[NumPy-Tickets] [NumPy] #1452: pickle memmap'd arrays (patch + tests)

[NumPy Trac]

#1452: pickle memmap'd arrays (patch + tests)
-------------------------+--------------------------------------------------
 Reporter:  brentp       |       Owner:  somebody    
     Type:  enhancement  |      Status:  needs_review
 Priority:  normal       |   Milestone:              
Component:  Other        |     Version:              
 Keywords:               |  
-------------------------+--------------------------------------------------

Comment(by charris):

 I think I'll just apply the previous patch. Because there is no
 __setstate__ this bit from the docs worries me:

 {{{
 object.__setstate__()

     Upon unpickling, if the class also defines the method __setstate__(),
 it is
 called with the unpickled state. [5] If there is no __setstate__() method,
 the
 pickled state must be a dictionary and its items are assigned to the new
 instance’s
 dictionary. If a class defines both __getstate__() and __setstate__(), the
 state
 object needn’t be a dictionary and these methods can do what they want.
 [6]

 }}}

 However, now I'm dubious about __getslice__, the problem being when the
 mmap is created with an offset:

 {{{
         if sys.version_info[:2] >= (2,6):
             # The offset keyword in mmap.mmap needs Python >= 2.6
             start = offset - offset % mmap.ALLOCATIONGRANULARITY
             bytes -= start
             offset -= start
             mm = mmap.mmap(fid.fileno(), bytes, access=acc, offset=start)
         else:
             mm = mmap.mmap(fid.fileno(), bytes, access=acc)
 }}}

 So the offset that works on the already mmapped file may not work with the
 file mmapped after unpickling, it looks like you may need the value of the
 start variable also. I don't think this is going to play well with the
 current version of __new__, although I think you could do it with __init__
 and the get/set state stuff. It is also possible that this will ruin
 portability of the pickled objects between python versions/operating
 systems, but I suppose we can live with that. Thoughts?

-- 
Ticket URL: <http://projects.scipy.org/numpy/ticket/1452#comment:6>
NumPy <http://projects.scipy.org/numpy>
My example project