Attached is a test-case boiled out of OpenOffice.org and bolted onto the xlib cairo-demo. It attempt to render the string MoWoW and the string MWooW with a very large Liberation Serif font. And the screenshot shows the results. Unless I'm being very slow today (always possible) there is something terribly screwed up going on. cairo-1.6.4-1.fc9.i386 I see this both with a i810 driver and a nv driver.
Created attachment 306499 [details] screen shot
Created attachment 306500 [details] test case
I've replicated this bug with the r100 driver as well. So it appears to be a fairly generic cairo-xlib or X server bug. Note that the bug does not appear when rendering to a cairo image surface. Also, the bug is present with at least both of "Liberation Sans" and "Bitstream Vera Sans". -Carl
I've also added a new large-font test to cairo's test suite to exercise this bug. It also verifies that all backends other than cairo-xlib pass the test. -Carl
Isn't this the bug reported here: http://lists.cairographics.org/archives/cairo/2008-May/014033.html
(In reply to comment #5) > Isn't this the bug reported here: > http://lists.cairographics.org/archives/cairo/2008-May/014033.html Could be. I hadn't done any analysis of that one, so hadn't remembered it. No fix yet? Behdad, does Vlad's suggestion sound right? I'm still looking close to try to understand this one myself. Where is the glyph data getting shuffled for example? -Carl
Vlad's fix doesn't work: http://lists.cairographics.org/archives/cairo/2008-May/014044.html I'll fix this today, perhaps using what I suggest in the above URL.
I've pushed a couple fixes to cairo master now. Please test.
Yes latest git version works just right for me
That probably means we can close this bug now, with cairo 1.7.4 in rawhide.