[Bug: 21.4.15] Bad interaction between scroll-conservatively, scroll-step and beginning-of-buffer.

[Jerry James]

I wrote:
> With the divide by zero fix I just sent to xemacs-patches, I found that
> the profiler returned pretty much immediately upon pressing M-<, and
> reported that too little time to measure had passed ... whereupon there
> was a long delay until the buffer of interest was redrawn.  So it looks
> like M-< executes pretty much instantly, and the problem is somewhere in
> redisplay.

Look at this code in redisplay.c (lines 6583 - 6598 in current CVS):

  /* We still haven't gotten the window regenerated with point
     visible.  Next we try scrolling a little and see if point comes
     back onto the screen. */
  if (scroll_step > 0)
    {
      int scrolled = scroll_conservatively;
      for (; scrolled >= 0; scrolled -= scroll_step)
	{
	  startp = vmotion (w, startp,
			    (pointm < startp) ? -scroll_step : scroll_step, 0);
	  regenerate_window (w, startp, pointm, DESIRED_DISP);

	  if (point_visible (w, pointm, DESIRED_DISP))
	    goto regeneration_done;
	}
    }

Setting a breakpoint here with GDB shows that this part of the code is
NOT hit when doing M->, but is when doing M-<.  Basically, on M-< with
Lawrence's variable settings, we are scrolling the buffer back one line
at a time and asking, "Are we there yet?  I have to go to the bathroom!
Jimmy's touching me!"

If the place we are trying to reach is "close", this makes sense.  So
why don't we get here when doing M->?  Does some code elsewhere note
that we are going "far away" and skip this altogether somehow?
-- 
Jerry James
http://www.ittc.ku.edu/~james/