What would be a good performance metric for this optimization? I've thrown together some update buffering, but scrolling through revlog is still significantly slowed down by the git log fetch. Could probably use some mechanism for async operation cancellation.

Last time I benchmarked the drawing was slower than the rate of input events so the idea was to allow two(or more) event processings before redrawing again. This would significantly change the architecture though so it should be carefully measured whether this has enough upsides to carry itself