Re: Recenter video

[Eli Zaretskii]

> From: Lennart Borgman <address@hidden>
> Date: Wed, 23 Jun 2010 02:41:08 +0200
> Cc: 
> 
> I can see the "jumping scrolling", but the bug here is not the same as
> the one I have tried to fix. Here I can see it enters the
> "clear_glyph" path in try_scrolling and I think this is what Eli have
> been working on.

If you still don't use the latest changes on the trunk, it's a small
wonder that you see recentering inside try_scrolling.

> I do not think Eli's changes cure the problem I have seen (but I am
> not sure) and my changes alone can't cure your problem (but it cured
> the problem I saw when scrolling window.c).

The problem is that your changes are incorrect: they violate the
redisplay protocol that doesn't allow using the current glyph matrix
if the last redisplay cycle was paused (a.k.a. interrupted) because
more input arrived.  The window_end_valid flag being nil means that
the redisplay was not finished, and in that case we cannot use the
try_scrolling method.  So your changes could work in some special
cases, but using them means we will certainly see redisplay bugs in
the future.

The only way to fix this case correctly is either to force redisplay
finish with its job, at least for a single window, before it pauses,
or improve the code that sets the window_end_valid flag so that it
does not give up too easily.

But to make such changes reliably I need to reproduce the problem on
my machine.  Until now, I can only reproduce it extremely rarely (like
once in the whole scroll of window.c, and only sometimes), even if I
load my CPU with 2 programs each of which eats up 100% of CPU
resources...