perf(generate): thread ast_table by value into codegen consumer#9555
Merged
IWANABETHATGUY merged 1 commit intoMay 26, 2026
Merged
Conversation
✅ Deploy Preview for rolldown-rs canceled.
|
Merging this PR will not alter performance
Comparing 
Footnotes
|
This was referenced May 26, 2026
Boshen
added a commit
that referenced
this pull request
May 26, 2026
…degen_ret_map `ast_table: IndexEcmaAst` is consumed by value on purpose — the whole point of #9555 is to have the compiler drop the per-module bumpalo arenas at scope exit, before the rest of `instantiate_chunks` runs. Clippy's `needless_pass_by_value` (pedantic) flags it because the body only borrows; suppress with a comment explaining the intent.
Boshen
changed the title
IWANABETHATGUY
approved these changes
May 26, 2026
Member
Author
Merge activity
|
## Summary Alternative to #9554. Same memory win on the realistic case, enforced by the type system instead of by a comment. Remove `ast_table` from `LinkStageOutput` and change `LinkStage::link()` to return `(LinkStageOutput, IndexEcmaAst)`. The AST table then flows by value through `bundle_up → GenerateStage::new → generate → finalize_modules → render_chunk_to_assets → instantiate_chunks → create_chunk_to_codegen_ret_map`, where the compiler drops it at scope exit (`create_chunk_to_codegen_ret_map` is the last reader). Adding a new post-codegen `ast_table` reader becomes a compile error — there is no field to read. ## Measured impact (3 runs each, verified locally with dhat + getrusage + `/usr/bin/time -ahl`) | Config | main | This PR | Δ vs main | Δ vs #9554 | |---|---:|---:|---:|---:| | **dhat At t-gmax (peak heap)** | | | | | | rome | 140.51 MiB | 108.07 MiB | **−32.4 MiB (−23.1 %)** | −2.08 MiB | | rome --minify | 159.78 MiB | 113.09 MiB | **−46.7 MiB (−29.2 %)** | +0.01 MiB | | **ru_maxrss (getrusage, 3-run avg)** | | | | | | rome | 172.05 MiB | 170.56 MiB | −1.49 MiB | +1.17 MiB | | rome --minify | 183.56 MiB | 175.71 MiB | −7.85 MiB | −5.70 MiB | | **time -ahl peak RSS (3-run avg)** | | | | | | rome | 180.99 MiB | 179.31 MiB | −1.68 MiB | +1.15 MiB | | rome --minify | 193.37 MiB | 184.69 MiB | −8.68 MiB | −6.42 MiB | The ~2 MiB additional saving on rome no-minify vs #9554 comes from dropping one frame earlier (inside `instantiate_chunks` rather than after it returns) — which matters when peak heap is reached during `instantiate_chunks`'s own `try_join_all`. On rome --minify, peak is firmly in `finalize_assets`, far past both drop points, so the two PRs are identical at the dhat level (the OS-RSS gap there is run-to-run noise — system malloc's page caching is sticky). ## Trade-offs vs #9554 - Pros: Compile-time guarantee that no post-codegen reader can exist. - Cons: 5 files instead of 1; signature changes through the codegen pipeline. Pick one to land — they are mutually exclusive. The minimal version (#9554) is simpler; this one is what you'd reach for if the team values type-system enforcement over diff size. Refs #9516.
