Behavior seen in profile - touchmove dispatched - mouse_event_shim handleTouchMove() called - C++ hit-testing frames on stack (elementFromPoint() impl) - mousemove emitted - homescreen handleEvent() invoked The C++ code that looks like elementFromPoint() is taking ~10ms on a critical path.

We were worried about elementFromPoint() calls when we created this shim. The shim can be set to not call elementFromPoint() for apps that do not need it. We can investigate whether the homescreen can use that. Basically, if the event management code is okay with mouse events behaving like touch events and always being delivered to the original element (or if the app uses setCapture()) then the shim doesn't need to call elementFromPoint() to simulate mouse event behavior where the event target is always the element under the pointer. So this is probably optimizable. (And I think early on in the development of this shim, I successfully tested homescreen panning without the elementFromPoint() calls.) Vivien probably knows more about the homescreen than I do, but feel free to assign this to me if you want. Are there perf keywords or whiteboards things to be set for bugs like this?

ISTR that we had to force-off the elementFromPoint() logic because it was too high of an overhead on homescreen, which is why seeing this in the profile surprised me. I think some folks are using [FFOS-perf] but I'm not sure for what overall purpose.

Sorry If I haven't understand this bug but currently (in master) the homescreen is working with touch events without shim lib. There aren't elementFromPoint calls in panning code.

upps sorry for my English: I haven't understood

The profile was made against gecko-18 / nightly.

Cristian is right: the homescreen doesn't use the mouse shim anymore, at least not in the master or the v1.0.0 branch. So there must be something else going on here.

It looks like the nightly branch is no longer maintained. The shim was removed the same day this bug was reported. Its still there in nightly, but gone from master and v1.0.0: The problem went away with this commit which removes the shim: https://github.com/mozilla-b2g/gaia/commit/7a505331d9fbdd277e5d1c59a9f230200d494c93#apps/homescreen/index.html And here it is for the 1.0.0 branch: https://github.com/mozilla-b2g/gaia/commit/8d7803635ced3f4507ed82d0a72a3cc12f5d1585#apps/homescreen/index.html Resolving this as a dupe of the bug that removed the shim