isinstance(None, numba.types.NoneType) is False

[david-zwicker]

While playing with generated_jit I stumbled upon this unexpected behavior:

>>> isinstance(None, numba.types.NoneType)

False

Clearly, standard python disagrees with this:

>>> isinstance(None, type(None))

True

Is there a rational for this behavior or is this simply a bug?

[stuartarchibald]

Thanks for raising this. I think this is because Numba's type system mimics that of Python/NumPy but does not inherit from it (kind of indicated by MRO)

In [8]: numba.types.NoneType.__mro__                                                                                            
Out[8]: 
(numba.types.misc.NoneType,
 numba.types.common.Opaque,
 numba.types.abstract.Dummy,
 numba.types.abstract.Type,
 object)

Equally isinstance(int, numba.types.Integer) is False.

Relating this to @generated_jit, type inference will first try and compile the function with anything that can be represented as a literal value represented as such (integers, strings, slices and make_function [note: None is not in this set of types]), if this fails it will try to compile the function with everything represented by the Numba type system types. The 0.41.0 Release demo notebook (Numba_041_release_demo.ipynb) at https://mybinder.org/v2/gh/numba/numba-examples/master?filepath=notebooks has a demonstration under the Literal value support section.

Another user asked a similar/related question yesterday, an example is also available there #3585 (comment)

Hope this helps?

[david-zwicker]

I don't quite get the analogy between isinstance(int, numba.types.Integer) and the original problem with None. I though int is a type itself and would therefore not be expected to be an instance of Integer. However, I interpret None as a (single) instance of NoneType and would thus expect the isinstance check to return True.

[stuartarchibald]

Apologies, that's an error, should read something like "isinstance(1, numba.types.Integer) also returns False", same reason as above applies.

[stuartarchibald]

@david-zwicker you may be interested in the For Developers section of this notebook relating to the 0.43 release https://github.com/numba/numba-examples/blob/master/notebooks/Numba_043_release_demo.ipynb

[stuartarchibald]

Closing this question as it seems to be resolved. Numba now has a discourse forum https://numba.discourse.group/ which is great for questions like this, please do consider posting there in future :) Thanks!