Re: tramp (2.1.15); TRAMP doesn't recognize subprocess death

[David Abrahams]

On Apr 13, 2009, at 12:38 PM, David Abrahams wrote:

> On Apr 12, 2009, at 2:33 PM, Michael Albinus wrote:
>
> > David Abrahams <address@hidden> writes:
> >
> > > Visit a remote file.
> > > `M-x compile'
> > > `ls'  ;; or whatever short command you please
> > > `M-x compile'
> > >
> > > Compilation mode will tell you a compilation is still running and  
> > > ask if
> > > you want it killed.  This is not a serious problem for compilation,
> > > although it is an inconvenience.  It's worse for something like  
> > > psvn,
> > > which relies on being able to know that the svn process has  
> > > finished.
> >
> > In Tramp 2.1.16-pre this was fixed some weeks ago. Do you see a  
> > chance
> > to test it, using Tramp's CVS repository? This code is not merged yet
> > with GNU Emacs.
>
>
> Looks like it works, thanks!

Hmm, I take it back, maybe.  It could also be a psvn.el bug, or maybe  
it's a bug in emacs' whole system for handling asynchronous processes;  
I'm not sure.
See 
http://svn.collab.net/viewvc/svn/trunk/contrib/client-side/emacs/psvn.el?revision=37144&view=markup#l1282

This code essentially launches a process and then sets the process  
sentinel.  What's happening is that -- sometimes only, maybe when the  
process output is large enough -- the process sentinel never gets  
called.  The Emacs documentation isn't very clear on what is supposed  
to happen in that case.  Is there a race condition between the process  
finishing and the setting of the sentinel?  If the process finishes  
first, is the sentinel still supposed to be called?  It seems like the  
answer /ought/ to be yes, since otherwise there is a race condition,  
and Emacs doesn't seem to give you anyway to set up the sentinel at  
the same time as (or before) launching the process.

--
David Abrahams
BoostPro Computing
http://boostpro.com