you may want to check against any of these tests that are relevant: https://github.com/cloudpipe/cloudpickle/blob/343da119685f622da2d1658ef7b3e2516a01817f/tests/cloudpickle_test.py#L1170

Can you give a short summary of what you feel is done and what is left to do? I'm heading toward a release, and trying to get some nagging issues/PRs tied up. I'll help with this if possible.

@mmckerns Status update
I spent a bit more time on this today, but it is still not complete.
There seem to be three remaining issues:

1. The implementation for <locals> is not satisfactory. The class path is not the same for the deserialized objects when they are locals rather than normal module objects.
2. The register function does not behave as expected based on the tests and issues observed in CloudPickle.
3. The current implementation does not work for python < 3.6. I believe the if registry is None clause of the cloud pickle implementation handles the post/pre 3.6 implementation change.

I don't think anything that previously worked would be broken by these changes, but I have not confirmed that ABC classes failed to pickle in python <= 3.6

@mmckerns @emfdavid Take a look at https://github.com/anivegesana/dill/tree/issue_332_abc_data and see if it is going in the direction that you want it.

@mmckerns @emfdavid Take a look at https://github.com/anivegesana/dill/tree/issue_332_abc_data and see if it is going in the direction that you want it.

@anivegesana: Yeah, this is a good direction.

@emfdavid Bump

@anivegesana: feel free to assume the answer is yes, as opposed to waiting.

Very cool @anivegesana - 2ce1c9e looks really good.
Tests are much improved using copy instead of dump and load and the implementation is cleaner. I think I can see how you are properly handling the registries now too.
Please don't block on me from here. Thank you!