fix(hash): keep chunk file names stable when an unrelated entry is added by hyf0 · Pull Request #9444 · rolldown/rolldown

hyf0 · 2026-05-18T13:15:53Z

Summary

Fixes #9339: adding or removing an isolated entry currently changes the file names of every other chunk in the bundle, even when those chunks are byte-identical between builds. The reported example uses entryFileNames / chunkFileNames with codeSplitting and observes the rolldown runtime chunk's hash flip across builds that should produce identical output.

Root cause

Hash placeholders (!~{000}~, !~{001}~, …) are assigned in rendering order, so their numeric indices shift whenever the chunk graph changes. Those indices leak into the hash from two places in finalize_chunks.rs:

Standalone content hash: the chunk's raw content includes placeholders pointing at sibling chunks (e.g. import "./shared-!~{001}~.js"), so when a sibling's index moves, this chunk's content bytes — and therefore its content hash — change.
Final hash: preliminary_filename (which contains the chunk's own placeholder) is mixed in directly, so when the chunk's own index moves, its final hash moves too.

Fix

Mirrors Rollup's generateFinalHashes:

Normalize generated placeholders before hashing content. A new visit_with_placeholders_defaulted helper in rolldown_utils::hash_placeholder walks the chunk content and streams it into Xxh3 byte-slice by byte-slice, replacing each rolldown-generated !~{xxx}~ with a zero-filled placeholder of the same shape. Placeholders rolldown didn't generate (literals in user source code that just happen to match the shape) are emitted verbatim so changes to them still flow into the hash — same placeholders.has(placeholder) check Rollup performs in replacePlaceholdersWithDefaultAndGetContainedPlaceholders. Streaming avoids materializing a chunk-sized normalized String per chunk; for a typical bundle with ~MB chunks containing cross-chunk imports, that's ~bundle-size's worth of throwaway allocations per build avoided.
Drop preliminary_filename from the final hash, deconflict at the file-name layer instead. Two chunks whose resolved file names would collide (case-insensitively, for HFS+/NTFS safety) are now disambiguated by rehashing the colliding chunk's hash until its file name is unique — the same do { … } while (collision) loop Rollup uses.

A design doc covering the three invariants (stability across builds, sensitivity to real content, uniqueness within a build), the two-phase parallel pipeline + sequential deconflict pass, and known limitations is added at meta/design/chunk-hash.md.

Why a separate PR, not a follow-up to #9356

#9356 fixed the content-hash side of the bug but had two issues that made it not the right shape to land as-is:

It kept the buggy preliminary_filename.hash(&mut hasher) line under an if options.sourcemap_debug_ids branch, so users with sourcemapDebugIds: true would still hit the original non-determinism.
It removed preliminary_filename from the final hash in the default mode without adding any collision handling, regressing the case Rollup explicitly covers in its deconflict-hashes test (two byte-identical entries + entryFileNames: '[hash].js' → Rollup produces two distinct file names via rehash; fix: stabilize chunk content hashes #9356 would silently produce two assets with the same name).

This PR takes the same overall direction (normalize placeholders in content) but lands the rehash loop alongside it so both invariants hold simultaneously, and applies uniformly regardless of sourcemapDebugIds. Thanks to @DaZuiZui for the original investigation and for surfacing the issue clearly — they're credited as co-author on the commit.

Tests

New packages/rolldown/tests/behaviors/hash-stability.test.ts: builds the same set of inputs twice, once with an extra unrelated entry, and asserts shared chunks (rolldown-runtime, react, a, b) keep the same file name and code. This is a multi-build invariant that single-build snapshot tests can't capture.
cargo test -p rolldown_utils covers visit_with_placeholders_defaulted, including the case where an unknown placeholder-shaped literal is emitted verbatim.
The crates/rolldown/tests/rolldown/hash/content_include_placeholder fixture (source console.log('_shared-!~{003}~.js');) doubles as a regression test for the user-literal carve-out — its snapshot would silently change if we ever started normalizing unknown markers again.
Existing hash-related snapshots have been refreshed; the changes are hash-only, content unchanged.

