fix(linker): resolve race condition in side effects computation for export-star#7728
Merged
IWANABETHATGUY merged 3 commits intorolldown:mainfrom Jan 1, 2026
Merged
fix(linker): resolve race condition in side effects computation for export-star#7728IWANABETHATGUY merged 3 commits intorolldown:mainfrom
IWANABETHATGUY merged 3 commits intorolldown:mainfrom
Conversation
…ce_needed_symbols During parallel processing in reference_needed_symbols, one thread may write to a module's side_effects while another thread reads from it. This adds atomic tracking to detect and panic on this race condition. The race occurs between: - Writes at lines 151-154, 168-172, 229-233 (setting importer.side_effects) - Reads at lines 187 and 217-218 (reading importee.side_effects) When a module both has `export * from 'wrapped-module'` (triggering a write) AND is imported by another module (triggering a read), the race can cause non-deterministic tree-shaking decisions.
…snapshot Fixes rolldown#7646 The race condition occurred because during parallel processing: - One thread reads `importee.side_effects` to determine statement side effects - Another thread writes to `importer.side_effects` via write_volatile When a module both has `export * from 'wrapped-module'` (making it a write target) AND is imported by another module (making it a read target), non-deterministic results occur. The fix: 1. Pre-compute which modules need side_effects upgraded to true before parallel section 2. Create a snapshot of side_effects values (including upgrades) for safe parallel reads 3. Defer side_effects writes to after the parallel section This ensures all parallel reads see a consistent view of side_effects values.
…de_effects phase Move the side effects computation for `export * from 'wrapped-module'` patterns from reference_needed_symbols to determine_side_effects. This: - Enables proper transitive side effects propagation through module chains - Eliminates the need for pre-computation and snapshot logic in parallel processing - Removes race condition concerns by computing all side effects before parallel section - Simplifies reference_needed_symbols by removing ~45 lines of deferred update code
Contributor
There was a problem hiding this comment.
Pull request overview
This PR fixes a race condition in side effects computation for export-star patterns that caused non-deterministic CJS bundle output. The fix moves side effects computation from parallel processing during reference_needed_symbols to the sequential determine_side_effects phase, eliminating the race where one thread could read stale side effects data while another thread was updating it.
- Adds export-star handling logic to
determine_side_effectsto mark modules withexport * from 'wrapped-module'patterns as having side effects - Removes unsafe parallel writes to
importer_side_effectsfromreference_needed_symbols - Ensures proper initialization order by computing side effects before parallel symbol referencing
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
crates/rolldown/src/stages/link_stage/tree_shaking/determine_side_effects.rs |
Added logic to detect export-star patterns with wrapped modules and mark them as having side effects based on wrap kind and dynamic exports |
crates/rolldown/src/stages/link_stage/reference_needed_symbols.rs |
Removed unsafe parallel writes to side effects field and related volatile pointer operations, now only reads pre-computed side effects |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
IWANABETHATGUY
approved these changes
Jan 1, 2026
This was referenced Jan 7, 2026
shulaoda
added a commit
that referenced
this pull request
Jan 7, 2026
## [1.0.0-beta.59] - 2026-01-07 ⚡ Inline Dynamic Imports for Statically Imported Modules - When a module is already statically imported, dynamic imports to that same module are now inlined instead of creating a separate chunk ### 🚀 Features - plugin_timings: add 3s threshold and doc link to warning message (#7741) by @shulaoda - improve treeshaking logic to handle empty parameter list in dynamic import .then() callbacks (#7781) by @Copilot - dev/lazy: don't include already executed modules (#7745) by @hyf0 - dev/lazy: support dynamic `import(..)