[Numpy-tickets] [NumPy] #299: Downcast unsafe when using array protocol?

[NumPy]

#299: Downcast unsafe when using array protocol?
------------------------+---------------------------------------------------
 Reporter:  faltet      |       Owner:  somebody   
     Type:  defect      |      Status:  new        
 Priority:  normal      |   Milestone:  1.0 Release
Component:  numpy.core  |     Version:  devel      
 Severity:  minor       |    Keywords:             
------------------------+---------------------------------------------------
 In numpy it seems that downcasting is possible:

 {{{
 >>> i1 = numpy.array([0,1], dtype='Int8')
 >>> i1[...] = numpy.array([8,9], dtype='Int32')
 >>> i1
 array([8, 9], dtype=int8)
 }}}

 even if an overflow does occur:

 {{{
 >>> i1[...] = numpy.array([800,900], dtype='Int32')
 >>> i1
 array([  32, -124], dtype=int8)
 }}}

 However, you cannot do the same with other objects compliant with the
 array protocol:

 {{{
 >>> i1 = numpy.array([0,1], dtype='Int8')
 >>> i1[...] = numarray.array([8,9], dtype='Int8')
 >>> i1
 array([8, 9], dtype=int8)   # Good. But...
 >>> i1[...] = numarray.array([8,9], dtype='Int16')
 Traceback (most recent call last):
   File "<stdin>", line 1, in ?
 TypeError: array cannot be safely cast to required type
 >>> i1[...] = numarray.array([8,9], dtype='Int32')
 Traceback (most recent call last):
   File "<stdin>", line 1, in ?
 TypeError: array cannot be safely cast to required type
 }}}

 The same happens with Numeric objects and for all kind of downcastings.

 It would be nice if numpy could support this feature.

-- 
Ticket URL: <http://projects.scipy.org/scipy/numpy/ticket/299>
NumPy <http://projects.scipy.org/scipy/numpy>
The fundamental package needed for scientific computing with Python.