Numpy.dot o

Problem: numpy dot is throwing an exception, but the docs say it is supported.

Numba version: 0.38.0

Minimal Example:

import numpy as np
from numba import jit

@jit(nopython=True)
def matmul2(size: int):
    a = np.random.random((size, size))
    b = np.random.random((size, size))
    return a.dot(b)

Expected: Should work Actual Result: Exception

Traceback (most recent call last):
  File "test.py", line 209, in <module>
    matmul2(1000)
  File "C:\Program Files\Python36\Lib\site-packages\numba\dispatcher.py", line 344, in _compile_for_args
    reraise(type(e), e, None)
  File "C:\Program Files\Python36\Lib\site-packages\numba\six.py", line 658, in reraise
    raise value.with_traceback(tb)
numba.errors.TypingError: Failed at nopython (nopython frontend)
Unknown attribute 'dot' of type array(float64, 2d, C)

File "test.py", line 204:
def matmul2(size: int):
    <source elided>
    b = np.random.random((size, size))
    return a.dot(b)
    ^

[1] During: typing of get attribute at test.py (204)

File "test.py", line 204:
def matmul2(size: int):
    <source elided>
    b = np.random.random((size, size))
    return a.dot(b)
    ^

[sirwhinesalot]

No, that did appear, I excluded it from my bug report up there to make it shorter.

The problem is that while the message is helpful, it just says “it is caused by an unsupported feature” which is correct, but the unsupported feature happens to be a trivial method call that should map to np.dot(), which is fully supported and hence the situation is confusing.

I’m sure people intimately familiar with numpy will understand that numpy.dot and numpy.ndarray.dot are different things but I didn’t and my colleagues were quite surprised as well. If your quick fix is not accepted, a little snipped saying “the method call form of numpy operations is not supported” would be quite helpful in the documentation.

[stuartarchibald]

No problem. The merge of https://github.com/numba/numba/pull/2946 will automatically close it.