Numba stubs

type stubs for numba.

https://numba.readthedocs.io/en/stable/index.html

https://github.com/numba/numba

What does this do

Correct all errors from calls, provide autocompletions for type signature, preserve decorated functions and class signatures.

Provide types for:

• jitclass
• jit
• guvectorize signature
• numba types (floats, integers, arrays) as well as Array inference from slicing calls

Notes

• It's currently impossible to correctly type both the guvectorize functions declarations, decorator signature, AND calls to the decorated func.

• numba types are generics in those stubs. This is not the case in the actual implementation, don't try to subtype them.

• Due to actual implementation design choices, overloads are mandatory for jit. This mean that the docstring from the actual implementation is lost (at least with Pylance).

• Signatures specs in jit are expected to be given with numba types, not strings.

Installation

uv add git+https://github.com/py-stubs/numba-stubs.git

Testing

At the root of the repository, run:

uv run tests/main.py