Known CI failure: `Vite Test Ubuntu`

The Vite Test Ubuntu check fails on two js-sourcemap.spec.ts cases (sourcemap is correct when preload information is injected, sourcemap is correct when using object as "define" value). These tests use toMatchInlineSnapshot with concrete chunk hash strings (e.g. dynamic-foo-B4JkVMbo.js, after-preload-dynamic-Dxsmo7dM.js.map) hardcoded in the Vite rolldown-canary branch — those values were produced by main's hashing algorithm. This PR changes the algorithm, so the snapshots no longer match.

This is purely a cross-repo snapshot-pinning artefact, not a regression in behavior: the sourcemaps themselves are correct, only the embedded chunk file-name hashes differ. The Vite-side inline snapshots will need to be refreshed once this PR lands and a new rolldown canary is published. Asking reviewers to accept this failure.

🤖 Generated with Claude Code

netlify · 2026-05-18T13:17:11Z

✅ Deploy Preview for rolldown-rs ready!

Name	Link
🔨 Latest commit	`8902861`
🔍 Latest deploy log	https://app.netlify.com/projects/rolldown-rs/deploys/6a0b110c1b76d40008ed94ce
😎 Deploy Preview	https://deploy-preview-9444--rolldown-rs.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.
🤖 Make changes	Run an agent on this branch

To edit notification comments on pull requests, go to your Netlify project configuration.

netlify · 2026-05-18T13:18:33Z

✅ Deploy Preview for rolldown-rs ready!

Name	Link
🔨 Latest commit	`6cc8eed`
🔍 Latest deploy log	https://app.netlify.com/projects/rolldown-rs/deploys/6a0c5b61c4bd0b00089441a2
😎 Deploy Preview	https://deploy-preview-9444--rolldown-rs.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.
🤖 Make changes	Run an agent on this branch

To edit notification comments on pull requests, go to your Netlify project configuration.

Copilot

Pull request overview

This PR fixes non-deterministic chunk [hash] filenames across builds when the chunk graph changes (e.g. adding an unrelated isolated entry), by making hashing independent of placeholder indices and adding Rollup-like filename collision deconfliction.

Changes:

Normalize hash placeholders in rendered chunk content before computing standalone content hashes, so placeholder index shifts don’t perturb hashes.
Remove preliminary_filename from the final-hash inputs and add a deterministic “rehash-until-unique” loop to avoid output filename collisions (case-insensitively).
Add a multi-build stability regression test and refresh affected snapshots.

Reviewed changes

Copilot reviewed 15 out of 15 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
packages/rolldown/tests/fixtures/plugin/context/emit-chunk/_config.ts	Updates inline snapshot expected chunk filename after hash behavior change.
packages/rolldown/tests/fixtures/plugin/augment-chunk-hash/_config.ts	Updates inline snapshot expected chunk filenames after hash behavior change.
packages/rolldown/tests/fixtures/output/sanitize-filename/function/_config.ts	Updates inline snapshot expected chunk filename after hash behavior change.
packages/rolldown/tests/fixtures/output/hash-filenames/normal/_config.ts	Updates inline snapshots for `[hash]` filenames and import/dynamic-import references.
packages/rolldown/tests/fixtures/misc/virtual/virtual-module-as-chunk/_config.ts	Updates inline snapshot expected virtual module chunk filename after hash behavior change.
packages/rolldown/tests/behaviors/hash-stability.test.ts	New regression test asserting stable filenames/code for shared chunks when adding an unrelated entry.
crates/rolldown/tests/rolldown/topics/new_url/nested_dirs/artifacts.snap	Refreshes expected hashed chunk filename and corresponding import path in snapshot output.
crates/rolldown/tests/rolldown/sourcemap/debug_ids/artifacts.snap	Updates expected debugId after hash/debug-id derivation changes.
crates/rolldown/tests/rolldown/hash/dynamic_import/artifacts.snap	Refreshes expected hashed filenames and dynamic import specifier in snapshot output.
crates/rolldown/tests/rolldown/hash/content_include_placeholder/artifacts.snap	Refreshes expected hashed filename in snapshot output.
crates/rolldown/tests/rolldown/hash/code_splitting/artifacts.snap	Refreshes expected hashed filenames and cross-chunk import specifiers in snapshot output.
crates/rolldown/tests/rolldown/hash/basic/artifacts.snap	Refreshes expected hashed filename in snapshot output.
crates/rolldown/tests/rolldown/function/entry_filenames/should_generate_specified_hash_length/artifacts.snap	Refreshes expected hashed entry filename for fixed-length hash output.
crates/rolldown/src/utils/chunk/finalize_chunks.rs	Normalizes placeholders prior to content hashing; removes preliminary filename from final hash inputs; adds Rollup-like filename collision deconfliction via rehash loop.
crates/rolldown_utils/src/hash_placeholder.rs	Adds `replace_placeholders_with_default` helper plus unit test to support stable content hashing.

