Bug 31931

Created attachment 40581 [details]
Screenshot without error

cairo builded with '--enable-xcb --enable-gl --enable-qt'

Created attachment 40582 [details]
Screenshot with error

cairo builded with '--enable-xlib-xcb --enable-xcb --enable-gl --enable-qt'

No idea what I might be doing wrong, but I can't reproduce this.

pixman: pixman-0.21.2-4-g4b5b5a2 (same version as reporter)
cairo: 1.10.0-187-g493aaf0 (also the same version)
libxcb-1.6, xcb-proto-1.6 both from current debian testing

cairo was built with ./autogen.sh PKG_CONFIG_PATH=/home/psychon/projects/pixman/ --enable-xcb --disable-ps --disable-pdf --disable-svg --enable-xlib-xcb

Tested via this command, result looks good to me (and according to lsof, this cairo is being used):
$ LD_PRELOAD=/tmp/cairo/src/.libs/libcairo.so /usr/lib/iceweasel/firefox-bin -ProfileManager -no-remote

Tested with libX11-1.3.3 & libxcb-1.6, the same black rectangles :/

Can xserver cause it?

Sure, which graphics driver do you use? Could you test via e.g. Xnest?

>This bug also appears in firefox via ssh -X, so I assume that it's not video driver bug.

xf86-video-ati/libdrm/mesa from git.

Can't reproduce with latest git.

Looks like lat time firefox havenn't built clean and was linked to bundled cairo copy.


I can reproduce this issue with firefox-4.0[beta], firefox-3.6 works ok (sorry that I haven't mentioned Fx version before and haven't tested 3.6). 

Reproducible with xserver-1.9.3/1.94[rc], libX11-1.3/1.4, libxcb-1.6/1.7 for me.

Building Fx with system cairo is broken ATM, you may need patches from https://bugzilla.mozilla.org/show_bug.cgi?id=623797

Still the same with f1d313e042af89b2f5f5d09d3eb1703d0517ecd7

Still the same with ed78510.

Still the same with b377380.

Still the same with cairo-git HEAD 6936a2b

I installed the iceweasel 5.0-1 package from debian experimental which is built with system cairo. I can reproduce the bug now (I used LD_PRELOAD to make it use my xlib-xcb cairo built).

With lots of try-and-error I noticed that sticking a cairo_surface_flush() into _cairo_xcb_surface_mark_dirty() makes the problem go away, but _cairo_xcb_surface_flush() doesn't help. In other words, _cairo_surface_detach_snapshots() is needed. Meh?

Digging a little deeper, I found out that the _cairo_surface_attach_snapshot() created in _cairo_xcb_surface_acquire_source_image() causes this problem. Dunno where to go from here.

Ah, finally. Could you check if the following commit fixes the issue for you? It's already pushed to cairo.git.
Also, sorry that this took me three bugspam-comments.

commit 7e1a9f1db0471f238a954a1fd248695a2a6920d7
Author: Uli Schlachter <psychon@znc.in>
Date:   Fri Jul 1 21:48:31 2011 +0200

    xlib-xcb: Fix cairo_surface_flush()
    
    This function called directly into the xcb's surface flush function. This means
    that snapshots for that surface weren't detached since that's normally done in
    cairo_surface_flush() before calling into the backend.
    
    Fix this by using surface_flush() instead of calling into the backend directly.
    
    Fixes: https://bugs.freedesktop.org/show_bug.cgi?id=31931
    
    Signed-off-by: Uli Schlachter <psychon@znc.in>

Fixed, thanks.

That also fixes the black on black I got.

The other xlib-xcb render corruption still exists, though.

@James Cloos: What are you talking about? I only see a single render corruption mentioned in this bug.

(In reply to comment #18)
> @James Cloos: What are you talking about? I only see a single render corruption
> mentioned in this bug.

Please check https://bugs.freedesktop.org/show_bug.cgi?id=38904
I know that report isn't extremely useful, but maybe you'll be able to reproduce this bug too.

> What are you talking about? I only see a single render corruption
> mentioned in this bug.

I may have only mentioned it in the thread on the list; I can’t remember
right now.  But the bug ojab referred to in commment #19, bug #38904, is
on point.  This one can be closed and further dialog moved there.