recarray.__getitem__ with field list gives "TypeError: Cannot change data-type for object array." when dtype contains object #3256
Description
This code works for me on numpy 1.6.2 but on 1.7.1 I get an exception, because in the new version numpy.core._internal._index_fields calls ary.view() and this doesn't work on object arrays.
import numpy as np
ra = np.recarray((2,), dtype=[('x', object), ('y', float), ('z', int)])
ra[['x','y']]
Example on 1.7.1:
In [13]: ra[['x','y']]
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
.../python-271-20110303-2/lib/python2.7/site-packages/numpy/core/records.pyc in __getitem__(self, indx)
455
456 def __getitem__(self, indx):
--> 457 obj = ndarray.__getitem__(self, indx)
458 if (isinstance(obj, ndarray) and obj.dtype.isbuiltin):
459 return obj.view(ndarray)
.../lib/python2.7/site-packages/numpy/core/_internal.pyc in _index_fields(ary, fields)
294
295 view_dtype = {'names':names, 'formats':formats, 'offsets':offsets, 'itemsize':dt.itemsize}
--> 296 view = ary.view(dtype=view_dtype)
297
298 # Return a copy for now until behavior is fully deprecated
.../lib/python2.7/site-packages/numpy/core/records.pyc in view(self, dtype, type)
495 if dtype.fields is None:
496 return self.__array__().view(dtype)
--> 497 return ndarray.view(self, dtype)
498 else:
499 return ndarray.view(self, dtype, type)
.../lib/python2.7/site-packages/numpy/core/records.pyc in __setattr__(self, attr, val)
437 if attr not in fielddict:
438 exctype, value = sys.exc_info()[:2]
--> 439 raise exctype, value
440 else:
441 fielddict = ndarray.__getattribute__(self, 'dtype').fields or {}
TypeError: Cannot change data-type for object array.