bug#32874: Unwanted scrolling in edebug `f' command when follow-mode is active

[Alan Mackenzie <acm@muc.de>]

Hello, Eli.

On Sun, Sep 30, 2018 at 10:35:50 +0300, Eli Zaretskii wrote:
> > Date: Sat, 29 Sep 2018 20:41:13 +0000
> > Cc: 32874@debbugs.gnu.org
> > From: Alan Mackenzie <acm@muc.de>

> > > I'd actually urge you to have a good look at
> > > window-scroll-functions as well.  (Follow mode already uses it, but
> > > I think it could use it for quite a lot more.)  This hook is called
> > > when Emacs concludes that a window may need to be scrolled to bring
> > > point into view.  This is exactly where Follow mode wants to be
> > > able to affect the decision of the display engine, right?  I think
> > > by making a few simple changes/extensions where this hook is
> > > called, we could make the work of Follow mode quite a lot easier,
> > > by letting it rely on the display engine instead of trying to
> > > maneuver the display engine to do what it wants.

> > I've had a look at window-scroll-functions, but I can't see what you
> > must be seeing.  Currently, the documentation warns against trying to
> > influence the scrolling, saying "it probably won't work anyway".

> But you don't want to scroll yourself, you just want to switch the
> selected window and move point so that Emacs won't need to scroll.

> AFAIU, follow-mode wants to kick in when point goes off the selected
> window.  And the call to window-scroll-functions is exactly the place
> where the display engine decides it needs to scroll the window, but
> didn't actually scroll it yet.  So that looks like a good place to have
> follow-mode do its thing.  We might need to add some simple facility
> for follow-mode to use, so that it could signal the display engine not
> to scroll the window.  Other than that, I think this possibility is
> worth exploring.

Follow-mode also needs to be active on explicit scrolling commands such
as C-v.  Also, after inserting a newline, subsequent windows need to be
scrolled down.  After either of these, follow-mode laboriously starts
determining where all its windows have to start and end.  There's nothing
in the display engine to help in this process.

> The advantage of using window-scroll-functions is that
> pre-redisplay-function is called much more frequently, in most cases
> follow-mode will need to do nothing at all.  You probably already have
> a logic for detecting when it should do something, but if you are
> invoked from window-scroll-functions, most if not all of that logic
> will be redundant.

> > Maybe it would be relatively simple to introduce new functionality.
> > Something like "scroll window so that window-end gets the given value".

> I'm not sure I understand how this could help follow-mode.  Please
> elaborate.

Currently when a middle or right hand window gets scrolled for any
reason, follow-mode has to determine how to scroll windows to the left of
it.  It does this by making a first guess at a window-start, does
set-window-start, then moves forward through the window to see how close
window-end is to where it needs to be.  If it's a line off, a different
starting position is chosen, and so on, until window-start gets correctly
placed.

If there were a function set-window-end, the display engine itself could
move back over the text lines to find window-start far more efficiently
and directly than follow-mode can.

-- 
Alan Mackenzie (Nuremberg, Germany).