Hello. The following seems a little weird:

Python 3.3.2 (v3.3.2:d047928ae3f6, May 16 2013, 00:03:43) [MSC v.1600 32 bit (Intel)] on win32
>>> m = memoryview(b'123')
>>> list(m[::-1])
[51, 50, 49]
>>> list(reversed(m))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: object of type 'memoryview' has no len()
>>>

The attached patch allows `reversed` to be called with memoryviews and it could potentially fix issue18690.

So the dilemma with len() was: does it return the number of bytes, or the number of items?
Given the new memoryview semantics, I'd say it should return the number of items.

Aye, with the 3.3 changes, I think we should continue down the
multi-dimensional array path. I suggest we run with whatever NumPy uses for
ndarray, which I believe is "number of items in the first dimension".

So, in the following case, len shouldn't return 4, but the number of items? Also, is it ok to assume that the number of items is
((*shape)[0] * ... * (*shape)[ndims-1])?

>>> x = np.array([[1, 2, 3], [4, 5, 6], [4,5,6], [4,4,4]], np.int32)
>>> x.shape
(4, 3)
>>> x.shape[0] * x.shape[1]
12
>>> len(x)
4
>>> memoryview(x)
<memory at 0x0217C378>
>>> len(memoryview(x))
4
>>> memoryview(x).shape
(4, 3)
>>>

No, you're right, it should probably return 4.

Yes, len() should return the number of items. +1 for making reversed()
work.

NumPy does this:

>>> x = numpy.array([1,2,3,4,5,6,7,8])
>>> list(reversed(x))
[8, 7, 6, 5, 4, 3, 2, 1]
>>>
>>> x = numpy.array([[1,2,3], [4,5,6]])
>>> list(reversed(x))
[array([4, 5, 6]), array([1, 2, 3])]
>>>

Hmm, I meant: Number of items in the first dimension, thus 4 in
Claudiu's example.

For multidimensional arrays it doesn't seem to work (yet).

>>> x = numpy.array([[1,2,3], [4,5,6]])
>>> list(reversed(x))
[array([4, 5, 6]), array([1, 2, 3])]
>>> x.data
<memory at 0x8032d06b8>
>>> list(reversed(x.data))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
NotImplementedError: multi-dimensional sub-views are not implemented
>>>

Stefan, what do you think about Claudiu's patch? Should a test be added to test_buffer as well?

Claudiu's patch looks correct.

> Stefan, what do you think about Claudiu's patch? Should a test be added to test_buffer as well?

I think the patch is good. We can add more tests when (if?) multi-dimensional
support is added to memoryview.

In that case we should probably do the same as NumPy and return a list of
subviews. So testing against tolist() like in the test case will only work
for one-dimensional views.

I can't commit right now (the machine with my ssh-key won't have Internet
access for some time), so if someone has time to do it ...

New changeset 0dc604d58949 by Nick Coghlan in branch 'default':
Close #19078: memoryview now supports reversed
http://hg.python.org/cpython/rev/0dc604d58949