SpecialPowers has a DOMWindowUtils property, which is a per-window object. attachSpecialPowersToWindow sometimes reuses the SpecialPowers of the .parent, meaning that we end up with SpecialPowers.DOMWindowUtils from the wrong window. Some DOMWindowUtils properties/methods use the window object that the DOMWindowUtils is related to, leading to weird behaviour/breakage. I remove the code to reuse SpecialPowers and pushed to try. The time to run tests didn't seem impacted at all, so I think we should stop reusing SpecialPowers objects.

Thanks, I don't think any of us really understand what we're doing with the DOM bits, what you see is just fumbling in the dark trying to get things working across the various Mochitest types.

Comment on attachment 8397090 [details] [diff] [review] v1 Review of attachment 8397090 [details] [diff] [review]: ----------------------------------------------------------------- r=me if this passes on try for all mochitest types on all platforms ::: testing/specialpowers/content/specialpowers.js @@ +103,5 @@ > try { > if ((aWindow !== null) && > (aWindow !== undefined) && > (aWindow.wrappedJSObject) && > + !(aWindow.wrappedJSObject.SpecialPowers)) { while I like this simplification, does this work in all cases?

Comment on attachment 8397090 [details] [diff] [review] v1 Review of attachment 8397090 [details] [diff] [review]: ----------------------------------------------------------------- ::: testing/specialpowers/content/specialpowers.js @@ +103,5 @@ > try { > if ((aWindow !== null) && > (aWindow !== undefined) && > (aWindow.wrappedJSObject) && > + !(aWindow.wrappedJSObject.SpecialPowers)) { This is just the existing code that used to be in the |else if|.

3 bugs backed out for failures in test_bug470212.html: https://tbpl.mozilla.org/?tree=Mozilla-Inbound&rev=8059895497c1&jobname=mochitest-9 https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?changeset=4fc31f234001

Relanded with a fix for the orange. Because we now use the SpecialPowers related to the window and not its parent, the test's mouse event were sent at slightly different coordinates (which seems to only affect the "B2G ICS Emulator Opt" build :-/). I adjusted the test so the coordinates are closer to the original values. https://hg.mozilla.org/integration/mozilla-inbound/rev/0fb101fc5976