`squeeze` accepts `axis=None`?

[mdhaber]

It looks like axis=None is not in the standard for squeeze, yet it is accepted by array_api_strict.

import array_api_strict as xp
x = xp.reshape(xp.arange(12), (3, 4))[..., xp.newaxis]
xp.squeeze(x, axis=None)
# Array([[ 0,  1,  2,  3],
#        [ 4,  5,  6,  7],
#        [ 8,  9, 10, 11]], dtype=array_api_strict.int64)

[ev-br]

I suspect it's an omission in the standard. ISTM all array libraries allow axis=None with the meaning of "remove all singleton dimensions":

In [17]: t = torch.empty((1, 2, 1, 3, 4, 1))

In [18]: torch.squeeze(t).shape
Out[18]: torch.Size([2, 3, 4])

torch does not allow squeeze(t, dim=None) which is fairly typical for torch's implementations of optional arguments, so will need to be worked around in the array-api-compat layer.

cross-ref data-apis/array-api#892

[ev-br]

Well, no, apparently, it's not an omission in the standard, it's deliberate, cf data-apis/array-api#892 (comment)

Let's then block axis=None in -strict : #125