numpy.array inplace exchange duplicates/overwrites elements (Trac #1057)

[thouis]

Original ticket http://projects.scipy.org/numpy/ticket/1057 on 2009-03-18 by trac user praeteri, assigned to unknown.

simply check this:

In [1]: from numpy import array

In [2]: a = array([ array([1,1]), array([2,2]), array([3,3]), array([4,4]), array([5,5]) ]) 

In [3]: a[0], a[1] = a[1], a[0]

In [4]: a
Out[4]: 
array([[2, 2],
       [2, 2],
       [3, 3],
       [4, 4],
       [5, 5]])

This data format is used e.g. by pylab.load.
The inplace exchange e.g. by random.shuffle.
Hence, it is a very common situation...

I don't know if this is the way it should be.

[thouis]

@charris wrote on 2009-03-18

It is the difference between objects and views. The views a[0], a[1] reference rows in a and the assignment a[0] = a[1] copies the data from row 1 to row 0. If the rows were themselves objects then things would work in the normal python way, but they aren't. So I think the current behaviour is correct and if the other behaviour is assumed in some applications then it is a bug.

[thouis]

@pv wrote on 2009-07-13

I don't think we can do much about this, except to document it. The items on the RHS never know they are being assigned to the same array.