Is there a reason why "split-by-experience" uses splitChunks.name as the default?

[rikisamurai]

Hi, I recently started a new project based on the React 19 template and noticed that rsbuild sets chunkSplit.strategy to "split-by-experience" by default. This strategy seems to assign a name to each cacheGroup.

However, according to the Webpack v4 to v5 migration guide To v5 from v4 | webpack, it's no longer recommended to use the name option.

From the SplitChunksPlugin documentation SplitChunksPlugin | webpack, I understand that specifying name can cause all modules matching the test condition to be bundled into a single chunk. This might lead to larger bundle sizes and even force some async resources into the initial chunk.

Could I ask the rationale behind using this configuration as the default?

    splitChunks: {
      chunks: 'all',
      cacheGroups: {
        react: {
          name: 'lib-react',
          test: /node_modules[\\/](?:react|react-dom|scheduler)[\\/]/,
          priority: 0
        },
        router: {
          name: 'lib-router',
          test: /node_modules[\\/](?:react-router|react-router-dom|history|@remix-run[\\/]router)[\\/]/,
          priority: 0
        },
        'lib-axios': {
          test: /node_modules[\\/]axios(-.+)?[\\/]/,
          priority: 0,
          name: 'lib-axios'
        }
      }
    },

[chenjiahan]

Sorry for the late reply. Rsbuild uses fixed chunk names like lib-axios and lib-polyfill for predictable caching - the same libraries always generate the same file names across builds, enabling reliable long-term browser caching.

[rikisamurai]

Thank you very much for your response!!!
Using splitChunks.name can cause a situation where entryA only uses react-router and entryB only uses remix, but this leads to both pages increasing unnecessary bundle size. Moreover, if dynamic import is used, it may cause the module to be directly included in the initial chunk.
Using splitChunks.idHint here can also split chunks according to the rules, but it will not cause redundant bundle size. The main issue this brings is that in production mode there won’t be semantically clear names because optimization.chunkIds is set to deterministic.

[chenjiahan]

Thank you for your suggestions. ❤️ The current implementation does indeed have the limitations you mentioned, which is why we only include a small number of commonly used packages in the "split-by-experience" strategy. And We have plans to re-design the current performance.chunkSplit configuration options and values, and I will take idHint into consideration. This will likely be included in Rsbuild 2.0.