codspeed-hq · 2026-05-18T13:23:22Z

Merging this PR will not alter performance

✅ 4 untouched benchmarks
⏩ 10 skipped benchmarks¹

_{Comparing yhf/fix-hash-stability-9339 (6cc8eed) with main (4a2ac11)}

10 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

…tion Previously `visit_with_placeholders_defaulted` normalized every syntactically valid `!~{...}~` marker to a zero-filled placeholder before hashing. That included literals in user source code that happened to match the shape — so a chunk whose code contains `'!~{foo}~'` and a chunk whose code contains `'!~{bar}~'` would collide on the same content hash, breaking content-addressable cache invalidation when the user-supplied string changed. Match Rollup's `replacePlaceholdersWithDefaultAndGetContainedPlaceholders` by only normalizing placeholders rolldown itself generated (registered in `ins_chunk_idx_by_placeholder`); unknown markers are emitted verbatim so changes to their bytes still flow into the hash. The `crates/rolldown/tests/rolldown/hash/content_include_placeholder` fixture is exactly this case (its source is `console.log('_shared-!~{003}~.js');`), so its snapshot hash updates to reflect the literal now being hashed verbatim. Also document the one remaining Phase 3 -> importer propagation gap as a known limitation in `meta/design/chunk-hash.md`; Rollup behaves the same way and fixing it would require topo-order processing. Reported by Codex review on PR #9444. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

`to_ascii_lowercase` only folds ASCII case, so two chunk filenames that differ only in non-ASCII case (e.g. `Á.js` vs `á.js`) wouldn't be detected as colliding on HFS+/NTFS — yet the doc comment promises case-insensitive deconfliction "safe to write on case-insensitive filesystems". Use `to_lowercase` to match Rollup's `toLowerCase()` and properly cover non-ASCII chunk names. The cost is negligible in this sequential post-pass. Reported in Copilot review on PR #9444. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

