Re: noverlay branch

[Gerd Möllmann]

> What I'm less clear about is the use of the `null` sentinel node.
> It seems that this node is sometimes modified (e.g. its color changed
> from black to red or vice versa, and maybe also its `parent` field),
> even though it's pointed to from lots of different nodes, so any help
> documenting the way it works (or why the value in those fields doesn't
> matter) would be welcome.

Maybe I can say something because I used a similar trick in alloc.c. 
The gist of that was to make searching more elegant, and faster:

static struct mem_node *
mem_find (void *start)
{
  struct mem_node *p;

...

  /* Make the search always successful to speed up the loop below.  */
  mem_z.start = start;
  mem_z.end = (char *) start + 1;

  p = mem_root;
  while (start < p->start || start >= p->end)
    p = start < p->start ? p->left : p->right;
  return p;
}

If that's done for the same reason in itree.c, I don't know.  A hint 
that it is not, might be that each tree has a separated null node...