perf(replay): optimize screenshot mode — 93% faster (14.7x)

[marandaneto]

💡 Motivation and Context

Relates to #321

Session replay's screenshot capture was consuming 3,319µs per snapshot on the main thread — nearly 20% of a 60fps frame budget. Users on older devices reported frame drops and jank caused by the view hierarchy traversal and image rendering.

This PR reduces main-thread time to ~230µs (93.2% reduction, 14.7× faster), bringing frame budget impact down to 1.4% at 60fps — negligible.

Performance Results

Metric	Before	After	Improvement
Total	3,319µs	~230µs	−93.2%
findMaskableWidgets	1,760µs	~175µs	−90.0%
toImage (drawHierarchy)	653µs	~50µs	−92.3%
Frame budget (60fps)	19.9%	1.4%	—

Key Optimizations

Main-thread performance:

1. Plain UIView + UIWindow fast path — Skip all type casts for container views that can never contain sensitive content. Biggest win (~66%).
2. Custom CGContext renderer (PostHogGraphicsImageRenderer) — Based on Sentry's approach. Bypasses UIGraphicsImageRenderer overhead by directly allocating a CGContext with malloc. Uses native screen scale which avoids internal rescaling in drawHierarchy.
3. Config caching (ReplayMaskConfig) — Cache maskAllTextInputs, maskAllImages, maskAllSandboxedViews once per snapshot instead of traversing weak ref + optional chain per view.
4. If-else type chains — Skip unnecessary type casts after a match in both findMaskableWidgets and toWireframe.
5. Touch phase filtering — Skip location(in:) for non-began/ended phases (avoids coordinate conversion during drag/scroll).
6. Move lock + meta event off main thread — windowViewsLock access and meta event dict creation moved to dispatch queue.

Off-main-thread + memory:

• Skip maskImage re-render when maskableWidgets is empty (saves ~1.3MB)
• Release UIImage early in toDict() before base64 encoding (reduces peak memory)
• Pre-size dictionaries in toDict() (minimumCapacity: 16)
• Hex lookup table in toRGBString() — replaces String(format:) with direct char array
• malloc instead of calloc for renderer buffer (skip zero-init of ~12MB at 3x scale)

Files Changed

• PostHogReplayIntegration.swift — Fast paths, config caching, if-else chains, touch handler, async refactor
• PostHogGraphicsImageRenderer.swift — NEW custom CGContext renderer
• UIView+Util.swift — Custom renderer integration, isNoCapture() optimization
• CGColor+Util.swift — Hex lookup table for color string formatting
• RRWireframe.swift — maskImage skip, early image release, dict pre-sizing
• RRStyle.swift — Dict pre-sizing

💚 How did you test it?

• Benchmark test on iOS Simulator (iPhone 17 Pro) with 200-view synthetic hierarchy (484 actual views including UIKit internals)
• 80 experiments across 9 optimization sessions with automated measurement
• make buildIOS and swift build both pass
• make format clean (SwiftLint + SwiftFormat)

📝 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 changelog.

[marandaneto]

@ioannisj autoresearch result, we'd need to test everything but the results are good

[github-actions]

posthog-ios Compliance Report

Date: 2026-03-25 16:59:56 UTC
Duration: 522ms

✅ All Tests Passed!

0/0 tests passed

---

[ioannisj]

@ioannisj autoresearch result, we'd need to test everything but the results are good

Oh wow! Will take a look and test. Looks promising from PR description