Re: Torture test for Nautilus

[Federico Mena Quintero <federico ximian com>]

On Wed, 2006-10-18 at 18:10 +0200, Alexander Larsson wrote:

> How long did you have to run it to get a failure? I ran if for a while
> and didn't see anything (HEAD). I'll start a longer run now to test.

[Hmm, I'm using Nautilus from GNOME 2.12.]

It ran for about 10 seconds, then got really slow (in the monitor code),
then crashed about 30 seconds afterwards when it got out of the
monitoring code.

> I made some performance fixes to gnome-vfs-monitor.c to make this less
> total ass performance-wise, but the fact is that you're running an
> application that creates file notification events as quickly as the cpu
> can generate them. We can't really do anything but spend 100% of the
> time handling incoming events. 

Yeah, good point.  It didn't have trouble when I inserted a very small
usleep() between each operation.  I agree that "instant changes" is not
a case we should contemplate performance-wise.

> > The torture program supports a "--seed" option that you can use to seed
> > the random number generator.  This is so that the sequence of "random"
> > torture steps will be reproducible.
> 
> However, it doesn't print the seed used on startup, so you don't know
> what seed to use the second time.

Good point! :)

Should I commit the script to nautilus/test/?  Then we can add this kind
of stuff easily.

Thanks for taking a look; I hope this is useful to find some weird
corner cases.

  Federico