fix: correct circular dependency check in facade elimination#9047
Conversation
How to use the Graphite Merge QueueAdd the label graphite: merge-when-ready to this PR to add it to the merge queue. You must have a Graphite account in order to use the merge queue. Sign up using this link. An organization admin has enabled the Graphite Merge Queue in this repository. Please do not merge from GitHub as this will restart CI on PRs being processed by the merge queue. This stack of pull requests is managed by Graphite. Learn more about stacking. |
✅ Deploy Preview for rolldown-rs ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
871500d to
32579f9
Compare
Merging this PR will not alter performance
Comparing Footnotes
|
32579f9 to
4aa5a26
Compare
28cb430 to
f6a94a0
Compare
Merge activity
|
## Summary This PR fixes an case of false-positive circular dependency check in facade chunk elimination that caused unnecessary facade chunks. The following is from a minimal reproduction case that is derived from vitejs/vite#21884. You may refer to the test case in this PR for details. `common.js` is a shared CJS dependency of both `ar_EG.js` and `en_US.js`. Due to `includeDependenciesRecursively`, `common.js` gets pulled into whichever group is processed first (`i18n-ar`). This makes `i18n-en-US` cross-chunk-depend on `i18n-ar` to access `require_common`: i18n-en-US → i18n-ar → runtime Since `i18n-ar` depends on the runtime, the old check — simulating merging runtime into `i18n-en-US` — falsely constructs a cycle: i18n-en-US → i18n-ar → (runtime = i18n-en-US) ← cycle here ## Fix Replaced `would_create_circular_dependency` with `is_reachable`. The old check is to simulate merging runtime into the target (i18n-en-US) instead of just checking if runtime has any reference to the target chunk. This causes a false positive on the case where there're multiple chunks that transitively or directly imports runtime. In this case, it's i18n-en-US that imports i18n-ar that imports runtime, which will be considered as cyclic deps. Changing to `is_reachable` fixes this as it only checks if runtime chunk has any imports to the target chunk. This makes more sense because we are not going to merge runtime into any of those chunks. ## Test plan - [x] New fixture `facade_elimination_with_shared_cjs_dep` — 4 chunks, no facades - [x] Existing snapshot updates from improved facade elimination ## Related issues Reduces facade chunks in Lobehub's case. Check out the report made by claude here: vitejs/vite#21884 (comment)
f6a94a0 to
f4aed8f
Compare
## [1.0.0-rc.16] - 2026-04-16 ### 🚀 Features - const enum cross-module inlining support (#8796) by @Dunqing - implement module tagging system for code splitting (#9045) by @hyf0 ### 🐛 Bug Fixes - rolldown_plugin_vite_manifest: handle duplicate chunk names for CSS entries (#9059) by @sapphi-red - improve error message for invalid return values in function options (#9125) by @shulaoda - await async export-star init wrappers (#9101) by @thezzisu - never panic during diagnostic emission (#9091) by @IWANABETHATGUY - include array rest pattern in binding_identifiers (#9112) by @IWANABETHATGUY - rolldown: set worker thread count with ROLLDOWN_WORKER_THREADS (#9086) by @fpotter - rolldown_plugin_lazy_compilation: escape request ID in proxy modules (#9102) by @h-a-n-a - treat namespace member access as side-effect-free (#9099) by @IWANABETHATGUY - relax overly conservative side-effect leak check in chunk optimizer (#9085) by @IWANABETHATGUY - runtime: release `cb` reference after `__commonJS` factory initialization (#9067) by @hyf0-agent - `@__NO_SIDE_EFFECTS__` wrapper should not remove dynamic imports (#9075) by @IWANABETHATGUY - rolldown_plugin_vite_import_glob: use POSIX path join/normalize for glob resolution (#9077) by @shulaoda - emit REQUIRE_TLA error when require() loads a module with top-level await (#9071) by @jaehafe - emit namespace declaration for empty modules in manual chunks (#8993) by @privatenumber - rolldown_plugin_vite_import_glob: keep common base on path segment boundary (#9070) by @shulaoda - prevent circular runtime helper imports during facade elimination (#8989) (#9057) by @IWANABETHATGUY - correct circular dependency check in facade elimination (#9047) by @h-a-n-a - docs: correct dead link