python-code-quality

Python Code Quality

Typing Rules

• Must annotate all function signatures (parameters and return)
• Must type HTTP responses with Pydantic models—never return raw dict from API endpoints
• Must use generics (list[str], dict[str, Any])—never bare list, dict, tuple
• Must use Pydantic models at API boundaries (requests, responses, external data)
• Should use T | None over Optional[T]
• Should use keyword-only args (*) for functions with 3+ parameters
• Should use TypeAlias for complex dict types instead of repeating dict[str, Any]
• Avoid Any except at true boundaries; prefer specific types or generics
• Never use mutable defaults (def f(items=[])); use None and initialize inside

Type Choice Guide

• Should use NewType for domain primitives (UserId, OrderId) to prevent mixing
• Avoid stringly-typed code—prefer enums or Literal for fixed values

Tooling

• Must use ruff for linting and formatting
• Must use ty (astral.sh) for type checking
• Must run before committing: ruff format . && ruff check --fix . && ty check .
• Must enable ANN, TCH, I, E, F, W ruff rules
• Should use line-length = 120

See ../project-setup/references/pyproject.template.toml for complete configuration.

Exceptions

• Must define a base ServiceError with message, code, status_code
• Must create specific exceptions inheriting from base (NotFoundError, ConflictError)
• Never swallow exceptions silently—always re-raise or wrap with context
• Should use logger.exception() to include traceback before re-raising

Logging

• Must use structured logging (structlog or loguru with JSON sink in production)
• Must include request_id in all log entries
• Never log sensitive data (tokens, passwords, PII)

Caching

• Should use @lru_cache for expensive pure functions
• Must set maxsize based on expected input cardinality
• Never cache functions with side effects or mutable state
• Should use async-lru (@alru_cache) for async functions (not bare @lru_cache)