DaZuiZui · 2026-05-19T11:22:08Z

thank

Hash placeholder indices (`!~{000}~`, `!~{001}~`, ...) are assigned in rendering order and shift whenever an unrelated chunk is added, so previously they leaked into both the standalone content hash (via raw chunk content) and the final hash (via `preliminary_filename`). Adding or removing an isolated entry could change every other chunk's file name even when the chunks were byte-identical. Two changes, mirroring Rollup's `generateFinalHashes`: - Normalize hash placeholders to a zero-filled placeholder of the same shape (`!~{000...}~`) before hashing chunk content, so the content hash depends only on the actual bytes and not on transient indices. - Stop mixing `preliminary_filename` into the final hash. Two chunks whose resolved file names would collide (case-insensitively) are now deconflicted by rehashing the colliding chunk until its file name is unique. Fixes #9339. Co-authored-by: DaZuiZui <66861267+DaZuiZui@users.noreply.github.com> Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Replace `replace_placeholders_with_default` (which materialized a chunk-sized normalized `String` per chunk) with `visit_with_placeholders_defaulted`, a `FnMut(&[u8])` visitor that feeds the normalized byte sequence straight into the Xxh3 streaming hasher. The byte sequence fed in is bit-for-bit identical to the old materialized version, so the resulting hash — and every snapshot — is unchanged; only the allocation is gone. For a bundle with N chunks averaging M bytes each, this drops ~N * M bytes of throwaway allocations per build. Rollup still materializes the normalized string at this point, so this is a place where rolldown intentionally diverges to be lighter on memory. Also add meta/design/chunk-hash.md describing the three invariants the hash mechanism has to satisfy (stability across builds, sensitivity to real content, uniqueness within a build) and how the two parallel phases plus the sequential deconflict loop combine to satisfy them. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…tion Previously `visit_with_placeholders_defaulted` normalized every syntactically valid `!~{...}~` marker to a zero-filled placeholder before hashing. That included literals in user source code that happened to match the shape — so a chunk whose code contains `'!~{foo}~'` and a chunk whose code contains `'!~{bar}~'` would collide on the same content hash, breaking content-addressable cache invalidation when the user-supplied string changed. Match Rollup's `replacePlaceholdersWithDefaultAndGetContainedPlaceholders` by only normalizing placeholders rolldown itself generated (registered in `ins_chunk_idx_by_placeholder`); unknown markers are emitted verbatim so changes to their bytes still flow into the hash. The `crates/rolldown/tests/rolldown/hash/content_include_placeholder` fixture is exactly this case (its source is `console.log('_shared-!~{003}~.js');`), so its snapshot hash updates to reflect the literal now being hashed verbatim. Also document the one remaining Phase 3 -> importer propagation gap as a known limitation in `meta/design/chunk-hash.md`; Rollup behaves the same way and fixing it would require topo-order processing. Reported by Codex review on PR #9444. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…n_single_items` The unit test built a one-element `FxHashSet` with `["..."].into_iter().collect()`, which clippy `iter-on-single-items` (denied as warning in CI) flags as unnecessary allocation/iteration on a single-element array. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

`to_ascii_lowercase` only folds ASCII case, so two chunk filenames that differ only in non-ASCII case (e.g. `Á.js` vs `á.js`) wouldn't be detected as colliding on HFS+/NTFS — yet the doc comment promises case-insensitive deconfliction "safe to write on case-insensitive filesystems". Use `to_lowercase` to match Rollup's `toLowerCase()` and properly cover non-ASCII chunk names. The cost is negligible in this sequential post-pass. Reported in Copilot review on PR #9444. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Findings from /simplify pass on the full PR: - finalize_chunks.rs: the 9-line block above the standalone-content-hash closure was largely duplicated by the doc comment on `visit_with_placeholders_defaulted` and by `meta/design/chunk-hash.md`. Collapse to one line pointing at the design doc. - finalize_chunks.rs: the 3-line block above the `deconflict_filenames` call duplicated the function's own doc comment. Drop. - chunk-hash.md: the header said "Two-Phase Pipeline" but the diagram below and the body describe three phases (Phase 1, 2, 3). Rename to "The Pipeline". - chunk-hash.md: the Phase 3 code snippet still used `to_ascii_lowercase`, drifted out of sync after the prior commit switched the code to `to_lowercase` per the Copilot review. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

