scuba change CL: cl/640025627

@goderbauer does it make sense to add an assert that makes sure nobody calls markNeedsLayout during idle / postFrameCallbacks? If markNeedsLayout was called during those two phases it's impossible (I think?) to recover before the next frame as the tree is finalized (and RenderBox.markNeedsLayout even destroys layout cache), and user actions such as hit-testing generally have asserts that ensure the render tree is clean: assert(!debugNeedsLayout).

I don't think we'll be able to just change the markNeedsLayout implementation since there are overrides (RenderBox), and adding the assert will probably mess up the case where we manually mark a render object dirty in tests. But it feels like a reasonable assert to add. Is there a way to somehow distinguish renderObject.markNeedsLayout() calls we manually write in tests?

Hm, I am not sure how we would practically achieve that given the existing (test) code. Also, do you think the assert is still correct for somebody who uses the rendering layer directly?

I don't know how people use the rendering layer directly, but I think

• "user inputs are handled during the idle phase" is the contract SchedulerBinding establishes
• our existing hitTest implementation does assume the tree is clean, and RenderBox.markNeedsLayout / RenderParagraph.markNeedsLayout destroys the current layout.

Given that mixin RendererBinding on BindingBase, ServicesBinding, SchedulerBinding, GestureBinding, SemanticsBinding, HitTestable it seems the best place to enforce that would be RenererBinding but doing that in the rendering library doesn't sound terribly wrong (as long as SchedulerBinding is being used).