RenderNodeContext

whether the nodes created with this context measure the area their content actually covers as they record it. Without it a node whose content is not confined to its own bounds -- one that does not clip, or that casts a shadow -- reports a conservative area instead, and whatever records that node widens its cull rect to match. It costs a bounding hierarchy built over every recording.

whether the nodes created with this context record their content into an immutable picture once and replay it until the content changes, instead of re-recording it for every frame. It trades memory for recording time, so it pays off for trees that are drawn far more often than they change, and costs for trees that are re-recorded every frame. A node that casts a shadow, and every node drawing it, replays its content directly either way, because the shadow it draws depends on where it is replayed.

Snapshotting is a property of the whole tree rather than of a single node: a snapshot inlines the drawing of the nodes below it, so those nodes have to report their changes upwards for it to be dropped in time.