Re: diagnosing emacs hangs

[Nick Dokos <nicholas.dokos@hp.com>]

Matt Price <moptop99@gmail.com> wrote:


> I'm now using emacs for almost everything and of course that's great,
> except that it is essentially a single-threaded OS that currently
> HANGS with some frequency (100% CPU usgte that will continue for hours
> if you let it go.  I think this probably has something to do with
> wanderlust or possibly org-mode (and/or misconfigurations i've made to
> both of these); but at present i cna't be sure since i have no idea
> how to diagnose these hangs.  Can someone give me some general
> directions on how to proceed with the diagnosis, and if you have them,
> some pointers on how you fixed a similar problem that you used to
> have?  Right now it's very frustrating -- I find myself losing
> substantial amounts of work when I kill emacs & maybe more
> importantly, i'm constantly losing my train of thought.
> 
> This is all under Ubuntu Lucid with emacs-snapshot 20090909,
> wanderlust=wl-beta 2.15.9+0.20100303, org-mode 6.34c (some of htese
> are debian sid packages).
> 

I assume only emacs is stuck, so you can open an xterm: what does ``ps
awlx | grep emacs'' say?  In particular, the state and the wchan are of
interest: normally, it should be in S state and waiting on select: idle
and waiting for input. If it's persistently in D state, it's stuck
somewhere in the kernel - the wchan gives an idea where. Do it a few times
to make sure that things are not changing.

The next step is to do ``strace -p<emacs_pid>'' to see whether it's going
in and out of the kernel (perhaps in an infinite loop).

If it is *not* going into the kernel, but accumulates CPU runtime (check
the ps awlx output a few times), then it's stuck in a loop in user
space. Attaching to it with ``gdb -p<emacs_pid>'' and getting a
backtrace should give an idea of where it's stuck. But if the loop is in
lisp code, the backtrace is not going to tell you where: it'll just be
in eval. If that's the case, then bisecting through your .emacs setup is
probably the best idea (maybe start by commenting out the org/wanderlust
stuff, particularly if you started getting these problems recently,
after making changes to their configuration.)

It's always a good idea to do these things with a working emacs first, so
that you learn what "normal" looks like. Then you have a better idea
of what's wrong when you try them on the stuck emacs.

This only scratches the surface but...

HTH,
Nick