AnnotationManager::getPointAnnotationsInBounds() should map IDs to screen coordinates

[1ec5]

For view-based annnotations (#1784 and #3276) to be performant, especially on Android, AnnotationManager::getPointAnnotationsInBounds() needs to supply the screen coordinates of each annotation whose ID it returns. That way we can call getPointAnnotationsInBounds() on each region change and avoid a potentially expensive series of calls to Map::pixelForLatLng() for each annotation in the returned list of annotations.

/cc @tobrun

[1ec5]

This API would also greatly improve performance in the OS X SDK when any annotation supplies a tooltip or cursor, since we need to update tooltip and cursor rects on each region change, just like we intend to do for view-based annotations.

[jfirebaugh]

Any implementation of this would likely just call Map::pixelForLatLng() for each annotation itself. Is there reason to believe this would be faster than having the SDK call it?

[1ec5]

This ticket has a big question mark on it because I filed it proacively before there were view-backed annotations to profile. @boundsj @tobrun, are you seeing this as a bottleneck currently? If not, we can close and move on.

[boundsj]

@1ec5 @jfirebaugh the implementation in #4801 only uses getPointAnnotationsInBounds to hit test visible annotation views at a touch point. If I select annotations over and over again for 30 seconds, getPointAnnotationsInBounds does not even register in the time profiler. For the same user activity, on the other end of the spectrum is SMCalloutView:presentCalloutFromRect:inLayer:ofView: constrainedToLayer:animated at about 30% and most of that is is a transparent image mask operation in a CALayer.

For testing if a view is in the viewport we currently use CGRectContainsRect and avoid getPointAnnotationsInBounds entirely. The expensive parts of this approach are currently:

• MGLAnnotationView:setCenter: 27%
• _mbglMap->getCameraOptions(padding): 17.2%
• _mbglMap->pixelForLatLng(latLng) (via MGLMapView:convertLatLng:toPointToView): 13.3%

[jfirebaugh]

It sounds like the current implementation of getPointAnnotationsInBounds is not a bottleneck, so closing here.