EXA glyph corruption
dsd at laptop.org
Thu Aug 28 12:01:15 PDT 2008
I'm working on OLPC's upcoming software release, including an update
from Fedora 7 to Fedora 9. Fedora 9 ships xserver 22.214.171.1246.
Unfortunately, we encounter a rendering bug with one of our applications
(etoys). Certain components that move around the screen render badly
(leave trails, etc) when interacting with the mouse cursor.
An etoys developer narrowed this down to a small program to reproduce
It moves a green square around the screen. If you let the green square
move onto the mouse cursor, you get graphical corruption. See
http://dev.laptop.org/~dsd/20080822/xshm-bug.png for a screenshot - the
mouse cursor didn't get included in the screenshot, but it is in the
white area inside the green square. The square should be solid green
even with the mouse cursor on top.
I noticed that Fedora recently added an "exa master upgrade" patch to
their xserver test builds - backporting the most recent EXA code from
master. Decided to give this a try in hope that it solves the bug.
Here is the patch:
The good news is that it solves the bug, here is a new screenshot:
The mouse cursor is on top of the green square in that screenshot with
no rendering glitches.
The bad news is that text rendering is now quite badly broken:
with the older X server build, that same screen looks like:
I backported the updated exa code from master again, this didn't help.
Jordan Crouse (geode driver author) also had a look at the text
corruption. He thinks that EXA might be passing him the wrong glyph
< CosmicPenguin> It looked to me based on whatever tag I was compiling
against the other day that the mask I was getting was
< CosmicPenguin> I only did enough work to verify that the driver was
< CosmicPenguin> keep in mind that I may or may not be missing
fontconfig or other support software for fully
effective font fun
< CosmicPenguin> I am not sure at what point from file to screen the
was being mismanaged - all I know is that I was doing
whatever I was told
< CosmicPenguin> and thats not half bad
We know that whatever caused the regression is some part of this patch:
More information about the xorg