Set gfx.canvas.azure.backends to skia. mach mochitest-plain content/canvas/test/test_2d.gradient.radial.outside3.html failed | pixel 1,1 is 255,0,0,255; expected 0,255,0,255 +/- 0 failed | pixel 50,1 is 255,0,0,255; expected 0,255,0,255 +/- 0 failed | pixel 98,1 is 255,0,0,255; expected 0,255,0,255 +/- 0 failed | pixel 1,25 is 255,0,0,255; expected 0,255,0,255 +/- 0 failed | pixel 50,25 is 255,0,0,255; expected 0,255,0,255 +/- 0 failed | pixel 98,25 is 255,0,0,255; expected 0,255,0,255 +/- 0 failed | pixel 1,48 is 255,0,0,255; expected 0,255,0,255 +/- 0 failed | pixel 50,48 is 255,0,0,255; expected 0,255,0,255 +/- 0 failed | pixel 98,48 is 255,0,0,255; expected 0,255,0,255 +/- 0

Related to different ways we handle this between Skia and non-Skia ( http://www.w3.org/TR/2dcontext/#dom-canvasgradient-addcolorstop ... If multiple stops are added at the same offset on a gradient, they must be placed in the order added, with the first one closest to the start of the gradient, and each subsequent one infinitesimally further along towards the end point (in effect causing all but the first and last stop added at each point to be ignored)... WebKit and Gecko differ on OS X - I'm not actually sure either of those is correct.

Works correctly on Linux. Both WebKit and Gecko are wrong on OS X.

Jeff and I discussed this a while back for Skia when we turned it on by default on Windows. We concluded at the time that it was pointless to get this test passing as no other browsers pass it and the specified behaviour is pretty odd.

In fact, it's already marked to skip on everything except CG: [test_2d.gradient.radial.outside3.html] skip-if = toolkit != 'cocoa'

This just removes a test from the suite, no try run on the change.