This is a genuine behavior change yeah? This means if a view doesn't have animation AND no view was changed, we used to not repaint and reuse the last element tree. Now we are going to redraw for that case

Is there anyway we could capture if this redraw is triggered by an animation?

It is a real behavior change, but I think in the opposite direction from what you're saying. The old condition was just a bad proxy for what we actually wanted to check.

Before, step 2 reused last_element when rendered_views was non-empty. That was meant to mean "invalidate ran, so the cache is trustworthy", but it's the wrong proxy: rendered_views holds cached child-view elements, and step 1 just popped the root out. So:

• If the root has no child views, the map is empty at that point even though invalidate ran. Step 2 gets skipped and we fall to step 3, a full fresh render of the root view on every draw, pointlessly.
• If the root has child views, the map stays non-empty across frames (cached child elements get re-inserted after every use), so the check couldn't actually detect "invalidate didn't run", which is the case its own comment said it was guarding. In practice every caller pairs invalidate + present so that path is unreachable, but the proxy was measuring map state, not the thing we cared about.

The new flag checks the intended condition directly: reuse the cache iff invalidate() ran this frame.

The reason this came up while building animation was that an animation repaint is a timer-triggered draw where nothing changed. The draw still runs invalidate(), but with an empty set, so nothing gets re-rendered and the root is never in rendered_views. Step 2 is the only thing that lets us repaint the cached tree; under the old emptiness check, a childless root would have re-run its whole render() at the spinner's cadence, which is bad lol.

and per the animation redraw capture, that should already be encoded. An animation-triggered draw is exactly a draw whose invalidation set is empty, so invalidate() re-renders nothing and present() reuses the cached tree. That reuse is the animation fast path. We could thread an explicit "draw reason" from the runtime, but it would just duplicate what the invalidation set already tells us.