python-libraries

name	python-libraries
description	Reach for this project's preferred, well-supported libraries instead of hand-rolling. Use when choosing a library for HTTP, retries, JSON, or parsing, or whenever you are about to write something the standard library or a maintained package already does.

Python Libraries

Prefer well-supported libraries (and the standard library) over hand-rolled code — it is less to maintain and less to get wrong than reinventing a solved problem.

Don't reinvent the wheel

• Don't hand-roll retry or backoff loops — use stamina (it wraps tenacity with sane defaults: capped attempts, jitter, and retrying only the exceptions you name). Set a timeout and exclude non-retryable errors, such as a 4xx that won't change on retry.
• Don't reimplement what itertools (stdlib) or more-itertools already provide (grouping, chaining, windowing, accumulating) — reach for them instead. See python-code-style for the broader "prefer stdlib idioms" rule.
• Don't parse HTML or XML with regex — use parsel (XPath and CSS selectors).

HTTP

• Use an HTTP client with async support — this rules out requests and urllib.
• Match the async client the repo already uses rather than introducing a second one, and don't add another HTTP client as a (dev) dependency without a concrete reason.
• Set an explicit timeout on outbound requests; a single client-level default is fine, but never rely on the library default, which is often None (waits forever).

JSON

• Use orjson over the standard library json for serialising and deserialising — it is faster and better-behaved.
• When a Pydantic model is involved, go straight to and from JSON bytes with Model.model_validate_json(raw) and model.model_dump_json() — one fast parse-and-validate pass.
• Don't call an HTTP response's .json() and hand the resulting dict to Pydantic: that parses twice (the client builds a dict, then Pydantic re-validates it). Pass the raw bytes (response.content) to Model.model_validate_json(...) instead.

See python-async for deploying on uvloop, and python-datetime for dates and times.