@IWANABETHATGUY

## [1.0.2] - 2026-05-20 ### 🚀 Features - devtools: emit package size in PackageGraphReady (#9434) by @IWANABETHATGUY - devtools: classify package dependency types (#9427) by @IWANABETHATGUY - devtools: map packages to modules and chunks (#9426) by @IWANABETHATGUY - devtools: mark used packages (#9423) by @IWANABETHATGUY - devtools: make duplicate packages discoverable (#9422) by @IWANABETHATGUY - devtools: emit package metadata (#9421) by @IWANABETHATGUY - update oxc to 0.132.0 (#9449) by @shulaoda - update oxc to 0.131.0 (#9424) by @shulaoda - allow checks.* to escalate emissions to hard errors (#9388) by @IWANABETHATGUY - dev: support watcher options `include` and `exclude` (#9395) by @h-a-n-a - emit warnings for invalid pure annotations (#9381) by @Kyujenius ### 🐛 Bug Fixes - hash: keep chunk file names stable when an unrelated entry is added (#9444) by @hyf0 - call `codeSplitting.groups[].name` in deterministic order (#9457) by @sapphi-red - dev/lazy: make `resolve_id` idempotent when the resolved id is already a lazy entry (#9439) by @h-a-n-a - chunk-optimization: publish absorbed dynamic-entry namespace cross-chunk (#9448) by @IWANABETHATGUY - treeshake: propagate pure annotation through compound exprs (#9431) by @Dunqing - finalizer: skip redundant init call when barrel executes in same chunk (#9354) by @IWANABETHATGUY - linking: initialize wrapped ESM re-export owners (#9353) by @IWANABETHATGUY - do not inherit __toESM across chunks for named-only external imports (#9333) (#9415) by @IWANABETHATGUY - watcher: don't write output or emit events after close() (#9328) by @situ2001 - chunk-optimization: avoid unsafe dynamic-only merges (#9398) by @IWANABETHATGUY - cjs: rename CJS-wrapped locals that would shadow chunk-scope names (#9392) by @hyf0 - dev/lazy: watch lazy modules added in rebuilds (#9391) by @h-a-n-a ### 🚜 Refactor - rolldown_dev: move dev example to break publish cycle (#9465) by @Boshen - binding: drop unsafe napi string helper, hoist transform ArcStr (#9456) by @hyf0 - ecmascript_utils: split rewrite_ident_reference off JsxExt trait (#9417) by @IWANABETHATGUY - use `ThreadsafeFunction::call_async_catch` (#9390) by @sapphi-red ### 📚 Documentation - devtools: document @rolldown/debug usage and package graph consumption (#9435) by @IWANABETHATGUY - replace `Inter` with system font stack in OG template SVG (#9240) by @yvbopeng - remove `output.comments` warning as all issues have been resolved (#9393) by @sapphi-red - in-depth: clarify @__PURE__ scope and document positions (#9389) by @Kyujenius - readme: remove release candidate notice (#9387) by @shulaoda ### ⚡ Performance - vite-resolve: cache importer existence checks (#9443) by @Brooooooklyn - binding: reduce plugin string clones (#9436) by @Brooooooklyn ### 🧪 Testing - enable `legal_comments_inline` test (#9394) by @sapphi-red ### ⚙️ Miscellaneous Tasks - bump pnpm to v11.1.2 (#9447) by @Boshen - deps: update rust crates (#9461) by @renovate[bot] - deps: update rollup submodule for tests to v4.60.4 (#9453) by @rolldown-guard[bot] - deps: update test262 submodule for tests (#9454) by @rolldown-guard[bot] - deps: update npm packages (#9430) by @renovate[bot] - deps: update github actions (#9429) by @renovate[bot] - deps: update dependency rolldown-plugin-dts to v0.25.1 (#9452) by @renovate[bot] - deps: update rust crates (#9428) by @renovate[bot] - revert allow checks.* to escalate emissions to hard errors (#9388) (#9438) by @IWANABETHATGUY - update mimalloc-safe to 0.1.61 (#9413) by @shulaoda - deny `todo`, `unimplemented`, and `print_stderr` clippy lints (#9412) by @Boshen - deps: update mimalloc-safe to 0.1.60 (#9410) by @shulaoda - remove `pip install setuptools` workaround for node-gyp on macOS (#9370) by @sapphi-red - renovate: disable automerge to require manual approval (#9386) by @shulaoda - deps: update napi (#9385) by @renovate[bot] ### ❤️ New Contributors * @yvbopeng made their first contribution in [#9240](#9240) Co-authored-by: shulaoda <165626830+shulaoda@users.noreply.github.com>

ryanto · 2026-05-20T14:47:01Z

thanks @hyf0 !

