Rolldown chunking in the wild
summary
Framer migrates to Rolldown from esbuild to get deterministic chunking and aggressive grouping, shipping in production August 2024 and replacing esbuild where needed. Benchmarks show a sweet spot at 20–25 initial chunks versus pathological cases like 276 tiny chunks, and Rolldown reduces median JavaScript by 36% median JS reduction which yields ~4% LCP improvement for 1–2MB sites, 41% for ≥2MB sites, and P90 LCP ~11% faster; very large builds use >10GB memory versus ~9.2GB for esbuild.
AI-written summary · about
chapters (16)
- 0:00Introduction to bundling
- 0:51Optimal chunking strategies
- 2:50Limitations of esbuild
- 4:52The case for bundling
- 5:10Network and compression impact
- 6:48Browser and cache performance
- 8:50Impact on slow devices
- 10:36Evaluating alternative bundlers
- 12:23Adopting Rolldown
- 13:39Rolldown versus esbuild
- 15:32Bundling at Framer
- 17:05Advanced chunking configuration
- 19:35Optimization techniques
- 21:06Analyzing and visualizing chunks
- 23:42Debugging tool improvements
- 25:03Results and conclusion
related talks
29mModern JavaScript Tooling: Exploring Rolldown, Oxc, and Beyond
22mOwning your Build-step – Owning your Code
23mDecoding Javascript: Loading Performance
16mRealtime Webpack - Pushing on-demand bundling to the limits
21mTree Shaking — Deploy Even Less Code to Production
34mBuild-time Optimizations in JavaScript Applications
12mReducing bundle size: Previewing a new Firebase for the web
45mBundler Bonanza