data leakage with S0 in record dtype (Trac #1992) #2585

Original ticket http://projects.scipy.org/numpy/ticket/1992 on 2011-12-12 by trac user hpaulj, assigned to unknown.

As of numpy 1.5.1 (and earlier)

>>> x = np.zeros(1, 'S,i')
>>> x['f0']='A'
>>> x
array([('', 65)], dtype=[('f0','|S0'),('f1','<i4')])

In contrast, if the assigned dtype is
'S1,i'
, x is then ('A',0)

So not only is there ambiguity as to what a 'S' dtype produces (as discussed in other tickets), but assignment to that record leaks over to the adjacent field.

I first encountered this in an introduction to numpy for Ironpython, but it occurs in all versions that I've tested.