[MPlayer-users] Segmentation fault in pre5

[Martin Simmons]

> On Sat, Oct 16, 2004 at 11:45:35PM +0100, Martin Simmons wrote:
> > >>>>> On Sat, 16 Oct 2004 18:06:50 -0400, D Richard Felker III <dalias at aerifal.cx> said:
> > > On Sat, Oct 16, 2004 at 08:33:26PM +0200, Stefan Sassenberg wrote:
> > > > On Fri, Oct 15, 2004 at 07:19:32PM -0700, Corey Hickey wrote:
> > > > > Stefan Sassenberg wrote:
> > > > > > Hello,
> > > > > > 
> > > > > > I frequently have segmentation faults when mencoding a movie.
> > > > > > Unfortunately this happens at different positions. I can provide some
> > > > > > information about this problem. I should add that I had other problems
> > > > > > of similar kind that I only solved by rebuilding mencoder with
> > > > > > ./configure --disable-mmx2 which I'm currently using.
> > > > > > 
> > > > > 
> > > > > Are you saying that it happens at a different position each time you try
> > > > > the same command on the same file? If so, then you probably have
> > > > > unstable hardware. I've found that mencoder is very intolerant of
> > > > > overclocking, probably from a combination of working the CPU very hard
> > > > > and the time it takes to encode.
> > > > > 
> > > > > If you are overclocking, try overclocking a little less; or, try cooling
> > > > > your CPU better.
> > > > > 
> > > > Correct. Just like the other problem I once had that I mentioned above.
> > > > I also thought about hardware problems, but mencoder is the only application
> > > > that shows errors like these and I'm not overclocking. Heat problems
> > > > might be the cause anyway.
> > > 
> > > anything that happens at a random location each time you run a program
> > > is _definitely_ heat-caused. mencoder doesn't read from /dev/urandom
> > > and pick what to do based on the results. the results of each run are
> > > entirely deterministic and won't vary from one to another.
> > 
> > There is plenty of other non deterministic problems that are not caused by
> > heat or urandom.  Particular cases that I've seen (not in mencoder yet):
> > 
> > - timing-related issues for interaction between threads or processes
> not threaded. this is one reason threaded apps suck.

Yes, I know mencoder is not using threads (though it does link with libpthread
so let's hope nothing calls those fns :-).

The other issue is processes -- mencoder calls gettimeofday() like there is no
tomorrow, so anything else running on the machine could cause subtle changes
in the behaviour of time-critical code.


> > - OS bugs (causing registers to be corrupted)
> never seen this myself...maybe windows? :o

Solaris and VxWorks.

__Martin