@jorenham any advice for the NumPy dtype typing failures that have appeared upon upgrading mypy?

@jorenham any advice for the NumPy dtype typing failures that have appeared upon upgrading mypy?

You're comparing a np.dtype instance using is with a type[np.float32], which indeed is always false. You probably want to instead do dtype.type is np.float32

@jorenham any advice for the NumPy dtype typing failures that have appeared upon upgrading mypy?

You're comparing a np.dtype instance using is with a type[np.float32], which indeed is always false. You probably want to instead do dtype.type is np.float32

So the problem is that we're annotating np.float32 with np.dtype[Any]? How are we supposed to type that?

@jorenham any advice for the NumPy dtype typing failures that have appeared upon upgrading mypy?

You're comparing a np.dtype instance using is with a type[np.float32], which indeed is always false. You probably want to instead do dtype.type is np.float32

So the problem is that we're annotating np.float32 with np.dtype[Any]? How are we supposed to type that?

is dtype an instance of np.dtype, e.g. dtype=np.dtype("float32")', or is is a scalar type like dtype=np.float32?

A scalar type.

https://github.com/data-apis/array-api-extra/blob/main/tests/test_funcs.py#L209

A scalar type.

Ah ok, then

should instead be dtype: type[np.generic] or dtype: type[np.floating] or something like that

BTW, hypothesis also has a dedicated strategy for this: https://hypothesis.readthedocs.io/en/latest/reference/strategies.html#hypothesis.extra.numpy.floating_dtypes

It's for np.dtypes though :P

Sounds like we should use that then!

thanks, I think 11d5472 should do it