Python Best Practices

Python Best Practices

This skill applies modern Python mechanisms without adding annotation or packaging ceremony. It favors inference, precise structural contracts, explicit runtime validation, side-effect-free imports, and context-managed resource ownership.

Effective Strategies for Python

Read the linked guidance that governs the current task before writing or reviewing Python.

1. Make type contracts precise without annotation ceremony.
   • Let the checker infer obvious implementation results; add annotations only when inference cannot express the contract.
   • Depend on the smallest required behavior with protocols, not a vendor's concrete client.
   • In Python 3.13+, express caller-relevant relationships with PEP 695 generics, avoiding module-level TypeVar declarations and needless generics.
   • Make a new finite-state member a type-checking failure with exhaustive handling.
   • Narrow optional values before use so the owning layer—not a cast or fabricated fallback—decides whether absence is preserved or rejected.
2. Validate untrusted data and make invalid states explicit.
   • Parse serialized input once at the controlled boundary so untrusted mappings do not spread inward.
   • Do not turn missing required data into a plausible value; preserve absence or fail at the owning boundary.
   • Keep required runtime checks active under optimization; do not use assertions for validation.
   • Catch only expected exceptions at the operation that raises them through narrow exception handling.
3. Keep packages explicit, inert, and independently distributable.
   • Keep imports inert and consumer-facing names deliberate with explicit public exports.
   • Prevent checkout-dependent imports by following the Python naming and src layout.
   • Publish intentional inline types with py.typed, reserving third-party stubs for dependencies that lack complete inline types instead of adding markers merely to silence diagnostics.
   • Give every independently buildable distribution its own declared metadata and direct dependencies.
   • Make packaging errors visible by not mutating import paths.
4. Keep asynchronous resource and traversal ownership visible.
   • Make async cleanup inseparable from acquisition with async context managers.
   • Preserve page boundaries and consumer control over remote traversal through async pagination.