Skip to content

Roadmap

Errium's core (errium_core) is framework-agnostic by design; the work ahead is mostly about adding thin adapters on top of it (see ARCHITECTURE.md for how an adapter is structured) and rounding out the classification coverage in the core itself.

Done

  • Core classification pipeline — ClassificationEngine, the ExceptionClassifier protocol, StandardizedError/ClassifiedError contracts, DefaultFormatter, ValidationNormalizer, dev/prod sanitization via ERRIUM_DEBUG.
  • FastAPI adapter (errium) — ErriumMiddleware for uncaught exceptions and HTTP exceptions, validation_exception_handler with beautified validation details.
  • Flask adapter (errium_flask) — ErriumFlask extension covering Werkzeug HTTP exceptions, generic exceptions, and beautified pydantic validation errors raised manually inside views.
  • Django Ninja adapter (errium_ninja) — register_errium(api) covering Ninja's ValidationError/HttpError (and its AuthenticationError/AuthorizationError/Throttled subclasses), Django's Http404, and generic exceptions, overriding Ninja's built-in defaults (which otherwise re-raise uncaught exceptions in production instead of returning JSON). Also strips Ninja's synthetic parameter-name wrapper from validation error locations so details keys match the other adapters.
  • Django REST Framework adapter (errium_drf) — errium_exception_handler, wired via REST_FRAMEWORK["EXCEPTION_HANDLER"], covering ValidationError (keeping DRF's own 400 status code), the APIException family, Django's Http404, and PermissionDenied. Achieves full catch-everything coverage — including raw uncaught exceptions — since it never returns None the way DRF's own default handler does; this corrects an earlier assumption (below, when this was still "planned") that a separate Django middleware would be needed for that. Has its own flatten_drf_errors normalizer, since DRF's ErrorDetail tree shape differs from pydantic's loc/type/msg. Also mirrors DRF's WWW-Authenticate/Retry-After header behavior for auth challenges and throttling.
  • Database error classification — DatabaseExceptionClassifier maps SQLAlchemy errors to DATABASE_ERROR, and integrity errors that look like uniqueness violations to DUPLICATE_RESOURCE, without a hard SQLAlchemy dependency.
  • Published on PyPI — errium 0.1.0 is available at https://pypi.org/project/errium/ (pip install errium), with flask, ninja, and drf extras. All five packages ship py.typed. The distribution depends on fastapi rather than fastapi[standard], so Flask and Django users don't inherit a server and CLI toolchain.
  • CI — GitHub Actions workflow running ruff check, ruff format --check, mypy, and pytest across Python 3.11/3.12 on every push and PR to main.

Planned

  • Express.js adapter (JavaScript port) — a separate package outside the Python distribution; would need its own port of the core contracts, not just an adapter.
  • AI-powered developer suggestions & self-healing hints — extending DefaultFormatter._get_debug_hints (currently a static, status-code-keyed lookup) with more context-aware, possibly model-generated suggestions in debug mode.

Notes for contributors

When picking up a new framework adapter, follow the pattern documented in ARCHITECTURE.md ("Design goal: core vs. adapters"): a new src/errium_<framework> package containing framework-specific ExceptionClassifiers plus one integration point that builds a ClassificationEngine, classifies, formats, and returns the framework's native response type. Nothing in errium_core should need to change to support a new framework.