Re: making window managers faster

On Sat, 26 Oct 2002 16:27:05 +0300 (EET DST) Jouni Tulkki <[email protected]>
babbled:

> I have seen people complaining about slow window redraw when moving/resizing
> windows. I had noticed this problem myself and have found one way to reduce
> it. X server sends a lot of mouse events when moving the mouse. I haven't
> calculated how much but it could be something like 100 - 200 or more per
> second. This is way too much for window moving. 
> 
> The solution is to add a time check to the window manager that makes sure that
> windows are not moved more often then (for example) 50 times per second. Same
> applies to window resize, but because window resize is more expensive the
> maximum refresh rate should be lower, maybe about 25 1/s.

this isn't x's job. the wm should handle this. ie it should be a state machine
and only evaluate its when it goes idle. since X is a buffered protocol, what
shoudl be happening is

loop:
read_events()
move...
move...
move...
move...
move...
move...
end_read_events()
move_window()
wait_for_events()

so all the move events come in in 1 read, since they are buffered. the window
move only happens when all the events have been read and processed and the wm
has evaluated its current state. this way the system never fal;ls behind the
user on processing events and 2, will only update as fast as the system is able
to handle

(disclaimer: the above loop is an overly simplified version of the event loop.
it's missing other state processing, other event handling and synchronization at
idle handling)

now there are a LOT of window managers. do you want to add the check to all of
them? and what's the point? 10 mouse events come in. they result in 1 window
move (if the wm is written well). all events are buffered in a pipeline of
protocol requests/replies.

same applies to resizes. often applications and some widget sets aren't well
written and don't compress configurenotify events sent by the wm and only
re-evaluate their redraw on idle. apps should do things this way for optimum
performance. this is more a result of lack of knowledge of how x really works
and how to best work within its design to achieve optimal results.

-- 
--------------- Codito, ergo sum - "I code, therefore I am" --------------------
The Rasterman (Carsten Haitzler)    [email protected]
                                    [email protected]
Mobile Phone: +61 (0)413 451 899    Home Phone: 02 9698 8615