bug#1112: 23.0.60; Child process not cleaned up properly

[Brent Goodrick]

Hi Sven,

I agree about the part about the process permissions: Emacs definitely
cannot kill the process because it is a sudo process.  However, if
Emacs fails to kill the process, Emacs should emit a warning message
as to the reason why the process could not be killed (e.g., "process
id 12345 could not be killed: operation not permitted").  This would
be appropriate in the case where the user kills the process buffer on
a long running process (a process that is not attempting to read from
standard input as this one is; see next paragraph).  This warning is
appropriate because otherwise, the user thinks that the process died
on its own and will have to discover the hard way via some degree of
head-scratching that the process could not be killed.

There is more to this problem than process permissions. The process
has opened up standard input for the Y/N prompt.  When I run
compilation mode on some process, I expect no user prompts (if I
wanted to handle prompts,I would execute it under the ever so handy
shell mode). But, when that process does read standard input for
prompting, I want that process to exit immediately (and most well
behaved apps do this these days, with the exception of Bourne shell
scripts using the "read" operator which doesn't behave correctly IMO).
 As a proof of concept, when I type "sudo apt-get some_package <
/dev/null" into a M-x compile prompt, I see reasonable behavior:

  <snip>
  After this operation, 27.5MB of additional disk space will be used.
  Do you want to continue [Y/n]? Abort.

  Compilation exited abnormally with code 1 at Wed Oct  8 07:18:54

So, is there any harm in having compilation mode close standard input
when it spawns any and all processes (perhaps only under the direction
of a customizable and properly documented defvar variable so as to
avoid backlash from users expecting the existing behavior)?  If so,
this would fix this issue in the majority of cases.   Actually, this
closing of standard input should be an option on the lowest level
Elisp command that compile uses to start processes, if it is not
already exposed as such: I imagine that having to hack this into the
compile mode Elisp code itself would be problematic given that the "<
/dev/null" construct would have to vary quite a bit on the two axes of
type of shell and execution platform.

Thanks,
Brent

On Tue, Oct 7, 2008 at 12:24 PM, Sven Joachim <svenjoac@gmx.de> wrote:
>
> On 2008-10-07 17:15 +0200, Brent Goodrick wrote:
>
> > 1. M-x compile
> > 2. Enter in: sudo apt-get install gimp-help-en
> > 3. See the apt-get prompt:
> >    Reading package lists... 0%
> >    <snip>
> >    The following extra packages will be installed:
> >      gimp-help-common
> >    The following NEW packages will be installed:
> >      gimp-help-common gimp-help-en
> >    0 upgraded, 2 newly installed, 0 to remove and 1 not upgraded.
> >    Need to get 15.9MB of archives.
> >    After this operation, 27.5MB of additional disk space will be used.
> >    Do you want to continue [Y/n]?
> > 4. Kill the buffer, and expect the underlying process to die, just
> >    like you would have if you had typed the above command in a shell
> >    buffer.
>
> Won't work for processes run under sudo, see below.
>
> > 5. Open up a shell, and type ps to see that the apt-get process still
> >    exists
> > 6. Go through step 1 again and notice now that a lock is being
> >    reported by the second apt-get session because the first process
> >    was not properly torn down by the act of killing the previous
> >    compilation buffer.
> >
> > My assessment: The shell mode somehow works differently than the
> > compilation mode since the compilation mode does not allow user
> > input. Fair enough, but the two modes should work the same in terms of
> > tearing down the two processes if the buffers are killed, and should
> > not ever leave dormant child processes.
>
> The real problem is that sudo is suid root and thus the compilation
> process runs with superuser rights.  Emacs is simply lacking the
> privileges to kill it.
>
> You can try something similar in your shell:
>
> ,----
> | % sudo sleep 1000 &
> | [1] 2186
> | % kill %1
> | kill: kill %1 failed: operation not permitted
> | % sudo kill $(pidof sleep)
> | [1]  + 2186 terminated  sudo sleep 1000
> | %
> `----
>
> Sven