[Numpy-tickets] [NumPy] #930: empty_like and zeros_like behave differently from ones_like

[NumPy]

#930: empty_like and zeros_like behave differently from ones_like
------------------------+---------------------------------------------------
 Reporter:  jbosch      |       Owner:  somebody
     Type:  defect      |      Status:  new     
 Priority:  low         |   Milestone:          
Component:  numpy.core  |     Version:  1.1     
 Severity:  trivial     |    Keywords:          
------------------------+---------------------------------------------------
 The {{{empty_like}}} and {{{zeros_like}}} creation functions, which are
 defined in python in numeric.py, differ in their handling of array-subtype
 input from {{{ones_like}}}, which is defined in C as a ufunc.

 In particular, if a subtype of ndarray is passed to {{{empty_like}}} or
 {{{zeros_like}}}, neither {{{__array_wrap__}}} or {{{__array_finalize__}}}
 is called, which makes it impossible for subtypes to define any extra
 attributes they might need.

 I believe this behavior could be fixed entirely by switching from:

 {{{if isinstance(a,ndarray):}}}

 to

 {{{if type(a)==ndarray:}}}

 at the top of those two functions, though there are probably more
 efficient ways to accomplish the same thing.

-- 
Ticket URL: <http://scipy.org/scipy/numpy/ticket/930>
NumPy <http://projects.scipy.org/scipy/numpy>
The fundamental package needed for scientific computing with Python.