perf(formatter): pre-size output buffer using source text length#22594
Merged
Conversation
Merging this PR will not alter performance
Comparing 
Footnotes
|
3 tasks
Member
Author
1% - 2% performance improvement. Using the input content length is fine because, in practice, most files are already formatted, so they are exactly the same. |
leaysgur
approved these changes
May 20, 2026
Member
Merge activity
|
Member
|
@Dunqing The PR itself is LGTM, but for some reason, the CI is failing, so I cannot merge it... Claude said it was an irrelevant change to the lexer, so I tried reverting it, but it still fails. Why is there a difference in transformer allocs even though we only modified the formatter...? Refs: #22614 🤷🏻 |
graphite-app Bot
pushed a commit
that referenced
this pull request
May 20, 2026
## Summary Adds `kitchen-sink.tsx` — a comprehensive synthetic TypeScript+JSX fixture maintained at [oxc-project/benchmark-files](https://github.com/oxc-project/benchmark-files) — to both `TestFiles::minimal()` (bench input set) and `TestFiles::complicated()` (alloc-tracking input set). The existing files in each set are untouched; this is a strict append. ## Why The existing bench input set didn't reliably surface general-purpose perf wins above the ~1-2% measurement noise floor: - #22580 (semantic pre-reserve) — visible because `binder.ts` exercises it - #22594 (formatter buffer) — visible - #22596 (minifier `try_fold_concat`) — **not visible** on the old set - #22599 (semantic resolve-refs no-temp-Vec) — **not visible** - #22603 (semantic var-hoist SmallVec) — **not visible** The kitchen-sink fixes that by exercising every AST node, every transformer plugin, every minifier optimization opportunity, and every semantic step in one large file. Verified by re-benching #22596 against this fixture: **minifier mean −1.5%, min −3.7%** — above noise, signal confirmed. ## Fixture stats (cross-checked locally) | Metric | Value | |---|---| | Source size | 21,117 lines / 732.90 kB | | AST nodes | ~133,000 | | Scopes | ~4,750 | | Symbols | ~7,000 | | Resolved references | ~16,000 | | Semantic diagnostics | 0 errors / 0 warnings | ## Snap baselines `tasks/track_memory_allocations/allocs_*.snap` updated with the kitchen-sink row across all 5 pipelines (parser / semantic / transformer / minifier / formatter). Future PRs that change allocation behavior on this fixture will produce a snap diff in CI. ## Bench-cleaner fix `tasks/benchmark/benches/lexer.rs`'s `SourceCleaner` was missing `visit_ts_template_literal_type` — TypeScript type-level template literals (e.g. `` `${T}-${U}` `` in conditional / mapped types) are syntactically identical to value-level template literals, so the bench-mode lexer (without parser context) cannot distinguish them. Without the cleaner converting them to plain strings, kitchen-sink's type-level templates caused the lexer bench to swallow ~1 KB spans as a single `TemplateHead` and produce spurious "Unterminated string" / "Invalid Unicode escape" errors. One-line fix to mirror the existing `visit_template_literal` handling. AI disclosure: drafted with Claude Code, reviewed manually.
fa8e5b8 to
14bd1ac
Compare
) ## Summary `Printer` constructs its underlying `CodeBuffer` (a `Vec<u8>`) via `CodeBuffer::with_indent`, which starts the buffer at `cap = 0`. For multi-MB outputs, that's ~`log2(output_size)` growth reallocations of a multi-MB buffer — each one allocates a bigger block, copies all existing bytes, and frees the old one. Add a `Printer::with_capacity(capacity, options, sorted_tailwind_classes)` constructor and route `Formatted::print` / `print_with_indent` through it with `source_text.len()` as the hint. Formatted output is typically within a small factor of the input size, so a single pre-allocation eliminates the growth reallocs in one shot. `Printer::new` is kept (delegates to `with_capacity(0, ...)`) for back-compat with internal callers. ## Allocation impact `cargo allocs`, `allocs_formatter.snap`: | File | Size | Sys reallocs before | Sys reallocs after | |---|---|---|---| | `checker.ts` | 2.92 MB | 584 | **565** (−19) | | `cal.com.tsx` | 1.06 MB | 787 | **773** (−14) | | `RadixUIAdoptionSection.jsx` | 2.5 KB | 37 | **28** (−9) | | `pdf.mjs` | 567 KB | 356 | **342** (−14) | | `antd.js` | 6.69 MB | 4671 | **4652** (−19) | | `binder.ts` | 193 KB | 58 | **43** (−15) | Sys alloc counts unchanged. The 9-19 reductions per file match `log2(file_size)` — exactly the doubling-growth pattern of the output buffer. ## Timing Local `cargo bench --bench formatter` shows small wins (~3-7%) on tightly-measured benches (errors.ts, Search.tsx, core.js, handle-comments.js). Two files (next.ts, index.tsx) have wide Criterion intervals that swallow any signal in either direction. App.tsx (the biggest at 5.97 ms) shows essentially no change — modern allocators handle large `Vec<u8>` reallocations via mmap remap rather than physical copy, so the wins on big files turn out smaller than the realloc count suggests. CodSpeed (instruction-count-based) will give the deterministic answer. ## Test Plan - [x] `cargo test -p oxc_formatter --lib --tests` — 364 tests pass - [x] `cargo allocs` — formatter snapshot shows clean reallocation reductions - [x] `cargo bench --bench formatter` — no regressions outside noise ## Side fix The commit hook flagged a pre-existing `clippy::redundant_pub_crate` warning in `crates/oxc_parser/src/lexer/mod.rs:38` (`pub(crate) use` inside a private module). One-character fix to `pub use` included so the hook passes. AI disclosure: drafted with Claude Code, reviewed manually.