graphite-app
Bot
force-pushed
the
refactor/thread-ast-table-by-value-into-codegen
branch
from
6ce1d61 to
800f077
Compare
IWANABETHATGUY
deleted the
refactor/thread-ast-table-by-value-into-codegen
branch
Merged
shulaoda
added a commit
that referenced
this pull request
May 27, 2026
## [1.0.3] - 2026-05-27 ### 🚀 Features - transform: respect decorator strictNullChecks option (#9580) by @kylecannon - drop `defer` keyword (#9503) by @TheAlexLichter ### 🐛 Bug Fixes - ci: create target dir before cargo release-oxc update (#9584) by @shulaoda - ci: reorder prepare-release steps to avoid dirty git check failure (#9583) by @shulaoda - testing: canonicalize temp dir early and use platform-specific separator in test262 (#9582) by @shulaoda - testing: resolve symlinked temp dir in test262 snapshot normalization (#9581) by @shulaoda - testing: canonicalize temp dir path in test262 snapshot normalization (#9579) by @shulaoda - dev: `onOutput` called twice when initial build fails (#9552) by @hyf0 - dev: make `ensureCurrentBuildFinish` not returning error when engine closes (#9564) by @h-a-n-a - oxc-runtime: route require() to CJS helper variant (#9263) (#9526) by @IWANABETHATGUY - generator: use exporter chunk's export mode for CJS default re-exports (#9299) (#9529) by @IWANABETHATGUY - rolldown: always run reduced-atom static cycle check (#9441) (#9514) by @IWANABETHATGUY - apply transform.dropLabels before scanning (#9521) (#9522) by @IWANABETHATGUY - rolldown_watcher: take `rolldown` dep through the workspace (#9510) by @Boshen - cache: keep the scan-stage cache consistent when a build fails (#9495) by @h-a-n-a - skip JSON default-import namespace optimization for write targets (#9484) (#9489) by @IWANABETHATGUY - deps: skip pnpm frozen-lockfile on Netlify to dodge catalog mismatch bug (#9471) by @Boshen ### 🚜 Refactor - oxc-runtime: use Cow for helper path construction (#9538) by @IWANABETHATGUY - fold import defer phase drop into PreProcessor (#9524) by @IWANABETHATGUY - distinguish `map: null` vs `map: undefined` in transform hook output (#9497) by @sapphi-red ### 📚 Documentation - explain the policy for Rust crates (#9547) by @sapphi-red - cache: add design doc for cache (#9544) by @h-a-n-a - guide/troubleshooting: add TDZ error section (#9537) by @sapphi-red - dev-engine: add design doc for dev-engine (#9479) by @h-a-n-a - lazy-barrel: tweak some words (#9483) by @shulaoda - lazy-barrel: expand reasoning behind LARGE_BARREL_MODULES advice (#9477) by @shulaoda ### ⚡ Performance - generate: thread ast_table by value into codegen consumer (#9555) by @Boshen - finalizers: replace `_reExport` construction with a direct call to avoid calling `clone_in` (#9501) by @Dunqing - reorder hot-path boolean checks to short-circuit on cheap predicates first (#9523) by @Boshen ### 🧪 Testing - rolldown: regression fixture for #9401 (#9418) by @IWANABETHATGUY - failing test for #9441 (#9504) by @TheAlexLichter ### ⚙️ Miscellaneous Tasks - deps: upgrade oxc to 0.133.0 (#9563) by @Dunqing - deps: update crate-ci/typos action to v1.46.3 (#9576) by @renovate[bot] - deps: update mimalloc-safe to 0.1.62 (#9577) by @shulaoda - mimalloc-safe: update to a bug-fix branch for verification (#9569) by @shulaoda - deps: update test262 submodule for tests (#9551) by @rolldown-guard[bot] - point published crates' readme to root README.md (#9553) by @Boshen - replace actions-cool/issues-helper with gh CLI (#9543) by @Boshen - deps: update cargo-shear to 1.12.4 (#9541) by @Boshen - deps: update taiki-e/install-action action to v2.79.4 (#9535) by @renovate[bot] - deps: update github actions (#9532) by @renovate[bot] - deps: update rust crates (#9534) by @renovate[bot] - deps: update npm packages (#9533) by @renovate[bot] - gate experimental/testing-only items to silence dead_code in publish builds (#9517) by @Boshen - docs: deploy to Void (#9509) by @Boshen - release: set up cargo-release-oxc for publishing crates (#9476) by @Boshen - rolldown_plugin_lazy_compilation: add missing description (#9507) by @Boshen - mimalloc-safe: update to a bug-fix branch for verification (#9506) by @shulaoda - deps: update crate-ci/typos action to v1.46.2 (#9468) by @renovate[bot] ### ❤️ New Contributors * @kylecannon made their first contribution in [#9580](#9580)
shulaoda
pushed a commit
that referenced
this pull request
May 27, 2026
