python-guide（Python 指南）

Python Guide

Applies to: Python 3.11+, APIs, CLIs, Data Pipelines, Automation

Core Principles

1. Type Hints Everywhere: All function signatures, class attributes, and module-level variables must have type annotations
2. Explicit Over Implicit: No * imports, no mutable default arguments, no implicit type coercions
3. Virtual Environments Always: Never install into system Python; use venv, uv, or poetry
4. Pytest Over unittest: Use pytest for all testing; fixtures and parametrize over setUp/tearDown
5. PEP 8 + Ruff: Enforce style mechanically; never rely on manual formatting

Guardrails

Python Version

• Target Python 3.11+ (use match statements, ExceptionGroup, tomllib)
• Set requires-python = ">=3.11" in pyproject.toml
• Use from __future__ import annotations for forward references in 3.11
• Never use features removed in 3.12+ (distutils, imp, legacy typing aliases)

Code Style

• Run ruff check and ruff format before every commit
• Max line length: 88 characters (Black default)
• Imports: stdlib, blank line, third-party, blank line, local (enforced by isort/ruff)
• Naming: snake_case for functions/variables, PascalCase for classes, UPPER_SNAKE for constants
• No bare except: — always catch specific exceptions
• No mutable default arguments (def f(items=None): not def f(items=[]):)
• Prefer f-strings over .format() or % formatting
• Use pathlib.Path instead of os.path for all file operations

Type Hints

• All public functions MUST have full type annotations (params + return)
• Use collections.abc types: Sequence, Mapping, Iterable (not List, Dict)
• Use X | None union syntax (not Optional[X])
• Use TypeAlias for complex types: UserMap: TypeAlias = dict[str, User]
• Use Protocol for structural subtyping (duck typing with safety)
• Use @overload for functions returning different types based on input
• Run mypy --strict in CI (no type: ignore without explanation)

from collections.abc import Sequence

def find_users(
    ids: Sequence[str],
    *,
    active_only: bool = True,
) -> list[User]:
    """Fetch users by ID list, optionally filtering inactive."""
    ...

Error Handling

• Never use bare except: or except Exception: without re-raising
• Create domain-specific exception hierarchies rooted in a base class
• Use raise... from err to preserve exception chains
• Log at the boundary, raise in the interior (don't log-and-raise)
• Use contextlib.suppress() instead of empty except blocks
• Always close resources with with statements or contextlib.closing

Dependencies

• Define all deps in pyproject.toml (not setup.py or bare requirements.txt)
• Pin exact versions in lock files (uv.lock, poetry.lock, pip-compile output)
• Keep requirements.txt only as a generated artifact, never hand-edited
• Separate [project.optional-dependencies] for dev, test, docs
• Audit with pip-audit or safety before adding new packages
• Prefer stdlib solutions: tomllib, pathlib, dataclasses, enum, logging

Project Structure

myproject/
├── src/
│   └── myproject/          # Importable package (src layout)
│       ├── __init__.py
│       ├── py.typed         # PEP 561 marker for type stubs
│       ├── domain/          # Business logic, entities
│       │   ├── __init__.py
│       │   ├── models.py
│       │   └── exceptions.py
│       ├── service/         # Application services
│       │   └── __init__.py
│       ├── repository/      # Data access layer
│       │   └── __init__.py
│       └── api/             # HTTP/CLI interface
│           └── __init__.py
├── tests/
│   ├── conftest.py          # Shared fixtures
│   ├── unit/
│   └── integration/
├── pyproject.toml           # Single source of truth for config
├── uv.lock                  # Or poetry.lock
└── README.md

• Use src layout (src/myproject/) to prevent accidental local imports
• Keep conftest.py at test root for shared fixtures; nest for scope
• Include py.typed marker for downstream type checking
• No __init__.py in tests/ (pytest discovers without it)
• One module = one responsibility; split at ~200 lines

Error Handling Patterns

Exception Hierarchy

class AppError(Exception):
    """Base exception for the application."""

    def __init__(self, message: str, *, code: str = "UNKNOWN") -> None:
        super().__init__(message)
        self.code = code

class NotFoundError(AppError):
    """Raised when a requested resource does not exist."""

    def __init__(self, resource: str, identifier: str) -> None:
        super().__init__(
            f"{resource} with id '{identifier}' not found",
            code="NOT_FOUND",
        )
        self.resource = resource
        self.identifier = identifier

class ValidationError(AppError):
    """Raised when input data fails validation."""

    def __init__(self, field: str, reason: str) -> None:
        super().__init__(
            f"Validation failed for '{field}': {reason}",
            code="VALIDATION_ERROR",
        )

