XDamage extension over-reports window damage

[Jonathan Lennox]

On Wednesday, January 26 2005, "Keith Packard" wrote to "Jonathan Lennox, xorg at lists.freedesktop.org, Keith Packard" saying:

> 
> Around 15 o'clock on Jan 26, Jonathan Lennox wrote:
> 
> > Specifically, I've noticed two problems.  First of all, Xdamage reports
> > damage if a window *obscured* by the one I'm monitoring is drawn to.  E.g.,
> > if I have output going to an XTerm that is partially obscured by the window
> > I'm monitoring, the area of my window that covers the XTerm continually gets
> > damage reports.
> 
> I think this can probably be fixed; the limitation here is that damage 
> monitoring occurs only on the target holdin the physical pixels; when two 
> windows share the same physical target (e.g. the screen), then damage 
> monitoring on one will see damage done to the other; the trick is to clip 
> the damage so that it doesn't impact the underlying window.
> 
> It can't be fixed in all cases; the damage inside the server is reported 
> in rectangular units, so some (odd) drawing will still "leak" through to 
> the underlying window.

Can't you just clip the damage region to the window visible region, for
windows without backing store?  I've not been able to figure out enough
about X server internals to determine what structure describes the window
visible region, but the X server must have this information in order to draw
to the screen.

(On a related note, I'd love to be able as a client to retrieve and track
changes to the window visible region, so I know for what parts of the window
XGetImage is meaningful.  I can semi-track things with XVisibilityEvent and
XExposureEvent, but that doesn't suffice for the visibility region
shrinking.  The former part seems like a natural fit for Xfixes?)

> > In this case, with raw or delta damage report types, it seems like
> > damage reports should be sent just for the area of fvwm2's trim, but instead
> > I get a rectangle encompassing the entire window. 
> 
> Again, damage is internally recorded in rectangular units, so an operation 
> which ends up reporting only a single large rectangle can appear to damage 
> the internal window contents.
> 
> The guarantee Damage provides is that you will get notification whenever 
> the contents of the window changes; it doesn't guarantee that you will get 
> minimal notification.  Changes to the server which reduce overreporting 
> would be gratefully accepted.

Do you have any suggestions as to what's the best way to determine what's
causing the overreporting?  Will Xdamage and Xrecord timestamps correspond
for events triggered by the same client requests?

To motivate all of this, I'm trying to implement an X-based server for the
protocol defined in the Internet-Draft
<http://www.ietf.org/internet-drafts/draft-lennox-avt-app-sharing-00.txt>.
It's like VNC, but optionally window-based, so you can share individual
applications -- e.g., just a slideshow, as part of a multimedia conference
-- instead of your whole desktop.

-- 
Jonathan Lennox
lennox at cs.columbia.edu