Live Share support

[rchiodo]

Support live share with Python Interactive window.

This entails

• Passing commands back to the guest so they can cause the guest to open a window too
• Caching responses on the host to pass back to the guest (as the guest doesn't have real access to the ipython instance)
• Writing functional tests for the whole thing
• Port forwarding the jupyter server instance to the guest so when we get to jupyter extensions being hosted, they can communicate with the iPython console too.

[rchiodo]

It doesn't work because of this issue for now:
microsoft/live-share#1480

Waiting on feedback from live share team.

[rchiodo]

So I can kind of get this to work if I turn off all schemes for our code lens. But then the command actually runs on the remote machine so the user doesn't get any of the benefit of it.

This work is going to require much more thinking about this. We probably need to start with a meeting with the liveshare team. They don't publish any sort of API to get our stuff to talk with them. That's the starting point I'd say.

Otherwise we'd have to

• Somehow detect a command is from a remote user
• Send the results to both sides

[rchiodo]

@ronglums do you have any indicators on the priority of this? If it's rather high, I can start looking for who to ping for a meeting.

[ronglums]

@rchiodo let's get the conversation going sooner rather than later..

[rchiodo]

The results of our discussion

• We should be able to port forward a port on the host machine to the guest machine and use this as our connectioninfo in the guest. Jupyter's controls will just think this is a jupyter server running as the tcp server should just forward everything along
• CodeLens are provided by the host machine, not the guest machine. We probably just need to eliminate our use of schemes in our provider registration.
• We should create a custom server to mirror results between the host/guest. The whiteboard example the live share team is going to send us would be a good starting point for this.
• We should NOT permit running cells from the guest unless the host grants it.
• We can initiate the live share connection from the guest on extension activation

[rchiodo]

Live share has an npm called vsls:
https://www.npmjs.com/package/vsls

[rchiodo]

Current status

• Enabled behind a feature flag
• Local to local works for most scenarios
• Still need to write functional tests
• Still need to verify all of the react side commands work across boundaries

[rchiodo]

Closing this in favor of smaller tasks.

Work left to do:

• Submit change in algorithm for using addcode instead of command duplication
• Functional tests
• True remote test

[rchiodo]

See items

#3609
#3603
#3610

[damienrj]

What is the status of this, is it working now? Would like to use this for remote interviews. At least I have been been able to get the other participant to see the Interactive python window. Code does share a port, and the other person can see the kernal.

[rchiodo]

Yes it should be finished now. Both sides need the python extension. Be careful though. The other person can run any python code they want and it runs inside your jupyter kernel.

We're thinking about making a 'read-only' mode for the guest so that they can only see the interactive window but can't send cells to it.

[damienrj]

Read only sounds good as well. The kernel gets shared but I can't see the interactive window, and I can connect to the kernel if I go to the shared server so it is close to working. However, trying to run cells on the remote machine give me the message.

The host doesn’t allow running this command. If needed, ask them to enable it.

It sounds like this is supposed to be working so I can open a ticket if that is best.