…he banner.

Instead of requesting the kernel info a second time, just use the original kernel info we request for a kernel connection.

This makes sure we don’t miss messages during the restart, and helps the browser not have so many different websocket connections.

Now that we don’t reconnect when restarting, we’ll just show the banner right away.

… make it.

Tests are still failing. This commit is a checkpoint in the work.

…to true along with resolving the promise.

…anything)

This gets the test suite passing, and narrows down our work to the tests that are still failing.

I think the handler wasn’t getting removed after the check, so it was failing on the next signal emission. testEmission automatically removes the handler after the check, taking care of this issue.

…ring kernel state.

Before, as soon as one kernel future rejected, clearState would return. Now all outstanding futures must resolve or reject before the state is considered cleared.

If a session or kernel is disposed, the requestKernelInfo done promise might reject. In that case, we should do nothing.

Kernel info is requested by the onWSOpen function already, which is what sets the connected state, so no need to request the kernel info here too.

…e as well.

…rejection.

This fixes a number of tests as well.

Also clean up obsolete todo items.

jupyter/notebook#3705 noted some issues with kernel messages. I think these were mainly because we disconnected and reconnected the websocket on restart. We no longer do this, and these tests seem to pass, so let’s remove the TODO and keep an eye on it.

…n to match existing behavior.

Throwing an error when shutting down a dead kernel causes a lot of test failures. So we keep the existing behavior and update the documentation, rather than changing the behavior to match the documentation.