Check buffer bounds in ndarray creation (Trac #1910)

[numpy-gitbot]

Original ticket http://projects.scipy.org/numpy/ticket/1910 on 2011-07-18 by @pv, assigned to unknown.

Forwarded from Stefan Krah:

---

With a recent NumPy clone + Python3.3, I get this invalid access:

Python 3.3.0a0 (default:1dd6908df8f5, Jul 16 2011, 11:16:00) 
[GCC 4.4.3] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> from numpy import *
>>> x = ndarray(buffer=bytearray([1,2,3,4,5,6,7,8,9,10]), shape=[2,3], strides=[-3, -2], dtype="B", offset=5)
>>> x
array([[  6,   4,   2],
       [  3,   1, 251]], dtype='uint8')
>>> x[1][2]
251
>>> 

I think NumPy should probably refuse to create such an array, like here:

>>> x = ndarray(buffer=bytearray([1,2,3,4,5,6,7,8,9,10]), shape=[2,3], strides=[-3, -2], dtype="B", offset=0)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: strides is incompatible with shape of requested array and size of buffer

[numpy-gitbot]

trac user skrah wrote on 2011-08-04

Conversely, this array is currently rejected but should be allowed (I think):

>>> x = ndarray(buffer=bytearray(1), shape=[0], strides=[1], dtype="B")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: strides is incompatible with shape of requested array and size of buffer

But the rejected array can be created via a slice:

>>> x = ndarray(buffer=bytearray(1), shape=[1], strides=[1], dtype="B")
>>> y = x[0:0]
>>> y.shape == (0,)
True
>>> y.strides == (1,)
True

[seberg]

This is fixed with commit 161bfe7