indexing inconsistency between h5pyd and h5py (and numpy)

[vsisl]

There is an inconsistency in the shape of the returned array when querying slice of numpy ndarray from an HDF5 file using h5py and h5pyd.

Imagine 3D numpy ndarray my_ary stored in my_file.h5

my_ary = np.zeros((5, 5, 5))
select = (slice(0,1), slice(0,2), slice(0,1))
my_ary[select].shape

(1, 2 ,1)

my_file = h5py.File('my_file.h5', 'w')
my_file.create_dataset(name='my_ary', data=my_ary)
my_file.close()

my_file = h5py.File('my_file.h5', 'r')
my_file['my_ary'][select].shape

(1, 2, 1)

my_file.close()
my_file = h5pyd.File('my_file.hdfgroup.org', 'r', endpoint='xxx:yyy')
my_file['my_ary'][select].shape

(2,)

Is this intended behavior?