Context Managers for Cleanup

from contextlib import contextmanager
from collections.abc import Generator

@contextmanager
def managed_connection(url: str) -> Generator[Connection, None, None]:
    conn = Connection(url)
    try:
        conn.open()
        yield conn
    except ConnectionError as err:
        raise AppError("Database unavailable") from err
    finally:
        conn.close()

Error Chaining

def get_user(user_id: str) -> User:
    try:
        row = db.fetch_one("SELECT * FROM users WHERE id = %s", (user_id,))
    except DatabaseError as err:
        raise AppError(f"Failed to fetch user {user_id}") from err
    if row is None:
        raise NotFoundError("User", user_id)
    return User.from_row(row)

Testing

Standards

• Test files: test_*.py (same name as module: models.py -> test_models.py)
• Test functions: test_<unit>_<scenario>_<expected> (e.g., test_get_user_not_found_raises)
• Use conftest.py for fixtures shared across a directory
• Coverage target: >80% for business logic, >60% overall
• Mark slow tests: @pytest.mark.slow and exclude from default runs
• No unittest.TestCase — use plain functions with pytest assertions
• Use tmp_path fixture for file operations (auto-cleanup)

Fixtures and Parametrize

import pytest
from myproject.domain.models import User

@pytest.fixture
def sample_user() -> User:
    return User(id="u-123", name="Ada Lovelace", email="ada@example.com")

@pytest.mark.parametrize(
    ("email", "is_valid"),
    [
        ("user@example.com", True),
        ("user@.com", False),
        ("", False),
        ("user@domain", False),
    ],
)
def test_validate_email(email: str, is_valid: bool) -> None:
    assert validate_email(email) == is_valid

def test_get_user_returns_user(sample_user: User) -> None:
    repo = InMemoryUserRepo(users=[sample_user])
    result = repo.get("u-123")
    assert result == sample_user

def test_get_user_not_found_raises() -> None:
    repo = InMemoryUserRepo(users=[])
    with pytest.raises(NotFoundError, match="User.*not found"):
        repo.get("nonexistent")

Mocking External Dependencies

from unittest.mock import AsyncMock, patch

async def test_send_notification_retries_on_failure() -> None:
    mock_client = AsyncMock()
    mock_client.post.side_effect = [ConnectionError, None]

    with patch("myproject.service.notify.http_client", mock_client):
        await send_notification(user_id="u-123", message="hello")

    assert mock_client.post.call_count == 2

Tooling

pyproject.toml Configuration

[project]
name = "myproject"
requires-python = ">=3.11"

[project.optional-dependencies]
dev = ["ruff", "mypy", "pytest", "pytest-cov", "pytest-asyncio"]

[tool.ruff]
target-version = "py311"
line-length = 88

[tool.ruff.lint]
select = [
    "E",    # pycodestyle errors
    "W",    # pycodestyle warnings
    "F",    # pyflakes
    "I",    # isort
    "N",    # pep8-naming
    "UP",   # pyupgrade
    "B",    # flake8-bugbear
    "S",    # flake8-bandit (security)
    "A",    # flake8-builtins
    "C4",   # flake8-comprehensions
    "SIM",  # flake8-simplify
    "RUF",  # ruff-specific rules
]

[tool.mypy]
strict = true
warn_return_any = true
disallow_untyped_defs = true

[tool.pytest.ini_options]
testpaths = ["tests"]
markers = ["slow: marks tests as slow (deselect with '-m \"not slow\"')"]
asyncio_mode = "auto"

[tool.coverage.run]
source = ["src/myproject"]
branch = true

[tool.coverage.report]
fail_under = 60
show_missing = true
exclude_lines = ["if TYPE_CHECKING:", "pragma: no cover"]

Essential Commands

ruff check .                 # Lint (replaces flake8, isort, pyupgrade)
ruff format .                # Format (replaces black)
mypy .                       # Type check (strict mode)
pytest                       # Run all tests
pytest --cov=src -q          # Coverage summary
pytest -m "not slow"         # Skip slow tests
pip-audit                    # Check dependencies for vulnerabilities
python -m build              # Build sdist + wheel

Advanced Topics

For detailed patterns and examples, see:

• references/patterns.md -- Async patterns, dataclass/Pydantic models, context managers, decorators, type hints (generics, Protocol, TypeVar)
• references/pitfalls.md -- Common Python gotchas and do/don't examples
• references/security.md -- Input sanitization, secrets management, SQL injection prevention

External References

• PEP 8 -- Style Guide
• PEP 484 -- Type Hints
• Ruff Documentation
• mypy Documentation
• pytest Documentation
• Python Packaging Guide
• Effective Python (Brett Slatkin)