pointer event performance

[forresto]

I'm noticing that moving nodes is pretty slow, and I think the timeline is saying it is the pointer event polyfill's fault.

Found another polyfill that could be worth trying: https://github.com/Rich-Harris/Points

Since that one intercepts events at the window level, it could maybe allow gmaps-like panning while dragged outside of the window.

[bergie]

I wouldn't start performance tuning this part quite yet, and in any case part of it might be the fact that we have multiple gesture recognizers running on the same element (one for each the-behavior).

Node moving will feel quite different when we start snapping them to grid, as well.

[bergie]

See also jquery-archive/PEP#92

[dfreedm]

Feel free to file a performance bug if you determine that this is in PointerEvents.
FWIW, I refactored some low hanging fruit recently (jquery-archive/PEP#102) and only just now pushed it to bower.

[bergie]

Regarding multiple gesture recognizers, things should be a lot more efficient when we have get the-grid/the-behavior#2

[forresto]

Seems a little better with the-behaviors, but still <30hz. Wondering why we're getting more than one mousemove event per frame, and why they are so expensive. Does it make sense that we are doing more mousemove calculations than screen updates?

[bergie]

One reason can be that we're currently setting a drag="x y" attribute to
the node being dragged.

This causes the dragChanged method to be called on the web component. But
since the attribute change notifications are asynchronous in Polymer, this
could be a reason why we miss moves.

[forresto]

Tried using https://github.com/Polymer/PointerGestures instead of the-behavior. Seemed like a big improvement. On mobile, panning went from 140ms to 60ms.

Test in new branch: https://github.com/the-grid/the-graph/tree/pointergestures

noflo

polymer

[bergie]

I think big part of this is that we still run the whole gesture data gathering operation after each pointermove, while we really are only interested in the coordinates moved (and gesture end). It might make sense to set up a shortcut earlier in the flow than we now do once gesture has been detected. We really don't need things like angle, distance, and speed after that.

[forresto]

fixing in pointergestures