests

Tests run in parallel by default (-n auto via addopts). Override with -n0 (sequential) or -n4 (exact count).

uv run poe test                    # All tests (parallel, auto workers)
uv run pytest tests/unit/          # Unit only
uv run pytest -n0                  # Force sequential (debugging)
uv run pytest --cov                # With coverage report

Test Isolation

Every test must set up its own state and clean up after itself. Use tmp_path for files, monkeypatch for env vars, yield fixtures for teardown. Never rely on test ordering or shared mutable state. For heavy setup (containers, DB), isolate between groups — scope fixtures to session/module and use non-overlapping namespaces.

Tests that pass alone but fail in parallel are broken tests — fix isolation, don't disable parallelism.

Flaky Tests

A flaky test is worse than a broken one — broken tests block immediately, flaky tests erode trust silently. Never ignore a flaky test. Fix it, rewrite it, or if the root cause is complex — file a bug and report to the user. No other options.

Coverage Guidelines

Not targets to chase, but sanity checks:

If coverage is low but e2e tests cover the workflows, that's fine.

Test Validation

After all tests are written and passing, dispatch a separate sub-agent to validate test quality. The validation agent must check:

• Meaningful coverage — are tests verifying real behavior, or just producing green checkmarks by testing getters/setters/trivial glue?
• Correctness — are assertions actually testing the right thing? No tautologies, no asserting mocks return what they were told to return.
• No source code compromises — was production code incorrectly adjusted just to make tests pass? Logic changes that serve tests rather than users are bugs.
• No shortcuts — no # type: ignore to silence test failures, no overly broad exception catching, no tests that pass regardless of input.

This step is mandatory before submitting work as complete.

Heavyweight Testing

When lightweight testing isn't enough. Same philosophy, higher infrastructure complexity.

Status: Design document. Not yet fully implemented.

When to Use

• Project has external dependencies (APIs, databases, system services)
• Features depend on specific system state (installed binaries, running daemons)
• Integration failures are costly or hard to debug
• Project is long-lived and maintained by multiple people

Investment Decision

Ask before building heavyweight infrastructure:

1. Will this project live long enough to justify the setup time?
2. Are integration failures actually happening or just theoretical?
3. Can lightweight testing (real tmp dirs, pytest-httpserver) cover 80% of the risk?

If yes to all three: build it. If not: stick with lightweight.

Core Idea

Containerized Test Environments

tests/
├── containers/
│   ├── Dockerfile.test-env        # Base test environment
│   ├── Dockerfile.mock-api        # Mock API server
│   ├── docker-compose.test.yml    # Orchestration
│   └── mock-bins/                 # Custom mock binaries
│       ├── mock-telegram          # Fake Telegram Desktop
│       └── mock-ffmpeg            # Fake ffmpeg (returns predefined output)
├── integration/
│   └── test_with_containers.py
└── conftest.py                    # Container lifecycle fixtures

Docker Compose for Test Services

# tests/containers/docker-compose.test.yml
services:
  mock-api:
    build:
      context: .
      dockerfile: Dockerfile.mock-api
    ports:
      - "18080:8080"

  test-db:
    image: postgres:16-alpine
    environment:
      POSTGRES_DB: test
      POSTGRES_PASSWORD: test
    ports:
      - "15432:5432"

Container Lifecycle Fixture

@pytest.fixture(scope="session")
def test_services():
    """Start all test containers, yield, then tear down."""
    compose_file = Path(__file__).parent / "containers" / "docker-compose.test.yml"
    subprocess.run(
        ["podman-compose", "-f", str(compose_file), "up", "-d", "--wait"],
        check=True,
    )
    yield
    subprocess.run(
        ["podman-compose", "-f", str(compose_file), "down", "-v"],
        check=True,
    )

Mock Binaries

Instead of patching subprocess.run(), provide a real binary that behaves predictably:

#!/bin/env python3
# tests/containers/mock-bins/mock-telegram
import sys, time, os

print("Telegram Desktop Mock v1.0")
print(f"Working directory: {os.getcwd()}")

if "-many" in sys.argv and "-workdir" in sys.argv:
    print(f"Mock Telegram started in {sys.argv[sys.argv.index('-workdir') + 1]}")
    time.sleep(int(os.environ.get("MOCK_TELEGRAM_LIFETIME", "5")))
    sys.exit(0)

print("Unknown arguments", sys.argv, file=sys.stderr)
sys.exit(1)

@pytest.fixture
def mock_telegram_bin(tmp_path: Path) -> Path:
    mock_bin = tmp_path / "telegram"
    mock_bin.write_text(MOCK_TELEGRAM_SCRIPT)
    mock_bin.chmod(0o755)
    return mock_bin

async def test_start_instance(mock_telegram_bin: Path) -> None:
    result = await start_instance(profile, binary_path=mock_telegram_bin)
    assert result.is_ok
    pid = result.unwrap()
    assert pid > 0

Stateful Mock API

For APIs that need to maintain state across requests:

from http.server import HTTPServer, BaseHTTPRequestHandler
import threading, json

class MockAPIHandler(BaseHTTPRequestHandler):
    profiles: dict[str, dict[str, str | int | bool]] = {}

    def do_POST(self) -> None:
        if self.path == "/api/profiles":
            data = json.loads(self.rfile.read(int(self.headers["Content-Length"])))
            self.profiles[data["id"]] = data
            self.send_response(201)
            self.end_headers()
            self.wfile.write(json.dumps(data).encode())

    def do_GET(self) -> None:
        if self.path == "/api/profiles":
            self.send_response(200)
            self.end_headers()
            self.wfile.write(json.dumps(list(self.profiles.values())).encode())

@pytest.fixture(scope="session")
def mock_api() -> Generator[str, None, None]:
    server = HTTPServer(("127.0.0.1", 0), MockAPIHandler)
    port = server.server_address[1]
    thread = threading.Thread(target=server.serve_forever, daemon=True)
    thread.start()
    yield f"http://127.0.0.1:{port}"
    server.shutdown()

Real Service Testing (DBus)

@pytest.fixture(scope="session")
def dbus_session() -> Generator[str, None, None]:
    """Start a real DBus session daemon for tests."""
    process = subprocess.Popen(
        ["dbus-daemon", "--session", "--print-address", "--nofork"],
        stdout=subprocess.PIPE,
    )
    address = process.stdout.readline().decode().strip()
    os.environ["DBUS_SESSION_BUS_ADDRESS"] = address
    yield address
    process.terminate()
    process.wait()