Created attachment 361220 [details] screenshot of bug on planet.ubuntu.com STR: 1. Visit http://planet.ubuntu.com/ 2. Zoom out (Ctrl -) ACTUAL RESULTS: Weird gray lines appear in various places (see screenshot) EXPECTED RESULTS: No weird gray lines. Broken on mozilla-central and 1.9.1 branch. Tested these builds: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.2a1pre) Gecko/20090208 Minefield/3.2a1pre Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.1b3pre) Gecko/20090203 Shiretoko/3.1b3pre But, working in Firefox 3.0.5: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:184.108.40.206) Gecko/2008121622 Ubuntu/8.10 (intrepid) Firefox/3.0.5 Requesting blocking status, as this is a regression in 3.1 with respect to 3.0.
FWIW, I get gray lines when zooming in, too. Almost all zoomlevels (aside from 1.0x) give me weird gray lines *somewhere*, though the lines' location varies depending on the zoomlevel.
Linux only? Seems to WFM on Mac. Definitely need a regression window and a reduced testcase.
Created attachment 361406 [details] reduced test case I see it on Linux & I'm willing to bet it's bug 468496 (i.e. yet another place where it matters that we aren't using CAIRO_EXTEND_PAD). Reduced self-contained test case attached. (One of) the problem images is embedded; it'll be at the far right of the window. Scale the page up and down and you should see gray or black lines at its edges.
Created attachment 361413 [details] testcase 1 On planet.ubuntu.com, it seems the gray lines appear on the edges of of the "chat bubble" PNG images, which are displayed via the CSS "background:" property. Here's a reduced testcase, using a mozilla PNG image to display the same problem. I see gray or black lines on the bottom & right edges of the 300x300 div in this testcase at these zoomlevels: Zoomed out 1x Zoomed out 7x (that's the max that it lets you zoom out, btw) Zoomed in 3x Zoomed in 13x (where "zoomed out/in [N]x" means to start at default zoomlevel, hold ctrl, and press - or + N times)
Yup, I can see lines at the bottom & right edges of the image on my computer too, but at different zoomlevels: zoomed out 1x (gray) zoomed out 3x (black) zoomed out 5x (black) (it won't let me zoom out any farther than this) zoomed in 3x, 13x (black) (*only if* zooming is done with the scroll wheel, not with Ctrl-+ -- it looks like Ctrl-+ takes bigger steps)
(In reply to comment #5) > zoomed in 3x, 13x (black) (*only if* zooming is done with the scroll wheel, Oh, sorry -- I was in fact using the scroll wheel when making my counts in comment 4. :) I assumed the steps were the same, but I guess they aren't.
Should be able to test the EXTEND_PAD hypothesis by hacking it in and seeing what happens, assuming EXTEND_PAD actually works on your system...
Created attachment 361449 [details] [diff] [review] demo that EXTEND_PAD solves the problem Indeed, the attached patch makes the problem disappear for me on both test cases and on planet.ubuntu.com. However, it is also visibly very slow.
Comment on attachment 361449 [details] [diff] [review] demo that EXTEND_PAD solves the problem clarifying title. ALso, please ignore the changes to reftests/border-image/reftest.list in the patch, I forgot I had some other changes in my tree.
I'd still like to know the regression range, but I suspect this is going to end up in the "X drivers suck" department.
I have an idea, which I shall describe in bug 468496 later. Need to finish up some other stuff first.
FWIW: I can actually reproduce the bug by viewing a PNG image directly (e.g. http://www.mozilla.org/images/front-moz-store.png ) and then simply zooming that. No need to use "background: url()" after all.
Related to bug 467832?
Same regression range as bug 487996 (probable fallout from bug 458487).
Duplicate of bug 468496 indicated by comment #8.