Share
Embed
Optimizing tidy: Practical Insights with the React Compiler
Talk at the HTW x OEV Meetup Berlin
31.01.2025
Niklas Kiefer
September 15, 2025
Transcript
28 slides
-
Optimizing tidy: Practical Insights with the React Compiler HTW x
OEV Meet-up Berlin -
tidy frontend projects “Classic” React SPA tidy-frontend tidy-agent-frontend tidy-frontend-shared React
Component Library “Classic” React SPA … -
What to measure A click scenario from our manual integration
testing suite • Without the compiler • With the compiler • Without manual memoization Metrics Google Lighthouse • First contentful paint (FCP) • Interaction to next paint (INP) • Total blocking time (TBT) Metrics React Profiler • Initial rendering time (IRT) More Metrics • Execution time (ET) Note: You could do the same in mobile mode -
Key findings: performance The react compiler doesn’t make the performance
worse in both projects. The execution time was almost the same in all scenarios. The results vary between large design system page and application scenario. Tip: Run lighthouse in incognito mode *with compiler compared to without compiler -
Key findings: performance In the agent-frontend click scenario, all performance
metrics are slightly better or the same with the compiler and with manual memoization. *with compiler compared to without compiler First Contentful Paint +-0% Total Blocking Time +-0% Initial Rendering Time -6,9% Interaction to next paint +-0% -
Key findings: performance In the agent-frontend click scenario, leaving out
manual memoization and trusting the compiler makes it faster - except initial rendering time *without manual memoization compared to with First Contentful Paint +-0% Total Blocking Time -37,5% Initial Rendering Time +5% Interaction to next paint -31,25% -
Key findings: performance In the design system page, all performance
metrics are best with the compiler and with manual memoization. *with compiler compared to without compiler First Contentful Paint -3,4% Total Blocking Time -33,3% Initial Rendering Time -0,66% Interaction to next paint +-0% -
Key findings: performance In the design system page, leaving out
manual memoization and trusting the compiler makes it slower. *without manual memoization compared to with First Contentful Paint +1,16% Total Blocking Time +27,78% Initial Rendering Time +3,54% Interaction to next paint +6,67% -
Key findings: dev experience • It works with both React
18 and 19 • Configuration as easy as three lines of code • Less memos, less cognitive load • “Pull requests [without memoization] took 31-46% longer to author”* *Source: https://react.dev/blog/2024/10/21/react-compiler-beta-release#react-compiler-at-meta -
Key findings: dev experience However: Eslint errors make it hard
to simply run it • Components got skipped because some rules were disabled • Blindly following the React ESLint rules breaks the application • Needs component refactoring to proper integrate compiler -
Key findings: dev experience However: Refactoring out all manual memoization
is a pain (GPT helps) • Starting in a new application would be easier to handle for sure • But: Leaving out memoization creates an abstraction layer -
Conclusion • The React Compiler can save you some unnecessary
re-rendering + improve the performance of your application • At least it does not make things worse -
Conclusion • The React Compiler can save you some unnecessary
re-rendering + improve the performance of your application • At least it does not make things worse However: • More components => less reliability • It’s not production ready (working group) • It won’t catch all unnecessary re-rendering • You can’t simply forget everything you know about useMemo, useCallback & … • Good architecture over trusting a black-box compiler -
Further reading • React Compiler Playground • How React Compiler
Performs on Real Code • React Compiler Docs • Discussions · GitHub - reactwg react-compiler • React Compiler Stableness Update