Bug 1389149 allows us to send hit-testing info to webrender. This bug is to have APZ utilize webrender to do hit-testing.

https://treeherder.mozilla.org/#/jobs?repo=try&revision=54c14278c903be02d4adbcac73a91169ca20dbb3 https://treeherder.mozilla.org/#/jobs?repo=try&revision=2893b95c61ddf1642b8d6b5ce87e179aad77b506

Comment on attachment 8929507 [details] Bug 1417519 - Hook up APZ to use the hit-testing API. https://reviewboard.mozilla.org/r/200788/#review206048

Comment on attachment 8929504 [details] Bug 1417519 - Log layers id as a hex value. https://reviewboard.mozilla.org/r/200782/#review206082

Comment on attachment 8929506 [details] Bug 1417519 - Add a mechanism for APZ to get the WebRenderAPI instance. https://reviewboard.mozilla.org/r/200786/#review206092

Comment on attachment 8929507 [details] Bug 1417519 - Hook up APZ to use the hit-testing API. https://reviewboard.mozilla.org/r/200788/#review206096 ::: gfx/layers/apz/src/APZCTreeManager.h:526 (Diff revision 1) > AsyncPanZoomController* GetTargetApzcForNode(HitTestingTreeNode* aNode); > AsyncPanZoomController* GetAPZCAtPoint(HitTestingTreeNode* aNode, > const ParentLayerPoint& aHitTestPoint, > HitTestResult* aOutHitResult, > HitTestingTreeNode** aOutScrollbarNode); > + already_AddRefed<AsyncPanZoomController> GetAPZCAtPointWR(const ScreenPoint& aPoint, It feels weird for GetAPZCAtPoint() to take a ParentLayerPoint but GetAPZCAtPointWR() to take a ScreenPoint. GetAPZCAtPoint() has a single call site, in GetTargetAPZC(), where the input ParentLayerPoint is obtained from a ScreenPoint via a cast with justification "ScreenIsParentLayerForRoot". I would prefer that we change GetAPZCAtPoint() to take its input as a ScreenPoint, to match GetAPZCAtPointWR(), and perform the cast inside GetAPZCAtPoint(). (Back when GetAPZCAtPoint() was recursive, we couldn't do this without introducing a helper, but now that it uses ForEachNode, we can.) ::: gfx/layers/apz/src/APZCTreeManager.cpp:2203 (Diff revision 1) > - RefPtr<AsyncPanZoomController> target = GetAPZCAtPoint(mRootNode, point, > - &hitResult, &scrollbarNode); > + target = GetAPZCAtPoint(mRootNode, point, &hitResult, &scrollbarNode); > + } > + > + if (gfxPrefs::WebRenderHitTest()) { > + HitTestResult wrHitResult = HitNothing; > + RefPtr<AsyncPanZoomController> wrTarget = GetAPZCAtPointWR(aPoint, wrHitResult); I assume doing both hit tests is a temporary state for testing? ::: gfx/layers/apz/src/APZCTreeManager.cpp:2230 (Diff revision 1) > return target.forget(); > } > > +already_AddRefed<AsyncPanZoomController> > +APZCTreeManager::GetAPZCAtPointWR(const ScreenPoint& aPoint, > + HitTestResult& aOutHitResult) For consistency with GetAPZCAtPoint(), use a pointer for the out-param. ::: gfx/layers/apz/src/APZCTreeManager.cpp:2257 (Diff revision 1) > + result = FindRootApzcForLayersId(layersId); > + MOZ_ASSERT(result); > + } > + > + aOutHitResult = HitLayer; > + if (hitInfo & gfx::CompositorHitTestInfo::eDispatchToContent) { Not strictly related to this patch, but can the enums HitTestResult and CompositorHitTestInfo be unified?

Comment on attachment 8929505 [details] Bug 1417519 - Don't allow things to get raw pointers to WebRenderAPI. https://reviewboard.mozilla.org/r/200784/#review206162

(In reply to Botond Ballo [:botond] from comment #10) > It feels weird for GetAPZCAtPoint() to take a ParentLayerPoint but > GetAPZCAtPointWR() to take a ScreenPoint. > > GetAPZCAtPoint() has a single call site, in GetTargetAPZC(), where the input > ParentLayerPoint is obtained from a ScreenPoint via a cast with > justification "ScreenIsParentLayerForRoot". > > I would prefer that we change GetAPZCAtPoint() to take its input as a > ScreenPoint, to match GetAPZCAtPointWR(), and perform the cast inside > GetAPZCAtPoint(). (Back when GetAPZCAtPoint() was recursive, we couldn't do > this without introducing a helper, but now that it uses ForEachNode, we can.) Good point, fixed. > ::: gfx/layers/apz/src/APZCTreeManager.cpp:2203 > > + RefPtr<AsyncPanZoomController> wrTarget = GetAPZCAtPointWR(aPoint, wrHitResult); > > I assume doing both hit tests is a temporary state for testing? Yup. > ::: gfx/layers/apz/src/APZCTreeManager.cpp:2230 > > + HitTestResult& aOutHitResult) > > For consistency with GetAPZCAtPoint(), use a pointer for the out-param. Done > ::: gfx/layers/apz/src/APZCTreeManager.cpp:2257 > > + aOutHitResult = HitLayer; > > + if (hitInfo & gfx::CompositorHitTestInfo::eDispatchToContent) { > > Not strictly related to this patch, but can the enums HitTestResult and > CompositorHitTestInfo be unified? I guess they could be. I'd go in the direction of removing HitTestResult and using CompositorHitTestInfo everywhere, since I'm going to be adding more stuff to CompositorHitTestInfo for scrollbars. I can file a follow-up for that.

Filed bug 1418541 for the unification.

Pushed by kgupta@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/2eaf8ed1338b Log layers id as a hex value. r=botond https://hg.mozilla.org/integration/autoland/rev/273eac46d15c Don't allow things to get raw pointers to WebRenderAPI. r=sotaro https://hg.mozilla.org/integration/autoland/rev/f9e128e8c8a6 Add a mechanism for APZ to get the WebRenderAPI instance. r=botond https://hg.mozilla.org/integration/autoland/rev/ff8e0f72dd9d Hook up APZ to use the hit-testing API. r=botond,mstange

https://hg.mozilla.org/mozilla-central/rev/2eaf8ed1338b https://hg.mozilla.org/mozilla-central/rev/273eac46d15c https://hg.mozilla.org/mozilla-central/rev/f9e128e8c8a6 https://hg.mozilla.org/mozilla-central/rev/ff8e0f72dd9d