Essential Libraries Highlights Summary

Every highlight bullet from the 12 pages in this section, gathered on one page and grouped by the page it came from.

The Python Dependency Model

• The standard library is a broad, stable floor - PyPI is where domain depth and speed actually live
• A virtual environment isolates one project's dependency graph from every other project on the machine
• The strongest place to reach for a dependency is a boundary - request bodies, env vars, scraped HTML - not deep inside business logic
• A version pin in pyproject.toml is a promise a maintainer makes, not a guarantee your installer enforces on its own
• Every added dependency is also added attack surface and maintenance liability, not just added capability

httpx & requests

• httpx mirrors requests API and adds async, HTTP/2, and timeouts
• Always set connect/read timeouts - never rely on infinite defaults
• Use context managers or client instances for connection pooling
• raise_for_status() after calls; map HTTP errors to domain exceptions
• Prefer httpx for new code; keep requests only where migration cost is high

Rich & Textual

• Rich formats tables, progress bars, and tracebacks without ANSI boilerplate
• Textual builds full terminal UIs with widgets and async event loops
• Use Console.print() instead of print() for structured CLI output
• Log to stderr; reserve stdout for machine-readable pipe data
• Textual apps need explicit keyboard and resize handling

pydantic-settings

• BaseSettings loads and validates env vars at import/startup
• Field aliases map DATABASE_URL-style names to Python attributes
• Fail fast on missing required secrets before serving traffic
• Use model_config for env file paths and nested delimiter
• Separate Settings per deployable - do not share one god object

tenacity

• Decorator-based retries with exponential backoff and jitter
• Retry only on specific exceptions - not blanket catch-all
• stop_after_attempt caps total tries; stop_after_delay caps wall time
• Idempotent operations only unless you implement deduplication
• Log before_sleep to make retry storms visible in observability

pendulum / arrow

• pendulum and arrow wrap stdlib datetime with human-friendly APIs
• Prefer zoneinfo + datetime for new code when possible
• Always store UTC in databases; convert at display boundaries
• Parsing user input needs explicit timezone or offset
• Avoid naive datetime arithmetic across DST boundaries

Pillow

• Pillow (PIL fork) opens, transforms, and saves raster images
• Always close images or use context managers to free handles
• EXIF orientation must be applied before resize/thumbnail
• Prefer thumbnail() for web previews - preserves aspect ratio
• Validate upload MIME and size before Pillow decode (decompression bombs)

openpyxl / python-docx / reportlab

• openpyxl reads/writes Excel xlsx without Excel installed
• python-docx builds Word documents from templates
• reportlab generates PDFs programmatically for invoices and reports
• Never trust uploaded office files - scan size and macro risk
• For complex layouts, start from a template file rather than zero code

beautifulsoup4 & lxml

• BeautifulSoup provides a forgiving DOM API over html.parser or lxml
• lxml is faster and stricter - good for large documents and XPath
• Respect robots.txt, rate limits, and terms of service
• Prefer official APIs over scraping when available
• Never use scraped HTML with innerHTML-equivalent without sanitization

celery

• Celery workers execute async tasks from a broker (Redis/RabbitMQ)
• Tasks must be idempotent - retries are normal
• JSON serialization only for JSON-serializable payloads
• Separate broker URL, result backend, and beat scheduler config
• FastAPI BackgroundTasks for in-process work; Celery for cross-process scale

SQLModel / SQLAlchemy

• SQLAlchemy is the core ORM/SQL toolkit for Python
• SQLModel combines SQLAlchemy tables with Pydantic models
• Use 2.0 style select() and session.execute - not legacy Query
• Migrations belong in Alembic - not manual CREATE in app code
• See databases section for deep coverage - this page orients essentials

Essential Libraries Best Practices

• Prefer httpx with explicit timeouts over one-shot requests calls
• Validate env config with pydantic-settings at startup
• Retry only idempotent operations with tenacity
• Store UTC timestamps; parse timezones at boundaries
• Cap upload bytes before Pillow decode
• Use Celery for cross-process work, not long HTTP handlers