Closed
Description
Describe the issue:
I'm trying to use typed np.ndarray with mypy.
I find many cases where computations involving np.ndarray eventually become type Any.
I am not certain if the fault lies in numpy or in mypy.
One obvious example is that a.__add__(b) behaves differently from np.add(a, b), as in the following code example.
In the final line, even when starting with a typed array b (numpy.ndarray[Any, numpy.dtype[numpy.floating[numpy.typing._64Bit]]]), the computation np.add(b, b) + np.add(b, b) produces Any.
Reproduce the code example:
import numpy as np
from typing import Any
def function(a: np.ndarray[Any, Any]) -> None:
b = np.ones(10)
reveal_type(a) # numpy.ndarray[Any, Any]
d1 = a + a
d1b = a.__add__(a)
d2 = np.add(a, a)
reveal_type(d1) # Any
reveal_type(d1b) # Any
reveal_type(d2) # numpy.ndarray[Any, numpy.dtype[Any]]
reveal_type(b) # numpy.ndarray[Any, numpy.dtype[numpy.floating[numpy.typing._64Bit]]]
d3 = b + b
d4 = np.add(b, b)
reveal_type(d3) # numpy.ndarray[Any, numpy.dtype[numpy.floating[Any]]]
reveal_type(d4) # numpy.ndarray[Any, numpy.dtype[Any]]
reveal_type(d3) # numpy.ndarray[Any, numpy.dtype[numpy.floating[Any]]]
d5 = d3 + d3
d6 = np.add(d3, d3)
reveal_type(d5) # numpy.ndarray[Any, Any]
reveal_type(d6) # numpy.ndarray[Any, numpy.dtype[Any]]
reveal_type(d4) # numpy.ndarray[Any, numpy.dtype[Any]]
d7 = d4 + d4
d8 = np.add(d4, d4)
reveal_type(d7) # Any
reveal_type(d8) # numpy.ndarray[Any, numpy.dtype[Any]]
reveal_type(np.add(b, b) + np.add(b, b)) # Any !Error message:
No response
NumPy/Python version information:
1.22.4 3.10.4 (main, Jun 29 2022, 12:14:53) [GCC 11.2.0]
mypy==0.971
mypy-extensions==0.4.3
either numpy==1.22.4 or numpy==1.23.1