Large websocket responses get truncated and cannot be parsed

Description

Sometimes a very long response from the language server gets truncated and the relevant action is not performed. This appears to be somewhat random, because subsequent attempts can sometimes succeed.

Uncaught SyntaxError: Unexpected end of JSON input
    at JSON.parse (<anonymous>)
    at i.readMessage (index.js:formatted:1)
    at index.js:formatted:1
    at WebSocket.e.onmessage (index.js:formatted:1)
readMessage @ index.js:formatted:1
(anonymous) @ index.js:formatted:1
e.onmessage @ index.js:formatted:1

The message truncates at less than 7k characters

Similar issue was reported in a different project using PHP backend here: https://github.com/beyondcode/dusk-dashboard/issues/20.

Reproduce

1. Open an R notebook
2. Type base:: (potentially press <kbd>Tab</kbd>, but should auto-invoke)
3. See lack of completion suggestions and error in console

Expected behavior

1. We may want to try to recover truncated JSON messages by parsing the viable portion (e.g. for completion it would make sense to show something rather than nothing, especially for extra long namespaces such as base in R)
2. The error should not happen in the first place

Context

• Operating System and version: Ubuntu 20.10
• Browser and version: Chrome
• JupyterLab version: 2.2.x

master on 2.1.2

[dpriedel]

I’ll try using my hack for a day or so and if I don’t run into any problems, I’ll create a PR

[dpriedel]

I believe I’m using the fixed code from above. I have jupyter-lsp 1.1.1 and jupyterlab-lsp 3.2.0

I’m seeing this truncated output with the Jedi language server, pretty much all the time.

By way of an experiment, I changed line 122 in the _read_content method of the stdio.py from ‘break’ to ‘continue’.

The problem went away.

I wonder if the problem is that this async consumer routine is reading output from an async producer routine which just is not ready with its next chunk of data – jedi language server not being the fastest tool in the shed.

So, maybe changing the ‘break’ to ‘continue’ and max_parts to some smaller number is a better way to go ??

I’m going to keep this change for a while and see how it goes.

Thanks