Randomly losing connections under emacs-ipython-notebook with version 1.11

[millejoh]

I have haphazardly documented in millejoh/emacs-ipython-notebook#559. The problems seem to start sometime after commit 0d96ba2.

Unfortunately I have very little information to contribute as all the debug buffer states is [WS] State change to connection broken by remote peer. This does appear to be an issue with the current release (1.11) of websocket, though, since if I revert to an earlier version I stop seeing the disconnects.

Let me know if there is anything else I can do to help track down the cause of this.

[ahyatt]

I'll see if I can reproduce. None of the commits recently look like they should effect connections at all, although maybe the recent change in error handling behavior caused some hidden problem to surface. Anyway, if I can reproduce it, I'll fix it.

[ahyatt]

According to @dickmao, #59 fixes this issue, which I've merged into head.

Can you confirm this fixes the issue? If so, I'll rev. the version to 1.11.1.

[ahyatt]

I've also made an addition small fix beyond the basic fix @dickmao provided in 37ef222. However, I was never able to reproduce this - any clues how I can try to reproduce this sort of thing in a functional test would help me make sure I don't break you again.

[nnicandro]

I ran into this issue as well. The server I was communicating with was closing the websocket connection due to an invalid frame being sent. It looks like encode-coding-string causes issues if used in websocket-mask that are dependent on the specific byte pattern used for the masking.

Here is a test case that will always fail when encode-coding-string is used in websocket-mask

(cl-letf ((text "{\"parent_header\":{},\"header\":{\"msg_id\":\"a2940bc8-619e-4872-97bd-4c8d6fb93017\",\"msg_type\":\"history_request\",\"version\":\"5.3\",\"username\":\"n\",\"session\":\"409cf442-74ba-462f-8183-6652503005af\",\"date\":\"2019-06-20T02:17:43.925049-0500\"},\"content\":{\"output\":false,\"raw\":false,\"hist_access_type\":\"tail\",\"n\":100},\"metadata\":{},\"buffers\":[],\"channel\":\"shell\"}")
          ((symbol-function #'websocket-genbytes)
           (lambda (&rest _) "\10\206\356\224")))
  (let ((frame (websocket-read-frame
                (websocket-encode-frame
                 (make-websocket-frame :opcode 'text
                                       :payload (encode-coding-string text 'raw-text)
                                       :completep t)
                 t))))
    (should frame)
    (should (equal (websocket-frame-payload frame) text))))

[millejoh]

@ahyatt - thanks for the quick response. Yes, I can confirm that EIN is working again. I wish I were smart enough to come up with a good test case, but hopefully the one @dzop provided will serve as check against this happening again.