Reflections on a Year of Compose
summary
Pinterest spends 18 months integrating Jetpack Compose, builds experiments that force AOT via a tiny red pixel and later full rewrites, and finds a persistent performance regression tied only partly to the jit hit that yields at best 3–5% recovery after fixes. Stability issues prove manageable and workarounds exist, but baseline profile tooling and months of perf hunting don’t pay off, so the team leaves Compose in greenfield use only and schedules quarterly rechecks with baseline profile rules.
AI-written summary · about
chapters (8)
- 0:00Start
- 20:13Baseline Profile
- 28:27We care about performance. A lot.
- 30:38In absence of external validation, we second guessed ourselves.
- 31:33We prioritized visual polish over learning quickly.
- 32:20We got distracted by external advice ill suited to our specific situation.
- 34:32We compared something easy to measure to something hard to measure.
- 37:48We didn't realize a key part of adopting for developer velocity gains
related talks
01h 08mAdopting Jetpack Compose @ Scale
46mOptimizing Compose: Issues, Experience, Tactics, Strategy
21mPerformance best practices for Jetpack Compose
36mOptimizing Render Performance of Jetpack Compose
29mCompose
21mMore performance tips for Jetpack Compose
28mCompose at Scale: Stop Unnecessary Recomposition
37mAdopting Jetpack Compose Safely