Picking wrapped classes

[rabernat]

Thanks for this very useful package! It solves a genuine problem with python's "all or nothing" approach to async programming.

I noticed that the classes produced by Synchronizer.create_blocking are not pickleable, even if the classes they wrap are.

I then had a look at the tests and saw this comment:

synchronicity/test/pickle_test.py

Lines 12 to 20 in f65617b

	@pytest.mark.skip(reason="Let's revisit this in 0.2.0")
	def test_pickle():
	s = Synchronizer()
	BlockingPicklableClass = s.create(PicklableClass, Interface.BLOCKING)
	obj = BlockingPicklableClass()
	assert obj.f(42) == 1764
	data = pickle.dumps(obj)
	obj2 = pickle.loads(data)
	assert obj2.f(43) == 1849

This makes me think there is some history here...

If I wanted to make my wrapped classes pickleable, could you give me any pointers on where to start?

[erikbern]

honestly, i don't remember – i just tried to run the test again and it fails in some very opaque way :(

synchronicity is overdue for a refactoring – i think it would be good to revisit things like this!

[freider]

iirc, one of the tricky aspects of this is to figure out which Synchronizer instance to attach an unpickled instance to, in a way that makes unpickles compatible both with processes having the "original" synchronizer and those that don't have an existing synchronizer.

One idea would be to always associate objects to synchronizers by some kind of global name (typically a library's package name?) and then just make that part of the pickled object. Then when unpickling you'd "get or create" that synchronizer.

Adding this as a TODO for a future 1.0/2.0 synchronicity release, since it's a pretty important feature to enable things like multiprocessing support.