We need to setup a case where the visual viewport is smaller than the layout viewport. Faking a pinch zoom through touch events is imprecise, whereas calling windowUtils.setResolutionAndScaleTo(value > 1) to simulate the equivalent effect of a pinch-zoom on the page led to the subsequent scroll attempt failing quite often even after calling waitForApzFlushedRepaints(). So instead, we set up a very wide layout viewport through the meta viewport tag, which on Desktop means additionally enabling the dom.meta-viewport.enabled pref.

(In reply to Jan Henning [:JanH] from comment #1) > https://treeherder.mozilla.org/#/ > jobs?repo=try&revision=e3d1d3f851c0bcdb8122b6b84a290c1d42d79b37 Great, the original push is restricted only to that particular test paths, but the retriggers totally ignore that and run something completely different... So the test isn't fundamentally broken anywhere, but that Try push is totally useless with regards to possible intermittent failures.

Pushed by mozilla@buttercookie.de: https://hg.mozilla.org/integration/autoland/rev/0355f47af6ce Test that resetting WindowUtils.isFirstPaint doesn't clobber APZ scroll position. r=botond

Pushed by mozilla@buttercookie.de: https://hg.mozilla.org/integration/autoland/rev/81378c45a359 Test that resetting WindowUtils.isFirstPaint doesn't clobber APZ scroll position. r=botond