This simplified logic quite a bit.

Most of these ideas came from previous work in jupyterlab#4724

This also (finally!) resolves the testing issues with kernel info messages being outstanding when the kernel is disposed.

…tatus.

…cture.

…r, etc.

…to the kernel manager.

This simplifies the responsibilities of each module.

…e transferred.

…ng syntax.

Lots of things in the tests still are wrong, but this is a snapshot of the current work.

…mespaces.

Also break out the kernelspec tests from the kernel tests.

Tests pass except for a small number of failures that still need investigating.

I left the term "Kernels" in the ux tooltip since users probably are familiar with that term more than "sessions"

There is still more to do in the completers to handle when there is no session, but this is a start, and the TODO is noted in the code.

…tart.

This won’t handle the restarting logic for other kernel connections to this same kernel. We should figure out what other kernel connections will do. Perhaps the kernel manager should broadcast the restarting across all kernel connections it knows about.

All services/kernel tests are passing!

This documents the change in behavior that a session context is not tied to the session lifecycle.

… when not connected.

We need these to know when/if the initial kernel has been started.

In addition to resolving merge conflicts, we changed to lumino in files without merge conflicts.

I think it’s better to be explicit and have only one place to access it: sessionContext.session?.kernel

…essions” rather than just “kernels”

…method.

Previously, if you tried to create a comm that already existed, you could get the comm someone else had created, which means you would overwrite their message handlers, etc. This was a bad design decision.

This changes the API so that you can only create comms with ids that do not exist. You can also query to see if a given id is used.

See jupyterlab#7284 for more discussion.

This means that ‘disconnected’ is no longer a terminal state, and can be recovered from with an explicit call to reconnect()

We also change the reconnect timeouts so that the first reconnection happens immediately, and subsequent ones happen at random times in growing intervals.

Also make second parameter optional, since each property was already optional.

We don’t want to accept a Partial<Session.IModel> since what we are really doing is setting the values of the top-level .path/.type/.name attributes.

Namespace the ISessionConnection interfaces. Make the connectTo and various function signatures more uniform.

… the manager startNew methods.

This also more generically supports any extra kernel startup options used by servers other than the notebook server, which encompasses the changes in jupyterlab#7499.

Fixes jupyterlab#7640

…xt in the signal connection.

This gives a sensible status to show a user (i.e., if connected, the kernel status, if not, then the connection status). Also heuristically create a “initializing” status for an initially starting kernel.

…eady before using it.

…efaults of true.

… into sessionpersistence

… CI.