[Upstream Bug] Please notify NumPy that numpy.typing.ArrayLike is broken

[felixchenier]

Hi folks

When I run this code:

from beartype import beartype
import numpy as np
from numpy.typing import ArrayLike

@beartype
def test(a: ArrayLike, b: float) -> ArrayLike:
    return a + b
    
print(test(np.eye(2), 4.0))

I get this:

Traceback (most recent call last):
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_util/cls/pep/utilpep3119.py", line 124, in die_unless_type_isinstanceable
    isinstance(None, cls)  # type: ignore[arg-type]
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/typing.py", line 1146, in __instancecheck__
    issubclass(instance.__class__, cls)):
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/abc.py", line 123, in __subclasscheck__
    return _abc_subclasscheck(cls, subclass)
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/typing.py", line 1208, in _proto_hook
    raise TypeError("Instance and class checks can only be used with"
TypeError: Instance and class checks can only be used with @runtime_checkable protocols

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/Users/felix/test.py", line 6, in <module>
    def test(a: ArrayLike, b: float) -> ArrayLike:
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_decor/_cache/cachedecor.py", line 77, in beartype
    return beartype_object(obj, conf)
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_decor/decorcore.py", line 239, in beartype_object
    return _beartype_func(  # type: ignore[return-value]
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_decor/decorcore.py", line 579, in _beartype_func
    func_wrapper_code = generate_code(bear_call)
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_decor/_wrapper/wrappermain.py", line 215, in generate_code
    code_check_params = _code_check_args(bear_call)
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_decor/_wrapper/wrappermain.py", line 475, in _code_check_args
    reraise_exception_placeholder(
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_util/error/utilerror.py", line 212, in reraise_exception_placeholder
    raise exception.with_traceback(exception.__traceback__)
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_decor/_wrapper/wrappermain.py", line 450, in _code_check_args
    ) = make_func_wrapper_code(hint)
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_util/cache/utilcachecall.py", line 339, in _callable_cached
    raise exception
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_util/cache/utilcachecall.py", line 331, in _callable_cached
    return_value = params_flat_to_return_value[params_flat] = func(
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_decor/_wrapper/_wrappercode.py", line 71, in make_func_wrapper_code
    ) = make_check_expr(hint)
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_util/cache/utilcachecall.py", line 339, in _callable_cached
    raise exception
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_util/cache/utilcachecall.py", line 331, in _callable_cached
    return_value = params_flat_to_return_value[params_flat] = func(
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_check/expr/exprmake.py", line 1712, in make_check_expr
    hint_curr_expr=add_func_scope_type(
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_check/expr/_exprscope.py", line 196, in add_func_scope_type
    die_unless_type_isinstanceable(cls=cls, exception_prefix=exception_prefix)
  File "/Users/felix/opt/miniconda3/envs/test/lib/python3.9/site-packages/beartype/_util/cls/pep/utilpep3119.py", line 165, in die_unless_type_isinstanceable
    raise exception_cls(exception_message) from exception
beartype.roar.BeartypeDecorHintPep3119Exception: @beartyped __main__.test() parameter "a" type hint <class 'numpy._typing._nested_sequence._NestedSequence'> uncheckable at runtime (i.e., not passable as second parameter to isinstance(), due to raising "Instance and class checks can only be used with @runtime_checkable protocols" from metaclass __instancecheck__() method).

I'm wondering if it's normal, and if it is, if there are plans to support ArrayLike.

Thanks!

[leycec]

Gah! numpy.typing strikes back. Sadly, this is an upstream NumPy issue. NumPy internally needs to explicitly decorate all of its PEP 544-compliant protocols (like numpy.typing._array_like._SupportsArray) by @typing.runtime_checkable. See also issue #130, which this is basically a depressing duplicate of.

This is currently afflicting every runtime type-checker – including typeguard. I suppose we'd all just hoped that someone else would notice this and do something about it. And here we are, slow-walking on the road to perdition.

@braniii: Did you ever get a chance to gently present this bug to NumPy devs? If not, no worries whatsoever. In that case, would you mind doing so instead, @felixchenier? I know, I know. This isn't what anyone wants to do on Wednesday. 😞

[leycec]

EDIT: Actually, don't sweat this one, @braniii and @felixchenier. I submitted an upstream NumPy issue tonight calling for clarification. This is a pretty deeply entangling subject. It was probably best that I charge the guns on that one. 🏇

[leycec]

I have good news. I have bad news. The bad news is that NumPy devs don't appear to care that their type hinting API is mostly unusable at runtime. The good news is that... okay, I've got nuthin'. There's no good news, actually.

@braniii and @felixchenier: would either of you mind chiming in at numpy/numpy#22352? That might help to lend some credibility to this issue and bring a NumPy dev or two around. They're really making it hard for us to love NumPy type hints, aren't they? 😭

[felixchenier]

I did put my emoji support. Good news is that there is now a PR for it 😁

[rtbs-dev]

just now ran into this same problem here. good to see the PR!

[leycec]

Thanks a basket of blueberries for all the cheerleading! When NumPy patches in runtime support for this, we'll immediately list ArrayLike as officially supported in our Features matrix without having to do any actual work.

Which is great, because you know how I feel about actual work. 😮‍💨

[felixchenier]

Hey @leycec
Long time no wrote :-)

By the time this was fixed, I had written a lazy-poor type checker that fulfilled my needs "a bit", but now I'm moving from 3.8- annotations to 3.10 and it's time to revisit my choices. I got back to where we left, upgraded everything, and now the code that crashed on the OP doesn't crash anymore. I guess numpy has been released, or you patched something, but in any case this question is not relevant anymore since it now works as expected.

Therefore I'll revisit the possibility of using beartype in my project :-)