« Win32 backend renderer revisited | Main | Call for testing: cairo backend for win32 »
September 18, 2005
Call for testing: cairo-1.0.0
We would like to upgrade the version of cairo used by Firefox 1.5 from 0.5.0 to 1.0.0 in order to fix a number of missing geometry issues on OS-X. Preliminary reports are that it also speeds up SVG on OS-X between 10% and 75% depending in the test. We've landed this updated cairo on the cvs tag MOZILLA_1_8_SVG_BRANCH, and have a couple precompiled versions for testing:
Unfortunately we don't have a linux tarball for download, but if you're running current Red Hat rawhide then Firefox 1.5b1 is using the system cairo-1.0.0.
The win32 build pointed to here still uses GDI+ for SVG; we're mainly interested in <html:canvas> testing on that build.
Remember when testing a new version to close any Firefox windows when switching between browsers, otherwise you just get a new window from the currently running Firefox. We'd like to hear about an regressions between these versions and the official 1.5b1. Thanks.
Posted by tor at September 18, 2005 6:03 PM
Comments
> Unfortunately we don't have a linux tarball for download, but if you're running current Red Hat rawhide then Firefox 1.5b1 is using the system cairo-1.0.0.
Or Debian
Posted by: glandium at September 18, 2005 7:06 PM
The Mac os X build works on 10.3.9.
Posted by: Ludovic at September 18, 2005 7:40 PM
If no idea if they're relevant, but I went through the tests at http://annevankesteren.nl/test/html/canvas/
and found no differences in rendering between the win32 build and the current nightly.
Normal browsing seems fine so far too.
Posted by: kadigan at September 18, 2005 10:13 PM
Why not put the new version on trunk where it will get better testing, and the consequences aren't so important? Besides, surely it will be going onto the trunk at some point regardless of whether it gets branch approval...
Posted by: djc at September 18, 2005 11:45 PM
http://fuchsia-design.com/CanvaSVG/Tests/Tiger.xml
In this test, written in CanvaSVG, the first tiger is written in SVG, whereas the second is written in CanvaSVG, which translates markup to SVG through canvas (please correct me if I'm wrong).
The second lion scrolls far quicker than the SVG. It didn't before the update, if memory serves. Again, please correct me if I'm wrong.
Posted by: Foxtrot at September 18, 2005 11:45 PM
Judging from Foxtrot's previous example (http://fuchsia-design.com/CanvaSVG/Tests/Tiger.xml) it seems that there's a major difference in rendering between the latest trunk, and the branch.
I'm not sure if the same issue appears on the branch builds, but here's a screenshot:
http://img114.imageshack.us/my.php?image=difference4hc.jpg
Posted by: Caleb at September 19, 2005 5:57 AM
That "current trunk nightly" image looks like the test and replacement of libpixman for vc6 in the build process isn't working for some reason.
Posted by: tor at September 19, 2005 6:16 AM
When you refer to Cairo 1.0.0, am I correct in assuming that you refer to the latest 1.0.x release whenever this change is considered? 1.0.2 in particular looks to be released pretty soon, and according to the mailing lists it'll add most-of-the-way support for dashed curves as demonstrated in the following SVG file:
http://whereswalden.com/files/mozilla/testcases/dashed-curve.svg
Posted by: Jeff Walden at September 19, 2005 7:51 AM
> The Mac os X build works on 10.3.9.
And on 10.2.8.
Posted by: Torben at September 19, 2005 8:52 AM
tor, I used the Win32 build with Cairo enabled from. Please see http://www.codedread.com/. When you bring up some Firefox menus overtop of the "CodeDread" masthead and then collapse the menu, the portion of the text that was under the menu has its fill removed. This can be restored by forcing the browser to repaint (switching tabs, etc).
This happens with the "Go", "Bookmarks", "Tools", or "Help" - but note that "File", "Edit" and "View" DO NOT cause this problem. Must be something to do with detection of repaint areas...?
This does not happen in the GDI+ backend. Is it an already-raised bug?
Posted by: Jeff Schiller at September 19, 2005 5:01 PM
I tested the blobs on Windows/XP and Mac OS X 10.4. 6 tests with map coordinates SVG. Basically on Mac, Mozilla is a little slower but Windows Mozilla CRASHES or misbehaves on three of the six tests.
Tests and writeup at: http://www.the325project.org/SVG/Tests/MapTests/mapTests.html.
As each blob uses a different back end, is the problem in the back end windows port (some chance it's my Windows I suppose).
Posted by: conor at September 20, 2005 5:41 AM
I have just filed a bug report against Mozilla SVG which relates to a huge performance degradation that I have experienced on Windows (GDI). When tried on MacOSX using cairo 1.0.0, the performance degradation is not experienced. Does any one have the ability to readily test both GDI and Cairo? It would help to determine if the renderer is the issue.
The bug report is at https://bugzilla.mozilla.org/show_bug.cgi?id=309765
and contains an attachmnent to test:
https://bugzilla.mozilla.org/attachment.cgi?id=197186
If it turns out that this is a GDI issue, it might motivate an earlier switch to Cairo on Windows.
Posted by: JP Fiset at September 23, 2005 5:19 PM
Normal browsing seems fine so far too.
Posted by: Stella at December 8, 2005 6:04 PM