numpy.ndarray.copy() incorrect when operating on less common array layouts

[QB3]

Description

The combination of transpose + reshape operations gives different results in numba and numpy.

• I am using numba version 0.41.0
• I have included below a minimal working reproducer

import numpy as np
from numba import njit

@njit
def resh_numba(a):
    return a.transpose(1, 0, 2).copy().reshape(2, 12)

x = np.arange(24).reshape(2, 2, 6)

print("numpy")
x_numpy = x.transpose(1, 0, 2).copy().reshape(2, 12)
print(x_numpy)

print("numba:")
x_numba = resh_numba(x)
print(x_numba)

[stuartarchibald]

Thanks for the report. I think the problem is with Numba's np.ndarray.copy implementation:

import numpy as np
from numba import njit

def transp(a):
    return a.transpose(1, 0, 2)

def transp_copy(a):
    return a.transpose(1, 0, 2).copy()

x = np.arange(24).reshape(2, 2, 6)

def test(fn):
    print("Testing %s" % fn.__name__)
    np.testing.assert_allclose(fn(x), njit(fn)(x))
    print("ok\n")

if __name__ == "__main__":
    test(transp)
    test(transp_copy)

gives:

$ python issue3557.py 
Testing transp
ok

Testing transp_copy
Traceback (most recent call last):
  File "issue3557.py", line 19, in <module>
    test(transp_copy)
  File "issue3557.py", line 14, in test
    np.testing.assert_allclose(fn(x), njit(fn)(x))
  File "<path>/lib/python3.7/site-packages/numpy/testing/_private/utils.py", line 1452, in assert_allclose
    verbose=verbose, header=header, equal_nan=equal_nan)
  File "<path>/lib/python3.7/site-packages/numpy/testing/_private/utils.py", line 789, in assert_array_compare
    raise AssertionError(msg)
AssertionError: 
Not equal to tolerance rtol=1e-07, atol=0

(mismatch 91.66666666666667%)
 x: array([ 0,  1,  2,  3,  4,  5, 12, 13, 14, 15, 16, 17,  6,  7,  8,  9, 10,
       11, 18, 19, 20, 21, 22, 23])
 y: array([ 0,  4,  8, 12, 16, 20,  2,  6, 10, 14, 18, 22,  1,  5,  9, 13, 17,
       21,  3,  7, 11, 15, 19, 23])

[QB3]

Thanks a lot for this quick answer.
Shall I rename the issue and change the minimal example then ?

[stuartarchibald]

@QB3 no problem. Title updated and the sample in #3557 (comment) is sufficient thanks. I'm slightly suspicious that once the copy is fixed that the reshape will also need attention.

[stuartarchibald]

On mainline for 0.55, neither the OP reproducer or that in #3557 (comment) reproduces any more. Closing assume fixed.