Websocket channels

[rgbkrk]

This completes our roundup of the kernels API. Usage:

const socketSubject = jupyter.kernels.connect(serverConfig, kernelID);

socketSubject.subscribe(message => ...);

socketSubject.send(JSON.stringify(message)); // message is a Jupyter message

Careful notes though, which bears caution of a changing API to make things easier:

• Jupyter WebSockets are now multiplexed, which has an effect on the messages - they include a channel key for iopub, shell, or stdin.
• Jupyter messages on the websockets appear to extract and include a few keys out of the header, namely msg_type and msg_id (they do also appear in the header). It's not clear to me if we have to set these at a higher level for the server.
• The WebSocketSubject appears to automatically deserialize messages from the server on the websocket into JavaScript Objects - however (and wisely so as I'll point out in a second), you must JSON.stringify messages you send to the backend
• Binary buffers are not handled here

In order to make this conform to the same interface we use with enchannel-zmq, it will require some extra adaptation. All that being said, this gets us started!

/cc @minrk for awareness

[minrk]

Cool!

It's not clear to me if we have to set these at a higher level for the server.

You shouldn't, and we probably don't even mean to be including them over the network if we are. The Python API copies msg_id and msg_type out of the header to the top-level for convenience, but this is not part of the network-level spec. A message is only a collection of dicts: header, parent_header, content, metadata; plus the buffers list. It has no top-level attributes per the spec.