RenderAndroidView._setOffset lacks the hasSize guard its sibling gained in #83481 — uncatchable 'RenderBox was not laid out' from its frame callback #190833

Steps to reproduce

1. Embed an Android platform view in virtual-display hosting mode (AndroidView — in our case via mapbox_maps_flutter's MapWidget, default hosting mode).
2. Mount the widget while a route transition is in flight, so the platform view attaches before the new route's first layout completes.
3. RenderAndroidView._setOffset's self-rescheduling frame callback runs before the render box has a size.

Expected results

The offset update is skipped (or deferred) until the box is laid out — the same way _handleGlobalPointerEvent in the same class handles it.

Actual results

RenderBox was not laid out: RenderAndroidView#... — in production crash reporting this surfaces with no application frames (frame-callback context), 6 events / 3 users on our closed-test cohort within a week.

Analysis

RenderAndroidView._setOffset (packages/flutter/lib/src/rendering/platform_view.dart:200 at 3.44.6 stable / ee80f08) reschedules itself every frame guarded only by _isDisposed and attached:

void _setOffset() {
  SchedulerBinding.instance.addPostFrameCallback((_) async {
    if (!_isDisposed) {
      if (attached) {
        await _viewController.setOffset(localToGlobal(Offset.zero));
      }
      // Schedule a new post frame callback.
      _setOffset();
    }
  }, debugLabel: 'RenderAndroidView.setOffset');
}

localToGlobal requires completed layout — on an attached-but-unlaid-out box it throws RenderBox was not laid out. Its sibling in the same file, _handleGlobalPointerEvent (line 359), carries exactly the guard _setOffset lacks — added for the same crash class on the iOS side in #83481 (RenderUiKitView._handleGlobalPointerEvent is not checking for null size, closed by adding a hasSize check):

// Don't receive pointer events if not laid out. ...
if (!hasSize) {
  return;
}

So the fix that closed #83481 for pointer events never reached the offset path. Adding the same if (!hasSize) return; (letting the already-scheduled next-frame callback retry after layout) removes the crash without behavior change.

Notes

• Reproduces on stable; the unguarded code is present at current master.
• Workaround for map users: androidHostingMode other than virtual display (only VD produces RenderAndroidView), but that is a performance-profile change, not a fix.