Make inverted index used for index predicates persistent

[fgregg]

http://www.zodb.org/en/latest/documentation/tutorial.html

[fgregg]

Right now the difficulty is in dealing with the incompatibilty between multiprocessing and standard ZODB.

This would be resolved if we used forkserver or spawn, because we don't actually do the multiprocessing until after getting a handle on the indexes.

[fgregg]

Got a minimally working code here: 7e06593

This depends on https://github.com/zodb/relstorage/ which does not support python3. So that is a dead end.

[fgregg]

Okay, I'm going to go another way.

I'm adding methods for reading and writing indices. These should be separate from reading and writing settings, because we usually want to write the indices after training, in particular after we index the real data.

I'm just pickling. When we tried to pickle the indices before, we ran into maximum recursion http://stackoverflow.com/questions/2134706/hitting-maximum-recursion-depth-using-pythons-pickle-cpickle

However, I can't get that error to reproduce. f56de14

[fgregg]

This reliably reproduces on mysql-example. The key thing we need to do is have an index that can be pickled or use a form of pickling that does not depend on recursion: zopefoundation/BTrees#44

[fgregg]

In the meantime, we could provide a much more useful error message.

[fgregg]

closed by #788