` (#7726) by @hyf0 - inline dynamic imports that imports statically imported modules (#7742) by @IWANABETHATGUY - option: add experimental option to control chunk optimization (#7738) by @IWANABETHATGUY ### 🐛 Bug Fixes - inline dynamic entry to user defined entry with esm wrap kind (#7783) by @IWANABETHATGUY - use canonical namespace reference for property access (#7777) by @IWANABETHATGUY - dynamic entry merged into common chunk with cjs and esm wrap kind (#7771) by @IWANABETHATGUY - tla: should not await non-tla-related modules (#7768) by @hyf0 - dynamic entry captured by common chunk with CJS format (#7757) by @IWANABETHATGUY - module_loader: mark emitted chunks as user-defined entry when already loaded (#7765) by @shulaoda - normalize preserveModulesRoot path (#7737) by @IWANABETHATGUY - linker: resolve race condition in side effects computation for export-star (#7728) by @camc314 ### 🚜 Refactor - plugin_timings: filter out plugins with duration < 1s from timing warnings (#7785) by @shulaoda - module_loader: remove unnecessary collect before extend (#7769) by @shulaoda - rename _id suffixes to _idx for oxc_index types (#7767) by @IWANABETHATGUY - remove duplicate `preserve_entry_signatures` from `AddEntryModuleMsg` (#7762) by @shulaoda - module_loader: pass `user_defined_entries` by reference (#7756) by @shulaoda - dev/lazy: get proxy entry's `ResolvedId` correctly (#7746) by @hyf0 - simplify try_rewrite_import_expression control flow (#7753) by @IWANABETHATGUY - module_loader: remove unnecessary dynamic import handling for runtime module (#7754) by @shulaoda - inline __toDynamicImportESM (#7747) by @IWANABETHATGUY - use From impl for ModuleLoaderOutput conversion (#7732) by @shulaoda - remove duplicate fields from `ModuleLoader` (#7731) by @shulaoda - tweak `resolve_user_defined_entries` (#7727) by @shulaoda ### 📚 Documentation - add rolldown-string reference to native MagicString compatibility section (#7778) by @Copilot - improve comments for export star side effects handling (#7730) by @IWANABETHATGUY ### 🧪 Testing - use assertion instead of console.log for some testcase (#7744) by @IWANABETHATGUY ### ⚙️ Miscellaneous Tasks - tweak some `output.dynamicImportInCjs` related rollup test results (#7776) by @sapphi-red - mark esbuild/dce/dce_of_symbol_ctor_call as passed (#7775) by @sapphi-red - deps: update oxc apps (#7772) by @renovate[bot] - vite-tests: allow running on PRs with `test: vite-tests` label (#7770) by @shulaoda - deps: update oxc apps (#7760) by @renovate[bot] - deps: update rollup submodule for tests to v4.55.1 (#7763) by @sapphi-red - deps: update test262 submodule for tests (#7764) by @sapphi-red - deps: update oxc to v0.107.0 (#7758) by @camc314 - deps: update taiki-e/install-action action to v2.65.13 (#7751) by @renovate[bot] - deps: update rust crates (#7750) by @renovate[bot] - deps: update npm packages (#7749) by @renovate[bot] - deps: update github-actions (#7748) by @renovate[bot] - deps: update dependency oxlint-tsgolint to v0.10.1 (#7729) by @renovate[bot] - deps: update crate-ci/typos action to v1.41.0 (#7725) by @renovate[bot] Co-authored-by: shulaoda <[email protected]>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
NOTE: I used AI for this, as I'm not familiar with the rolldown codebase. So feel free to close if this is the wrong fix. (That said, I manually validate each of the commits, including making sure that this fixes the issue, so I think it's OK.)
I've split this change into three commits
Firstly, I got AI to identify the cause of the issue, I cloned the repro as a submodule and got claude to investigate it. Eventually, this lead to identifying a race condition in reference_needed_symbols. This function runs in parallel across all modules and was both reading importee.side_effects and writing to importer.side_effects simultaneously. When module A reads module B's side effects while another thread is upgrading B's side effects, A could see stale data.
Commit 1: Validating the race condition
After working out what the problem was, I added debug assertions to validate that the race condition was real and reproducible. This involved adding assertions in the relevant parts of the code to check for inconsistent states.
I then ran this against the reproduction repo, and was able to consistently trigger the panics that I added.
Commit 2: First fix attempt
Given that the cause (race condition) was identified, I made an initial attempt to fix it (in that same file), the approach being:
truebefore entering the parallel sectionThis eliminates the race condition.
However, this doensn't really correctly handle `transitive side effects e.g:
A.js:export * from 'B'// B has WrapKind::NoneB.js:export * from 'C'// C is WrapKind::CjsC.js:module.exports = {}B gets upgraded because of C, but A's snapshot of B was captured before B was upgraded. The fix worked for direct dependencies but could miss chains.
Commit 3: Better solution
Rather than adding complexity to handle transitive cases in the snapshot approach (commit 2), I refactored for a cleaner architecture.
During the link stage, there is a dedicated
determine_side_effects. This runs beforereference_needed_symbold, already handling transitive effects via import.I've got no idea how to write a test for this given how flaky it is, so suggestions are appreciated.
fixes #7646