[MPlayer-cvslog] CVS: main/libmpdemux aviheader.c,1.63,1.64

[Reimar Döffinger]

Hi,
> > On Sat, Jan 01, 2005 at 03:39:04AM -0500, D Richard Felker III wrote:
> > > On Sat, Jan 01, 2005 at 01:16:55AM -0500, D Richard Felker III wrote:
> > > > On Fri, Dec 31, 2004 at 05:02:50PM +0100, Reimar Döffinger CVS wrote:
> > > > > CVS change done by Reimar Döffinger CVS
> > > > >
> > > > > Update of /cvsroot/mplayer/main/libmpdemux
> > > > > In directory mail:/var2/tmp/cvs-serv30117/libmpdemux
> > > > >
> > > > > Modified Files:
> > > > >  aviheader.c
> > > > > Log Message:
> > > > > fix quicksort to avoid stack overflow and extreme runtimes
> > > > >
> > > > >
> > > > > Index: aviheader.c
> > > > > ===================================================================
> > > > > RCS file: /cvsroot/mplayer/main/libmpdemux/aviheader.c,v
> > > > > retrieving revision 1.63
> > > > > retrieving revision 1.64
> > > > > diff -u -r1.63 -r1.64
> > > > > --- aviheader.c 20 Oct 2004 02:13:33 -0000 1.63
> > > > > +++ aviheader.c 31 Dec 2004 16:02:47 -0000 1.64
> > > > > @@ -44,15 +44,25 @@
> > > > >      return 0;
> > > > >  }
> > > > >
> > > > > -/*
> > > > > +/**
> > > > >   * Simple quicksort for AVIINDEXENTRYs
> > > > > + * To avoid too deep recursion, the bigger part is handled
> > > > > iteratively, + * thus limiting recursion to log2(n) levels.
> > > > > + * The pivot element is randomized to "even out" otherwise extreme
> > > > > cases. */
> > > > >  static void avi_idx_quicksort(AVIINDEXENTRY *idx, int from, int to)
> > > > >  {
> > > > >      AVIINDEXENTRY temp;
> > > > > -    int lo = to;
> > > > > -    int hi = from;
> > > > > -    off_t pivot_ofs = AVI_IDX_OFFSET(&idx[(from + to) / 2]);
> > > > > +    int lo;
> > > > > +    int hi;
> > > > > +    off_t pivot_ofs;
> > > > > +    int pivot_idx;
> > > > > +  while (from < to) {
> > > > > +    pivot_idx = from;
> > > > > +    pivot_idx += rand() % (to - from + 1);
> > > >
> > > > imnsho using rand() is forbidden. it corrupts the state of the caller.
> > >
> > > not to mention, it also makes debugging hell by making the program
> > > non-deterministic.
> >
> > I know that! Don't treat me like stupid an provide a working alternative
> > instead.
> 
> pivot_idx = (from + to)>>1;

Yes, that was the original method. And with the broken files that
mencoder creates the sorting can take more that 60s.

> a better method is to take the median of 3 or more elements

I didn't try that variant.
But as the glibc qsort is already used all over the place I have to
admit it is probably the better solution...

Greetings,
Reimar Döffinger