## [1.0.3] - 2026-05-27 ### 🚀 Features - transform: respect decorator strictNullChecks option (#9580) by @kylecannon - drop `defer` keyword (#9503) by @TheAlexLichter ### 🐛 Bug Fixes - ci: create target dir before cargo release-oxc update (#9584) by @shulaoda - ci: reorder prepare-release steps to avoid dirty git check failure (#9583) by @shulaoda - testing: canonicalize temp dir early and use platform-specific separator in test262 (#9582) by @shulaoda - testing: resolve symlinked temp dir in test262 snapshot normalization (#9581) by @shulaoda - testing: canonicalize temp dir path in test262 snapshot normalization (#9579) by @shulaoda - dev: `onOutput` called twice when initial build fails (#9552) by @hyf0 - dev: make `ensureCurrentBuildFinish` not returning error when engine closes (#9564) by @h-a-n-a - oxc-runtime: route require() to CJS helper variant (#9263) (#9526) by @IWANABETHATGUY - generator: use exporter chunk's export mode for CJS default re-exports (#9299) (#9529) by @IWANABETHATGUY - rolldown: always run reduced-atom static cycle check (#9441) (#9514) by @IWANABETHATGUY - apply transform.dropLabels before scanning (#9521) (#9522) by @IWANABETHATGUY - rolldown_watcher: take `rolldown` dep through the workspace (#9510) by @Boshen - cache: keep the scan-stage cache consistent when a build fails (#9495) by @h-a-n-a - skip JSON default-import namespace optimization for write targets (#9484) (#9489) by @IWANABETHATGUY - deps: skip pnpm frozen-lockfile on Netlify to dodge catalog mismatch bug (#9471) by @Boshen ### 🚜 Refactor - oxc-runtime: use Cow for helper path construction (#9538) by @IWANABETHATGUY - fold import defer phase drop into PreProcessor (#9524) by @IWANABETHATGUY - distinguish `map: null` vs `map: undefined` in transform hook output (#9497) by @sapphi-red ### 📚 Documentation - explain the policy for Rust crates (#9547) by @sapphi-red - cache: add design doc for cache (#9544) by @h-a-n-a - guide/troubleshooting: add TDZ error section (#9537) by @sapphi-red - dev-engine: add design doc for dev-engine (#9479) by @h-a-n-a - lazy-barrel: tweak some words (#9483) by @shulaoda - lazy-barrel: expand reasoning behind LARGE_BARREL_MODULES advice (#9477) by @shulaoda ### ⚡ Performance - generate: thread ast_table by value into codegen consumer (#9555) by @Boshen - finalizers: replace `_reExport` construction with a direct call to avoid calling `clone_in` (#9501) by @Dunqing - reorder hot-path boolean checks to short-circuit on cheap predicates first (#9523) by @Boshen ### 🧪 Testing - rolldown: regression fixture for #9401 (#9418) by @IWANABETHATGUY - failing test for #9441 (#9504) by @TheAlexLichter ### ⚙️ Miscellaneous Tasks - deps: upgrade oxc to 0.133.0 (#9563) by @Dunqing - deps: update crate-ci/typos action to v1.46.3 (#9576) by @renovate[bot] - deps: update mimalloc-safe to 0.1.62 (#9577) by @shulaoda - mimalloc-safe: update to a bug-fix branch for verification (#9569) by @shulaoda - deps: update test262 submodule for tests (#9551) by @rolldown-guard[bot] - point published crates' readme to root README.md (#9553) by @Boshen - replace actions-cool/issues-helper with gh CLI (#9543) by @Boshen - deps: update cargo-shear to 1.12.4 (#9541) by @Boshen - deps: update taiki-e/install-action action to v2.79.4 (#9535) by @renovate[bot] - deps: update github actions (#9532) by @renovate[bot] - deps: update rust crates (#9534) by @renovate[bot] - deps: update npm packages (#9533) by @renovate[bot] - gate experimental/testing-only items to silence dead_code in publish builds (#9517) by @Boshen - docs: deploy to Void (#9509) by @Boshen - release: set up cargo-release-oxc for publishing crates (#9476) by @Boshen - rolldown_plugin_lazy_compilation: add missing description (#9507) by @Boshen - mimalloc-safe: update to a bug-fix branch for verification (#9506) by @shulaoda - deps: update crate-ci/typos action to v1.46.2 (#9468) by @renovate[bot] ### ❤️ New Contributors * @kylecannon made their first contribution in [#9580](#9580)
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.
Summary
Alternative to #9554. Same memory win on the realistic case, enforced by the type system instead of by a comment.
Remove
ast_tablefromLinkStageOutputand changeLinkStage::link()to return(LinkStageOutput, IndexEcmaAst). The AST table then flows by value throughbundle_up → GenerateStage::new → generate → finalize_modules → render_chunk_to_assets → instantiate_chunks → create_chunk_to_codegen_ret_map, where the compiler drops it at scope exit (create_chunk_to_codegen_ret_mapis the last reader). Adding a new post-codegenast_tablereader becomes a compile error — there is no field to read.Measured impact (3 runs each, verified locally with dhat + getrusage +
/usr/bin/time -ahl)The ~2 MiB additional saving on rome no-minify vs #9554 comes from dropping one frame earlier (inside
instantiate_chunksrather than after it returns) — which matters when peak heap is reached duringinstantiate_chunks's owntry_join_all. On rome --minify, peak is firmly infinalize_assets, far past both drop points, so the two PRs are identical at the dhat level (the OS-RSS gap there is run-to-run noise — system malloc's page caching is sticky).Trade-offs vs #9554
Pick one to land — they are mutually exclusive. The minimal version (#9554) is simpler; this one is what you'd reach for if the team values type-system enforcement over diff size.
Refs #9516.