nrepl-server-filter called, even though process buffer is dead

[jonpither]

Trying to debug #390. This may or may not be related:

I often see an error when I kill CIDER in windows using cider-quit:

Debugger entered--Lisp error: (error "Selecting deleted buffer"))
nrepl-server-filter(#<process nrepl-server> "^CTerminate batch job (Y/N)? ")

This happens inside of nrepl-server-filter, specifically the line:

(with-current-buffer (process-buffer process))

TBH my knowledge of process buffers and sentinels etc is a bit naive. But it seems to me that CIDER kills the process buffer, yet this server-filter is still hanging around to intercept output from the nREPL process which is still on-going, in particular the output "Terminate batch job (Y/N)", which always happens on windows when you try to exit from batch jobs (.bat files).

So my question is, a) can you help me understand how this nrepl-server-filter is being called for a process with a dead corresponding process buffer, and b) could CIDER have some defensive code? I.e. don't call with-current-buffer on a potentially dead buffer.

[Malabarba]

So my question is, a) can you help me understand how this nrepl-server-filter is being called for a process with a dead corresponding process buffer, and

I'm never quite sure when it comes to Emacs+Windows+Subprocesses, so it is possible, and your analysis sounds reasonable.

b) could CIDER have some defensive code? I.e. don't call with-current-buffer on a potentially dead buffer.

If we are sure that this is happening (and I tend to agree), then yes, we should add defensive code (with comments explaining why).

[jonpither]

Thanks. I'll try out some defensive code on Monday, when I'm back in a windows environment.

[Malabarba]

Thanks. Feel free to open a PR when you have it. Should be a simple of matter of wrapping that code in a

;;; In Windows this can be false.
(when (buffer-live-p (process-buffer process))
  ...
  )

[jonpither]

Yes I'll do it tomorrow