This is a very incomplete patch. It seems to fix the memory leak, but breaks all
the tests.

http://zuza.appspot.com/81/diff/441/642
File translate/storage/cpo.py (left):

http://zuza.appspot.com/81/diff/441/642#oldcode123
Line 123: gpo.po_message_add_filepos.argtypes = [c_int, STRING, c_int]
I assume this solves the -1 issue? Crafty!

http://zuza.appspot.com/81/diff/441/642
File translate/storage/cpo.py (right):

http://zuza.appspot.com/81/diff/441/642#newcode143
Line 143: gpo.po_message_set_format.argtypes = [c_int, STRING, c_int]
Not sure what this has to do with addlocations?  You must keep patches separate.

http://zuza.appspot.com/81/diff/441/642#newcode386
Line 386: origins = ['translator', 'developer', 'programmer', 'source code']
Maybe call them note_types

http://zuza.appspot.com/81/diff/441/642#newcode402
Line 402: return unit
Generally I think this part looks good

http://zuza.appspot.com/81/diff/441/642#newcode461
Line 461: ]
I thought we could get a list of valid formats from the library

http://zuza.appspot.com/81/diff/441/642#newcode565
Line 565: gpo.po_message_set_msgctxt(self._gpo_message, context)
Again one piece at a time.  If you can separate out context from locations from
-1 , etc then there is more chance of a fix being committed.

http://zuza.appspot.com/81/diff/441/641
File translate/storage/test_cpo.py (right):

http://zuza.appspot.com/81/diff/441/641#newcode94
Line 94: assert not unit.hastypecomment("invalid-format")
Should we not raise an exception if we shouldn't be setting the format?

http://zuza.appspot.com/81/diff/441/641#newcode96
Line 96: 
My guess is that these should go into the common tests ie test_po.py

I'm splitting this patch, so you or someone else can review it in small pieces.
http://zuza.appspot.com/101
http://zuza.appspot.com/102
http://zuza.appspot.com/103

Thanks for the review. I'll update this patch when all the others was correct.