Remote REPL

[leviska]

Check for existing issues

• [X] Completed

Describe the feature

Run jupyter repl on remote machines

From what I could understand from code:

• Currently repl_store is global, and not updated when changing projects (i.e. connecting to remote project)
• JupyterSettings are global too, and project settings are ignored (there is a big chance that remote machine needs different kernel configs)

It seems that current code could work on remote, if it weren't global, and it would update to changing workspaces and connecting to remote.

If applicable, add mockups / screenshots to help present your vision of the feature

No response

[leviska]

I was trying to do this myself, but got a little bit stuck due to lack of knowledge around code base:

• I could get my local config to match remote config, but then it would actually read local jupyter kernelspec instead of remote. I guess that's because I first open local workspace, then remote, and the global repl_store is initialized with local fs, not remote (funny enough it's reading local kernelspec but then runs it on remote)
• I don't quite understand the difference between project/workspace/editor and it's relation to each other (see next)
• As far as I understand, we need to observe_new_views inside ReplStore::init, and create a new repl for each (workspace/editor)
• I was trying to look at file_finder to understand, and it works with workspaces, but here we also need editor too to get the code that we want to run, so I got a little stuck with a problem of matching both states correctly. It seems we need to have only one of (editor/workspace), and get the other from that, but I don't get it what the "parent/child" relationship here if it even exists

[pipegalera]

Same problem here:

[KorigamiK]

I am also facing this issue and adding a default kernel selection for the remote project in its settings.json also doesn't seem to fix this issue, even though the lsp seems to understand the remote file system correctly

[rgbkrk]

Confusingly, I'm working on support for remote jupyter servers which is not the same as the kernels via remoting (over SSH). I'll need some verbiage to differentiate between these.