Ian Beaver added the comment: Its not multi-dimensional slicing to get a subset of objects as in Numpy, but more the ability to slice a buffer containing a multi-dimensional array as raw bytes. Buffer objects in Python2.7 are dimensionality naive so it works fine. You were correct that I was testing against Python3.2, in Python3.3 the slicing of ndim > 1 works, however only for reading from the buffer. I still can't write back into a memoryview object with ndim > 1 in Python 3.3.
Python 2.7.3: >>> import numpy as np >>> arr = np.zeros(shape=(100,100)) >>> type(arr.data) <type 'buffer'> >>> arr.data[0:10] '\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00' >>> Python 3.2.3: >>> import numpy as np >>> arr = np.zeros(shape=(100,100)) >>> type(arr.data) <class 'memoryview'> >>> arr.data[0:10] Traceback (most recent call last): File "<stdin>", line 1, in <module> NotImplementedError >>> Python 3.3.3: >>> import numpy as np >>> arr = np.zeros(shape=(100,100)) >>> type(arr.data) <class 'memoryview'> >>> arr.data[0:10] <memory at 0x7faaf1d03a48> >>> However to write data back into a buffer: Python 2.7.3: >>> import numpy as np >>> arr = np.zeros(shape=(100,100)) >>> arr.data[0:10] = '\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00' >>> Python 3.2.3: >>> import numpy as np >>> arr = np.zeros(shape=(100,100)) >>> arr.data[0:10] = b'\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00' Traceback (most recent call last): File "<stdin>", line 1, in <module> NotImplementedError >>> Python 3.3.3: >>> import numpy as np >>> arr = np.zeros(shape=(100,100)) >>> arr.data[0:10] = b'\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00' Traceback (most recent call last): File "<stdin>", line 1, in <module> NotImplementedError: memoryview assignments are currently restricted to ndim = 1 >>> Also the slice in Python3.3 is not the same as just returning a chunk of raw bytes from the memory buffer, instead of a bytes object the indexing behaves similar to numpy array indexes and you get the (sub) array items back as Python objects. Python2.7.3: >>> import numpy as np >>> arr = np.zeros(shape=(100,100)) >>> arr.data[0:10] '\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00' len(bytes(arr.data[0:10])) 10 Python3.3.3: >>> import numpy as np >>> arr = np.zeros(shape=(100,100)) >>> arr.data[0:10] <memory at 0x7f109a71ea48> >>> len(bytes(arr.data[0:10])) 8000 This is not a big deal in my case since I already have numpy arrays I can just use bytes(arr.flat[start:end]) to scan through the array contents as byte chunks, but that would not be possible with just a memoryview object like it was with the Python2 buffer object without converting it to something else or dropping to ctypes and iterating over the memory addresses and dereferencing the contents. So in Python3.3 its halfway to the functionality in Python2.7, I can send chunks of the data through a compressed or encrypted stream, but I can't rebuild the data on the other side without first creating a bytearray and eating the cost of a copy into a memoryview. All I really need is a way to reconstruct the original memoryview buffer in memory from a stream of bytes without having to make a temporary object first and then copy its contents into the final memoryview object when it is complete. ---------- _______________________________________ Python tracker <rep...@bugs.python.org> <http://bugs.python.org/issue14130> _______________________________________ _______________________________________________ Python-bugs-list mailing list Unsubscribe: https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
