Moving to use jupyter_client?

[ngr-t]

The jupyter team provides jupyter_client package, which is the reference implementation of the Jupyter protocol. I'm considering to use this in Hermes. jupyter_client requires many dependencies, so I think its better to run a server on other interpreter than Sublime's internal python.

It will require large change about connecting to server / kernels, but I think the code should be much cleaner, easier to maintain, reduce problems Hermes has now. I'd appreciate your opinions, thanks.

[ngr-t]

Now package control has dependencies function, but I found not all jupyter_client's requirement packages are provided (tornado is not). I think of going with the way of making an independent server running on interpreter other than Sublime's internal.

[ngr-t]

Another way is to use only the part of the package we need to run Hermes.

[ngr-t]

Recently I work on making Hermes to use jupyter_client. It looks working without tornado with a little modification. The performance issues looks significantly improved by that and code will be much cleaner.

My main concern now:

• It will change the way to connect kernels largely, though I feel it will get less confusing than ever.
• The jupyter process spawned by the plugin will stop when the editor hangs up. Using connection file to existing process will be a workaround.
• How to connect remote jupyter process? Maybe we can connect with the way like this.

I'm going to support connection file and I believe it will help us on the situation like connecting to remote kernel or a kernel already running.

[ngr-t]

I've almost done with this, but have to deal with dependency issues.

[ngr-t]

Close.

[yaroslavyaroslav]

Tornado has added to dependencies. So this could be reconsidered.