How we chose Positron's Python type checker

12 hours ago

Positron evaluated four open-source Python type checkers (Pyrefly, ty, Basedpyright, Zuban) to select a default for enhancing Python data science.
Key evaluation dimensions included feature completeness, correctness, performance, ecosystem health, and development velocity.
Pyrefly, backed by Meta, uses aggressive type inference, catching bugs even in untyped code, and offers clean hover documentation with hyperlinks.
ty, developed by Astral (acquired by OpenAI), follows a 'gradual guarantee' philosophy, only flagging errors in annotated code, and is the fastest in performance.
Basedpyright, a community fork of Pyright, is mature and feature-complete but defaults to stricter type checking, which may overwhelm data scientists.
Zuban, from the author of Jedi, aims for mypy compatibility but is less mature, with limited third-party library completions and a two-step installation.
Positron chose Pyrefly as the default due to its balance of features, active development, MIT license, and suitability for data science workflows.
Users can easily switch language servers in Positron by installing alternatives and disabling Pyrefly, ensuring flexibility as tools evolve.
The type checker ecosystem is rapidly evolving, with ongoing improvements from all contenders, and Positron plans to continue evaluations.

Hasty Briefsbeta