[NumPy-Tickets] [NumPy] #1332: set_fill_value does not work properly for masked structured array

[NumPy Trac]

#1332: set_fill_value does not work properly for masked structured array
-------------------------+--------------------------------------------------
  Reporter:  robitaille  |       Owner:  pierregm
      Type:  defect      |      Status:  reopened
  Priority:  high        |   Milestone:          
 Component:  numpy.ma    |     Version:  devel   
Resolution:              |    Keywords:          
-------------------------+--------------------------------------------------

Comment(by robitaille):

 The test code I posted works fine - however, I had strange problems with a
 larger program and managed to narrow it down to the following example.

 The following code works as expected:

 {{{
 import numpy.ma as ma
 import numpy as np

 dtype = ('dec', np.dtype('float64'))

 a = ma.array([(1.,),(2.,),(3.,),(4.,)],
              mask = [(False,),(False,),(False,),(False,)],
        fill_value = (np.nan,),
             dtype = [('ra', '<f8')])

 newdtype = np.dtype([('ra', '<f8'), dtype])

 b = ma.empty(a.shape, dtype=newdtype)

 b['ra'] = a['ra']
 print repr(b)

 b['ra'].set_fill_value(a['ra'].fill_value)
 print repr(b)
 }}}

 with the output:

 {{{
 masked_array(data = [(1.0, 2.1304385512215302e-314) (2.0,
 2.1243859343164211e-314)
  (3.0, 2.1304549062766045e-314) (4.0, 5.5626846462680035e-309)],
              mask = [(False, False) (False, False) (False, False) (False,
 False)],
        fill_value = (1e+20, 1e+20),
             dtype = [('ra', '<f8'), ('dec', '<f8')])

 masked_array(data = [(1.0, 2.1304385512215302e-314) (2.0,
 2.1243859343164211e-314)
  (3.0, 2.1304549062766045e-314) (4.0, 5.5626846462680035e-309)],
              mask = [(False, False) (False, False) (False, False) (False,
 False)],
        fill_value = (nan, 1e+20),
             dtype = [('ra', '<f8'), ('dec', '<f8')])
 }}}

 However, simply commenting out the first print statement results in the
 set_fill_value no longer having an effect:

 {{{
 import numpy.ma as ma
 import numpy as np

 dtype = ('dec', np.dtype('float64'))

 a = ma.array([(1.,),(2.,),(3.,),(4.,)],
              mask = [(False,),(False,),(False,),(False,)],
        fill_value = (np.nan,),
             dtype = [('ra', '<f8')])

 newdtype = np.dtype([('ra', '<f8'), dtype])

 b = ma.empty(a.shape, dtype=newdtype)

 b['ra'] = a['ra']
 # print repr(b)

 b['ra'].set_fill_value(a['ra'].fill_value)
 print repr(b)
 }}}

 gives output

 {{{
 masked_array(data = [(1.0, 2.1304385522096615e-314) (2.0,
 2.1243859343164211e-314)
  (3.0, 2.1304549062766045e-314) (4.0, 5.5626846462680035e-309)],
              mask = [(False, False) (False, False) (False, False) (False,
 False)],
        fill_value = (1e+20, 1e+20),
             dtype = [('ra', '<f8'), ('dec', '<f8')])
 }}}

 So essentially, the presence or absence of print repr(b) is changing
 things.

-- 
Ticket URL: <http://projects.scipy.org/numpy/ticket/1332#comment:9>
NumPy <http://projects.scipy.org/numpy>
My example project