Cannot re-connect after repl quits unexpectedly

[pesterhazy]

Thanks for the awesome project! In recent versions of CIDER, I have to restart Emacs to connect to an nrepl after the previous lein process died.

STR

1. lein repl
2. cider-connect
3. In the repl terminal window, ^D
nREPL: Connection closed unexpectedly (connection broken by remote peer)
4. re-start the repl
5. cider-connect
6. Re-use zombie REPL buffers? Yes

Expected result

I can re-evaluate the buffer

Actual result

I see a beach ball (on OSX) and Emacs becomes unresponsive. Evaluating the buffer doesn't work.

*Messages* shows nrepl-send-sync-request: Sync nREPL request timed out (op clone id 2)

System

; CIDER 0.11.0snapshot (package: 20151220.156) (Java 1.8.0_51, Clojure 1.5.1, nREPL 0.2.10)

In this test I'm using leiningen with an ancient version of Clojure, but the same problem exists with boot and Clojure 1.7.

Not that I'm not using cider-nrepl-middleware because it slows down startup of a largish projects by many seconds.

In tests the problem doesn't occur if I cider-quit before exiting the external lein repl (but you sometimes don't have that luxury).

[bbatsov]

You should also post here your *nrepl-messages* log.

[Malabarba]

Sounds like we don't cleanup the buffer when the connection is lost, and then when starting the new connection we try to send a message to the dead connection for some reason (which times out).

[bbatsov]

I think the zombie repl handling code was supposed to do the cleanup, when the dead connection is reused. Don't recall any recent changes there, though.

[pesterhazy]

@bbatsov, sure! I've updated to current lein/nrepl and created an empty project, but the issue persists.

There are two buffers labeled "nrepl messages" (1.txt, 2.txt). I'm also attaching the relevant section of my messages (3.txt).

1.txt
2.txt
3.txt

[pesterhazy]

Also, this is in latest spacemacs (if that matters).

[jgeraerts]

same issue here with cider 0.10.0 on "GNU Emacs 24.5.1 (i686-pc-mingw32)\n of 2015-04-11 on LEG570"

[rfkm]

I think the cause of this issue is that nrepl client unexpectedly picks up an old session id.
The easiest way to fix this is to add (setq nrepl-session nil) into cider-repl-create function.

[bbatsov]

Yeah, we should make sure both the normal and the tooling sessions are reset for the reused REPL buffer. @rfkm feel free to submit a PR. I'm pretty sure this problem was introduced recently. Maybe we should search with bisect for the problematic commit.

[Malabarba]

It was probably introduced when I made the send-request function automatically add the session parameter to the request.
I agree with the suggestion of how to fix it, but I think we should also set that variable to nil in the process sentinel.

[bbatsov]

And the nrepl-tooling-session.