524173 - Off-by-one bug in Cairo's rendering of some degenerate linear gradients

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Zack Weinberg (:zwol)]

Attachment: test case patch

While developing the tests for bug 513395, we discovered a bug in Cairo's rendering of some degenerate linear gradients -- that is, gradients with two or more stops at the same point.  I'm attaching a pair of new reftests which demonstrate the problem using <canvas>.  linear-gradient-1a.html passes on all platforms, but linear-gradient-1b.html passes only on Mac; on Windows and Linux, the boundary between the blue and the red boxes is one pixel below where it ought to be (that is, the blue box is 151px tall and the red box is 149px tall, when both should be 150px).

The problem seems to be a function of where along the gradient line the degenerate cluster appears.  1a has two stops at t=0, whereas 1b has two stops at t=1.  I've also been able to reproduce the problem with two stops at t=0.5 under some circumstances (not shown in this attachment).

Note that for tryserver purposes I marked 1b fails-if(MOZ_WIDGET_TOOLKIT!="cocoa") in canvas/reftest.list, so don't be surprised if you run it and see no failures.

Comment 1

[Jeff Muizelaar [:jrmuizel]]

I believe this is fixed by the pixman update associated with bug 640250

Comment 2

[Zack Weinberg (:zwol)]

Could you please include the test cases from this bug with your landing of the pixman update, then? (removing the fails-if, natch)

Comment 3

[Jeff Muizelaar [:jrmuizel]]

http://hg.mozilla.org/mozilla-central/rev/128c34e6c3b8

Comment 4

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Except the fails-if wasn't actually removed.  But, interestingly, the test passes on Lion.  This bug may be the same as bug 673333 given that it passes on Lion, and also the same as the two other failures so annotated in https://hg.mozilla.org/mozilla-central/rev/84117219ded0.

Comment 5

[Jeff Muizelaar [:jrmuizel]]

Bug 895135 seems to fix this.