@DaZuiZui

…ded (rolldown#9444) Fixes rolldown#9339: adding or removing an isolated entry currently changes the file names of every other chunk in the bundle, even when those chunks are byte-identical between builds. The reported example uses `entryFileNames` / `chunkFileNames` with `codeSplitting` and observes the rolldown runtime chunk's hash flip across builds that should produce identical output. Hash placeholders (`!~{000}~`, `!~{001}~`, …) are assigned in rendering order, so their numeric indices shift whenever the chunk graph changes. Those indices leak into the hash from two places in `finalize_chunks.rs`: 1. **Standalone content hash**: the chunk's raw content includes placeholders pointing at sibling chunks (e.g. `import "./shared-!~{001}~.js"`), so when a sibling's index moves, this chunk's content bytes — and therefore its content hash — change. 2. **Final hash**: `preliminary_filename` (which contains the chunk's own placeholder) is mixed in directly, so when the chunk's own index moves, its final hash moves too. Mirrors Rollup's `generateFinalHashes`: 1. **Normalize generated placeholders before hashing content.** A new `visit_with_placeholders_defaulted` helper in `rolldown_utils::hash_placeholder` walks the chunk content and streams it into `Xxh3` byte-slice by byte-slice, replacing each rolldown-generated `!~{xxx}~` with a zero-filled placeholder of the same shape. Placeholders rolldown didn't generate (literals in user source code that just happen to match the shape) are emitted verbatim so changes to them still flow into the hash — same `placeholders.has(placeholder)` check Rollup performs in `replacePlaceholdersWithDefaultAndGetContainedPlaceholders`. Streaming avoids materializing a chunk-sized normalized `String` per chunk; for a typical bundle with ~MB chunks containing cross-chunk imports, that's ~bundle-size's worth of throwaway allocations per build avoided. 2. **Drop `preliminary_filename` from the final hash, deconflict at the file-name layer instead.** Two chunks whose resolved file names would collide (case-insensitively, for HFS+/NTFS safety) are now disambiguated by rehashing the colliding chunk's hash until its file name is unique — the same `do { … } while (collision)` loop Rollup uses. A design doc covering the three invariants (stability across builds, sensitivity to real content, uniqueness within a build), the two-phase parallel pipeline + sequential deconflict pass, and known limitations is added at `meta/design/chunk-hash.md`. made it not the right shape to land as-is: - It kept the buggy `preliminary_filename.hash(&mut hasher)` line under an `if options.sourcemap_debug_ids` branch, so users with `sourcemapDebugIds: true` would still hit the original non-determinism. - It removed `preliminary_filename` from the final hash in the default mode without adding any collision handling, regressing the case Rollup explicitly covers in its [`deconflict-hashes` test](https://github.com/rollup/rollup/tree/master/test/chunking-form/samples/hashing/deconflict-hashes) (two byte-identical entries + `entryFileNames: '[hash].js'` → Rollup produces two distinct file names via rehash; rolldown#9356 would silently produce two assets with the same name). This PR takes the same overall direction (normalize placeholders in content) but lands the rehash loop alongside it so both invariants hold simultaneously, and applies uniformly regardless of `sourcemapDebugIds`. Thanks to @DaZuiZui for the original investigation and for surfacing the issue clearly — they're credited as co-author on the commit. - New `packages/rolldown/tests/behaviors/hash-stability.test.ts`: builds the same set of inputs twice, once with an extra unrelated entry, and asserts shared chunks (`rolldown-runtime`, `react`, `a`, `b`) keep the same file name and code. This is a multi-build invariant that single-build snapshot tests can't capture. - `cargo test -p rolldown_utils` covers `visit_with_placeholders_defaulted`, including the case where an unknown placeholder-shaped literal is emitted verbatim. - The `crates/rolldown/tests/rolldown/hash/content_include_placeholder` fixture (source `console.log('_shared-!~{003}~.js');`) doubles as a regression test for the user-literal carve-out — its snapshot would silently change if we ever started normalizing unknown markers again. - Existing hash-related snapshots have been refreshed; the changes are hash-only, content unchanged. The `Vite Test Ubuntu` check fails on two `js-sourcemap.spec.ts` cases (`sourcemap is correct when preload information is injected`, `sourcemap is correct when using object as "define" value`). These tests use `toMatchInlineSnapshot` with concrete chunk hash strings (e.g. `dynamic-foo-B4JkVMbo.js`, `after-preload-dynamic-Dxsmo7dM.js.map`) hardcoded in the Vite `rolldown-canary` branch — those values were produced by `main`'s hashing algorithm. This PR changes the algorithm, so the snapshots no longer match. This is purely a cross-repo snapshot-pinning artefact, not a regression in behavior: the sourcemaps themselves are correct, only the embedded chunk file-name hashes differ. The Vite-side inline snapshots will need to be refreshed once this PR lands and a new rolldown canary is published. Asking reviewers to accept this failure. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: DaZuiZui <66861267+DaZuiZui@users.noreply.github.com> Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>

