"assert_almost_equal" does not fail when comparing a number and an empty list

The behavior of numpy.testing.assert_almost_equal seems incorrect when one of the arguments is an empty list.

The following cases behave as I expect:

# Compare an empty list to a list containing the number 2. These are not almost equal, so an AssertionError is correctly raised.

>>> from numpy.testing import assert_almost_equal
>>> assert_almost_equal([], [2])
AssertionError: 
Arrays are not almost equal to 7 decimals

(shapes (0,), (1,) mismatch)
 x: array([], dtype=float64)
 y: array([2])

# Compare the integer 2 to a list containing only 2. Since the arguments are treated as 'array-like', this should not raise an exception (and doesn't).

>>> from numpy.testing import assert_almost_equal
>>> assert_almost_equal(2, [2])
>>>

When one of the arguments is an empty list and the other is a number, they are presumably not almost equal. Nonetheless:

>>> from numpy.testing import assert_almost_equal
>>> assert_almost_equal([], 2)  # no AssertionError?
>>>

The same behavior can be observed in numpy.testing.assert_allclose.

[mhvk]

I think the behaviour is quite obviously not intended and would argue for ensuring that comparing arrays with size 0 and size non-zero always leads to failure. Since this problem arises because we’re already special-casing scalars (which is useful), one might as well do something like:

cond = (x.shape == () and y.size > 0 or y.shape == () and x.size > 0) or x.shape == y.shape

[ZedThree]

I’ve just bumped into this today, or at least I think this is the same issue. Here are some more examples of surprising behaviour with empty lists/zero-sized arrays:

>>> np.all([] == [12])
False

>>> np.all(np.array([]) == [12])
True

>>> np.array_equal(np.array([]), [12])
False

>>> np.allclose(np.array([]), [12])
True

I think I understand why these happen, but they don’t necessary feel correct. The documentation seems vague on this too.