Residual JVM After Quitting

[drguildo]

Whenever I quit emacs or run nrepl-quit I get an orphaned JVM process remaining.

Attached should be an image showing the emacs process tree after running nrepl-jack-in.

I also tested with CIDER 20131018.1553 from a fresh .emacs.d with just cider/clojure-mode and had the same problem.

[drguildo]

This Stack Overflow comment seems to indicate that the bug has been around since at least April.

[bbatsov]

I'll investigate the problem. Might be something Windows related, though - since I don't think I've seen orphaned processes on OSX and Linux.

[bbatsov]

This has to be something Windows specific, since I cannot reproduce it on OSX.

[drguildo]

I'd be happy to help you debug the problem if you don't have a Windows install handy.

[drguildo]

It's probably worth noting that I don't get this problem with any other programs that spawn a copy of the JVM. For example Leiningen spawns the same process structure but both java.exe processes are terminated when it exits.

[bbatsov]

You should debug what happens in cider--close-buffer when it's passed the connection buffer. You might instrument that function with C-u C-M-x and just invoke cider-quit. delete-process should normally kill the underlying Java process, but for some reason it seems that this is not happening on Windows.

[drguildo]

I'm not familiar enough with nrepl.el/CIDER or elisp to know what I should be looking for.

It seems the parent java.exe process is getting killed, just not the child for some reason.

[bbatsov]

@drguildo Might be an Emacs bug then, since delete-process is a core Emacs function. You may get more info regarding this if you post a question on the emacs-devel mailing list.

[drguildo]

I filed an Emacs bug and this is what one of the developers had to say:

Emacs on Windows can only monitor and kill its immediate subprocesses, it cannot monitor, let alone kill, any of their descendant processes, because it has no idea about them. And the OS doesn't automatically kill all the processes in the subprocess tree, and there's no way to send a signal to them all, as on Posix platforms. If killing the immediate child process doesn't cause some of its children to exit or abort, then those grandchildren will be left orphaned.

Why doesn't the child java.exe exit when it parent does? Can you arrange for that to happen? Failing that, I don't think there's a solution to this problem, sorry.

[bbatsov]

So I was correct when I assumed this was an Windows-related "bug". Since I don't use Windows, I don't know what we can do at this point. I guess we need some explicit tracking of processes there, but someone else will have to implement it.

[drguildo]

If I run (System/exit 0) in the REPL, all Java processes are killed. Would it be possible to do this or something similar in cider--close-buffer?

[bbatsov]

I guess we can. I wonder if there are even simpler solutions to this problem.

[noisesmith]

IMHO assuming that the OS will kill a child process if we exit is not simple, it is sloppy. My vote is for explicitly shutting down any owned VM on shutdown of the host, and considering anything created via jack-in as owned.

[bbatsov]

@noisesmith Perhaps you're right. On the other hand I don't recall ever seeing someone tracking explicitly child processed in Emacs Lisp code. I'll look into that.

@drguildo A friend of mine (who's a crazy Windows hacker) checked that claim on emacs-devel and created a patch that debunks it - see the patch he provided here.