@IWANABETHATGUY

## [1.0.2] - 2026-05-20 ### 🚀 Features - devtools: emit package size in PackageGraphReady (rolldown#9434) by @IWANABETHATGUY - devtools: classify package dependency types (rolldown#9427) by @IWANABETHATGUY - devtools: map packages to modules and chunks (rolldown#9426) by @IWANABETHATGUY - devtools: mark used packages (rolldown#9423) by @IWANABETHATGUY - devtools: make duplicate packages discoverable (rolldown#9422) by @IWANABETHATGUY - devtools: emit package metadata (rolldown#9421) by @IWANABETHATGUY - update oxc to 0.132.0 (rolldown#9449) by @shulaoda - update oxc to 0.131.0 (rolldown#9424) by @shulaoda - allow checks.* to escalate emissions to hard errors (rolldown#9388) by @IWANABETHATGUY - dev: support watcher options `include` and `exclude` (rolldown#9395) by @h-a-n-a - emit warnings for invalid pure annotations (rolldown#9381) by @Kyujenius ### 🐛 Bug Fixes - hash: keep chunk file names stable when an unrelated entry is added (rolldown#9444) by @hyf0 - call `codeSplitting.groups[].name` in deterministic order (rolldown#9457) by @sapphi-red - dev/lazy: make `resolve_id` idempotent when the resolved id is already a lazy entry (rolldown#9439) by @h-a-n-a - chunk-optimization: publish absorbed dynamic-entry namespace cross-chunk (rolldown#9448) by @IWANABETHATGUY - treeshake: propagate pure annotation through compound exprs (rolldown#9431) by @Dunqing - finalizer: skip redundant init call when barrel executes in same chunk (rolldown#9354) by @IWANABETHATGUY - linking: initialize wrapped ESM re-export owners (rolldown#9353) by @IWANABETHATGUY - do not inherit __toESM across chunks for named-only external imports (rolldown#9333) (rolldown#9415) by @IWANABETHATGUY - watcher: don't write output or emit events after close() (rolldown#9328) by @situ2001 - chunk-optimization: avoid unsafe dynamic-only merges (rolldown#9398) by @IWANABETHATGUY - cjs: rename CJS-wrapped locals that would shadow chunk-scope names (rolldown#9392) by @hyf0 - dev/lazy: watch lazy modules added in rebuilds (rolldown#9391) by @h-a-n-a ### 🚜 Refactor - rolldown_dev: move dev example to break publish cycle (rolldown#9465) by @Boshen - binding: drop unsafe napi string helper, hoist transform ArcStr (rolldown#9456) by @hyf0 - ecmascript_utils: split rewrite_ident_reference off JsxExt trait (rolldown#9417) by @IWANABETHATGUY - use `ThreadsafeFunction::call_async_catch` (rolldown#9390) by @sapphi-red ### 📚 Documentation - devtools: document @rolldown/debug usage and package graph consumption (rolldown#9435) by @IWANABETHATGUY - replace `Inter` with system font stack in OG template SVG (rolldown#9240) by @yvbopeng - remove `output.comments` warning as all issues have been resolved (rolldown#9393) by @sapphi-red - in-depth: clarify @__PURE__ scope and document positions (rolldown#9389) by @Kyujenius - readme: remove release candidate notice (rolldown#9387) by @shulaoda ### ⚡ Performance - vite-resolve: cache importer existence checks (rolldown#9443) by @Brooooooklyn - binding: reduce plugin string clones (rolldown#9436) by @Brooooooklyn ### 🧪 Testing - enable `legal_comments_inline` test (rolldown#9394) by @sapphi-red ### ⚙️ Miscellaneous Tasks - bump pnpm to v11.1.2 (rolldown#9447) by @Boshen - deps: update rust crates (rolldown#9461) by @renovate[bot] - deps: update rollup submodule for tests to v4.60.4 (rolldown#9453) by @rolldown-guard[bot] - deps: update test262 submodule for tests (rolldown#9454) by @rolldown-guard[bot] - deps: update npm packages (rolldown#9430) by @renovate[bot] - deps: update github actions (rolldown#9429) by @renovate[bot] - deps: update dependency rolldown-plugin-dts to v0.25.1 (rolldown#9452) by @renovate[bot] - deps: update rust crates (rolldown#9428) by @renovate[bot] - revert allow checks.* to escalate emissions to hard errors (rolldown#9388) (rolldown#9438) by @IWANABETHATGUY - update mimalloc-safe to 0.1.61 (rolldown#9413) by @shulaoda - deny `todo`, `unimplemented`, and `print_stderr` clippy lints (rolldown#9412) by @Boshen - deps: update mimalloc-safe to 0.1.60 (rolldown#9410) by @shulaoda - remove `pip install setuptools` workaround for node-gyp on macOS (rolldown#9370) by @sapphi-red - renovate: disable automerge to require manual approval (rolldown#9386) by @shulaoda - deps: update napi (rolldown#9385) by @renovate[bot] ### ❤️ New Contributors * @yvbopeng made their first contribution in [rolldown#9240](rolldown#9240) Co-authored-by: shulaoda <165626830+shulaoda@users.noreply.github.com>