in CodeSplittingGroup.tags JSDoc (#9051) by @hyf0 - emit DUPLICATE_SHEBANG warning when banner contains shebang (#9026) by @IWANABETHATGUY ### 🚜 Refactor - use semantic reference flags for member write detection (#9060) by @Dunqing - extract UsedSymbolRefs newtype wrapper (#9130) by @IWANABETHATGUY - dedupe await wrapping in export-star init emit (#9119) by @IWANABETHATGUY - calculate side-effect-free function symbols on demand (#9120) by @IWANABETHATGUY - extract duplicated top-level await handling into shared helper (#9087) by @IWANABETHATGUY - rolldown_plugin_vite_import_glob: use split_first for get_common_base (#9069) by @shulaoda - simplify ESM init deduplication with idiomatic insert check (#9044) by @IWANABETHATGUY ### 📚 Documentation - document runtime module placement strategy in code-splitting design (#9062) by @IWANABETHATGUY - clarify `options` hook behavior difference with Rollup in watch mode (#9053) by @sapphi-red - meta/design: introduce module tags (#9017) by @hyf0 ### ⚡ Performance - convert `generate_transitive_esm_init` to iterative (#9046) by @IWANABETHATGUY ### 🧪 Testing - merge strict/non_strict test variants using configVariants (#9089) by @IWANABETHATGUY ### ⚙️ Miscellaneous Tasks - disable Renovate auto-updates for oxc packages (#9129) by @IWANABETHATGUY - upgrade oxc@0.126.0 (#9127) by @Dunqing - deps: update napi to v3.8.5 (#9126) by @renovate[bot] - deps: update dependency @napi-rs/cli to v3.6.2 (#9123) by @renovate[bot] - move lazy-compilation design doc (#9117) by @h-a-n-a - deps: update dependency vite-plus to v0.1.18 (#9118) by @renovate[bot] - deps: update dependency vite-plus to v0.1.17 (#9113) by @renovate[bot] - deps: update oxc to v0.125.0 (#9094) by @renovate[bot] - deps: update dependency follow-redirects to v1.16.0 [security] (#9103) by @renovate[bot] - deps: update test262 submodule for tests (#9097) by @sapphi-red - deps: update crate-ci/typos action to v1.45.1 (#9096) by @renovate[bot] - deps: update rust crates (#9081) by @renovate[bot] - deps: update npm packages (#9080) by @renovate[bot] - remove outdated TODO in determine_module_exports_kind (#9072) by @jaehafe - rust/test: support `extendedTests: false` shorthand in test config (#9050) by @hyf0 - ci: extract shared infra-changes anchor in path filters (#9054) by @hyf0 - add docs build check to catch dead links in PRs (#9052) by @hyf0 ### ❤️ New Contributors * @thezzisu made their first contribution in [#9101](#9101) * @fpotter made their first contribution in [#9086](#9086) * @jaehafe made their first contribution in [#9071](#9071) * @privatenumber made their first contribution in [#8993](#8993) Co-authored-by: shulaoda <165626830+shulaoda@users.noreply.github.com>

Summary
This PR fixes an case of false-positive circular dependency check in facade chunk elimination that caused unnecessary facade chunks. The following is from a minimal reproduction case that is derived from vitejs/vite#21884. You may refer to the test case in this PR for details.
common.jsis a shared CJS dependency of bothar_EG.jsanden_US.js. Due toincludeDependenciesRecursively,common.jsgets pulled into whichever group is processed first(
i18n-ar). This makesi18n-en-UScross-chunk-depend oni18n-arto accessrequire_common:i18n-en-US → i18n-ar → runtime
Since
i18n-ardepends on the runtime, the old check — simulating merging runtime intoi18n-en-US—falsely constructs a cycle:
i18n-en-US → i18n-ar → (runtime = i18n-en-US) ← cycle here
Fix
Replaced
would_create_circular_dependencywithis_reachable. The old check is to simulate merging runtime into the target (i18n-en-US) instead of just checking if runtime has any reference to the target chunk. This causes a false positive on the case where there're multiple chunks that transitively or directly imports runtime. In this case, it's i18n-en-US that imports i18n-ar that imports runtime, which will be considered as cyclic deps.Changing to
is_reachablefixes this as it only checks if runtime chunk has any imports to the target chunk. This makes more sense because we are not going to merge runtime into any of those chunks.Test plan
facade_elimination_with_shared_cjs_dep— 4 chunks, no facadesRelated issues
Reduces facade chunks in Lobehub's case. Check out the report made by claude here: vitejs/vite#21884 (comment)