perf: Optimize mask widget rect collection to O(N)

[jeremiahseun]

Replaced List with Set in _collectMaskWidgetRects to improve performance from O(N^2) to O(N) by avoiding linear scan for contains checks. Included a benchmark test demonstrating ~17x speedup for a tree with ~5500 elements.

💡 Motivation and Context

The _collectMaskWidgetRects method previously used a List to track collected rectangles. It performed a !rectList.contains(element.rect) check for every element during the recursive traversal.

Since List.contains is an O(N) operation, and this check runs for every element in the tree, the overall complexity became O(N^2). For large widget trees (e.g., complex screens with thousands of elements), this caused significant performance degradation during session replay capture.

This change replaces the List with a Set for the accumulation phase. Set.contains and Set.add are O(1) operations, effectively bringing the total complexity down to O(N). The result is converted back to a List at the end to preserve the API signature.

🤔 Why did I do this / How did I discover this?

I was profiling our app's performance because we noticed intermittent stuttering (dropped frames) on our main 'Feed' screen, but only when Session Replay was enabled.

This screen is quite complex—it’s a long ListView with nested widgets, easily containing thousands of elements. I popped open the Flutter DevTools CPU Profiler to see what was eating up the frame time.

I expected the layout or painting to be the bottleneck, but I was surprised to see a significant chunk of time spent in ElementData.extractMaskWidgetRects. digging into the source code, I spotted the culprit immediately: a List.contains() check running inside a recursive loop.

In Computer Science terms, we were accidentally doing an O(N²) operation on every snapshot. For a small screen, you barely notice it. But on our feed with ~5,000 nodes, it was taking over 100ms—completely blocking the UI thread and causing visible lag.

💚 How did you test it?

I added a new benchmark test test/reproduction_benchmark.dart that generates a deep widget tree with ~5,500 ElementData nodes.

Benchmark Results (Avg per iteration):

Before: ~132 ms
After: ~7.6 ms
Improvement: ~17x faster
I also ran the full existing test suite (flutter test) to ensure no regressions in functionality.

📝 Checklist

• I reviewed the submitted code.
• I added tests to verify the changes.
• I updated the docs if needed.
• No breaking change or entry added to the changeling

Screenshots

[marandaneto]

@jeremiahseun makes sense, thanks
left a comment, otherwise mind running

posthog-flutter/Makefile

Line 16 in 2583d50

formatDart:

and then adding an item to the changelog? https://github.com/PostHog/posthog-flutter/blob/main/CHANGELOG.md#next (under next)

[jeremiahseun]

Hi @marandaneto, I have removed the test file and run the format command.