Copilot AI review requested due to automatic review settings May 18, 2026 13:15

github-actions Bot assigned hyf0 May 18, 2026

Copilot started reviewing on behalf of hyf0 May 18, 2026 13:16 View session

Copilot AI reviewed May 18, 2026

View reviewed changes

Comment thread crates/rolldown/src/utils/chunk/finalize_chunks.rs Outdated

hyf0 requested a review from sapphi-red May 19, 2026 06:51

hyf0 changed the title ~~fix(hash): stabilize chunk content hashes across builds~~ fix(hash): keep chunk file names stable when an unrelated entry is added May 19, 2026

sapphi-red approved these changes May 19, 2026

View reviewed changes

hyf0 and others added 10 commits May 19, 2026 20:45

[autofix.ci] apply automated fixes

0824b59

[autofix.ci] apply automated fixes

b4f3c78

test(hash-stability): link to issue #9339

638ec4e

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

[autofix.ci] apply automated fixes

3a034d8

hyf0 force-pushed the yhf/fix-hash-stability-9339 branch from 6c04184 to 6cc8eed Compare May 19, 2026 12:45

hyf0 enabled auto-merge (squash) May 19, 2026 12:45

hyf0 mentioned this pull request May 19, 2026

fix: stabilize chunk content hashes #9356

Closed

hyf0 merged commit 0f811f4 into main May 19, 2026
30 of 31 checks passed

hyf0 deleted the yhf/fix-hash-stability-9339 branch May 19, 2026 12:48

rolldown-guard Bot mentioned this pull request May 20, 2026

release: v1.0.2 #9478

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(hash): keep chunk file names stable when an unrelated entry is added#9444

fix(hash): keep chunk file names stable when an unrelated entry is added#9444
hyf0 merged 10 commits into
mainfrom
yhf/fix-hash-stability-9339

hyf0 commented May 18, 2026 •

edited

Loading

Uh oh!

netlify Bot commented May 18, 2026

Uh oh!

netlify Bot commented May 18, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

codspeed-hq Bot commented May 18, 2026 •

edited

Loading

Uh oh!

DaZuiZui commented May 19, 2026

Uh oh!

Uh oh!

ryanto commented May 20, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Conversation

hyf0 commented May 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Root cause

Fix

Why a separate PR, not a follow-up to #9356

Tests

Known CI failure: Vite Test Ubuntu

Uh oh!

netlify Bot commented May 18, 2026

✅ Deploy Preview for rolldown-rs ready!

Uh oh!

netlify Bot commented May 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

✅ Deploy Preview for rolldown-rs ready!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

codspeed-hq Bot commented May 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will not alter performance

Footnotes

Uh oh!

DaZuiZui commented May 19, 2026

Uh oh!

Uh oh!

ryanto commented May 20, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

hyf0 commented May 18, 2026 •

edited

Loading

Known CI failure: `Vite Test Ubuntu`

netlify Bot commented May 18, 2026 •

edited

Loading

codspeed-hq Bot commented May 18, 2026 •

edited

Loading