[Yum-devel] 2.9.9 or 3.0?

Michael Stenner mstenner at linux.duke.edu
Fri Oct 6 15:20:24 UTC 2006


On Wed, Oct 04, 2006 at 07:47:38AM +0100, Menno Smits wrote:
> I vote 2.9.9. This searchNevra slowness issue still needs to be fixed.
> 
> Also I've been thinking that a single CTRL-C should always kill Yum for 
> consistency with every other cli app out there. This has been mentioned 
> before on the Yum list. I think another signal like CTRL-\ or CTRL-Z 
> should be used to switch mirrors.
> 
> Thoughts?

I have no design problems with changing this feature, but there are
some major implementation changes.  The problem comes from the fact
that we write in python, and python is written in C.  Our access to
signal handlers is not as perfect as we might like, and so it's hard
to deal with signals well.  SIGINT is easy to deal with because python
has built-in handling of it via the KeyboardInterrupt exception.  To
do that, we do not need to set any signal handlers, but only catch the
exceptions.

If we want to deal with other signals, then we will have to set our
own signal handlers, and that gets tricky.  For example, system calls
(like disk IO, select(), and socket reads) can result in missed
signals.  This is exactly the kind of stuff going on during file
download, so I'd say the odds are decent that we'd run into trouble.

See here for some more details:
  http://docs.python.org/lib/module-signal.html

Bottom line: we might TRY changing the current behavior, but I would
encourage some serious testing.  We may also decide it's not worth it.
I'll pitch in on any urlgrabber changes either way.

					-Michael
-- 
  Michael D. Stenner                            mstenner at ece.arizona.edu
  ECE Department and Optical Sciences Center                520-626-1619
  University of Arizona                                         ECE 524G



More information about the Yum-devel mailing list