PR: Trade ipykernel.serialize for cloudpickle and remove use of publish_data

[minrk]

• use custom spyder_msg message type rather than IPython Parallel-specific data_pub.
  This removes the need for special __spy__ names because it is no longer a shared namespace.
• use cloudpickle for serializing objects rather than deprecated ipykernel.serialize
  that is reserved for IPython Parallel.

[pep8speaks]

Hello @minrk! Thanks for updating the PR.

Cheers ! There are no PEP8 issues in this Pull Request. 🍻

Comment last updated on November 29, 2017 at 17:51 Hours UTC

[ccordoba12]

@minrk, thanks a lot for taking the time to help us improve our architecture! You know better than anyone else the Jupyter/IPython intricacies, so we really appreciate it.

We were thinking to use dill instead of cloudpickle to improve serialization, because dill has also the ability to restore a full IPython interpreter session, something we'd like to add in the future.

Any particular advice for using cloudpickle instead?

[minrk]

No, I don't think there's a big separator between dill and cloudpickle. cloudpickle's just my default choice these days. No objection to switching to dill.

[ccordoba12]

Thanks a lot @minrk for this! It's a really great contribution!!

I decided it to move it to our stable branch to avoid Spyder to fail when ipykernel 5.0 is released. I also fixed several issues with Python 2.

[ccordoba12]

Pinging @ghisvail about this one because it adds cloudpickle as a new dep. We hope to not add more new deps in 3.x, but this one was really necessary.