ENH: helper for multidimensional slicing

In thinking about #10771, I realized that a simple helper function like the following might be handy:

def ndslice(sl):
    """ convert a slice of iterables into a tuple of slices (nd slice)"""
    start = sl.start
    stop = sl.stop
    step = sl.step
    if start is None and stop is None and step is None:
        return ()
    if start is None:
        start = itertools.repeat(None)
    if stop is None:
        stop = itertools.repeat(None)
    if step is None:
        step = itertools.repeat(None)
    return tuple(
        slice(art, op, ep)
        for art, op, ep in zip(start, stop, step)
    )

Which could be used as:

a = np.zeros((10, 10, 10))
a[ndslice(np.s_[[1,2,3]:[4,5,6]])]  # same as a[1:4, 2:5, 3:6]

Or possibly using a __getitem__ method like np.s_[], sugared to:

class ndslice_type(object):
    __getitem__ = lambda self, item: ndslice(item)
ndslice_ = ndslice_type()

a[ndslice_[[1,2,3] : [4,5,6]]]  # same as a[1:4, 2:5, 3:6]

[shoyer]

You can’t combine ... with a tuple. So that would need to be:

sliding_window_view(arr, shape)[(Ellipsis,) + np.ndslice_[::step_tup]]
sliding_window_view(arr, shape)[np.ndslice_[::step_tup] + (Ellipsis,)]

[pv]

np.ndslice_[...,::step_tup]? Not sure what the semantics for ... are in ndslice though?