14bd1ac to
78cf83f
Compare
Dunqing
added a commit
that referenced
this pull request
May 26, 2026
# Oxlint ### 🚀 Features - 10da26b linter: `no-misleading-character-class` add suggestions for regex literal (#22681) (Sysix) - b84941e linter/vue: Implement no-expose-after-await rule (#22675) (bab) - 98b98c1 linter/vue: Implement no-computed-properties-in-data rule (#22674) (bab) - 868e2e8 linter: Add suggestion for `no-misleading-character-class` (#22631) (Sysix) - 2d4c919 oxlint: Support `vite-plus/resolveConfig` for vite.config.ts (#22456) (leaysgur) - 2a60012 linter/vue: Implement require-render-return rule (#22613) (bab) - 9f227fd linter/vue: Implement no-deprecated-props-default-this rule (#21892) (bab) - 9cd28b3 linter: Add debug option to print files to be linted (#22546) (camchenry) - 87f065e linter/vue: Implement return-in-emits-validator rule (#21935) (bab) - ea0380c linter/unicorn: Implement `import-style` rule (#22173) (Hao Chen) - dde40fe linter/vue: Implement no-watch-after-await rule (#22006) (bab) - a735eb0 linter/vue: Implement valid-next-tick rule (#22531) (bab) - 6dc615d linter/vue: Implement no-shared-component-data rule (#21842) (bab) - a656418 linter/vue: Implement valid-define-options rule (#22107) (bab) - bb6f1b2 linter/vue: Implement require-slots-as-functions rule (#22244) (bab) - 5fa4774 linter/n: Implement `callback-return` rule (#22470) (Mikhail Baev) ### 🐛 Bug Fixes - 52bd016 linter: Respect allow unused disable directives in LSP (#22715) (camc314) - fa7c463 semantic: Correct TS enum member symbol spans (#22689) (camc314) - ed445ba linter: Respect flags overrides for `RegExp(/regex/i, "u")` (#22678) (Sysix) - 26b9396 semantic: Resolve parameter decorators outside parameter scope (#22623) (camc314) - 203952d linter: `no-misleading-character-class` fix `is_unicode_code_point_escape` check (#22655) (Sysix) - 2f64d3d linter: `no-misleading-character-class` own diagnostic for surrogate pairs without u flag (#22654) (Sysix) - 0c6ebc2 linter/eslint/no-lone-blocks: Do not flag empty loops (#22649) (Mikhail Baev) - 2a7562e linter/no-focused-tests: Mark fixer as a suggestion (#22645) (camc314) - dbe644f linter: Respect args none for unused rest parameters (#22627) (camc314) - d0211b0 linter: Allow undefined in DummyRuleMap index (#22626) (camc314) - 36fc0ec oxlint/lsp: "ignore this" actions merge with existing directive (#22604) (Sysix) - f7f370e linter/vitest/prefer-expect-type-of: Recommend `toBeTypeOf` instead of `expectTypeOf` (#22612) (Mikhail Baev) - 77063e5 linter/consistent-indexed-object-style: Preserve interface modifiers in fixes (#22579) (camc314) - 4f33aa7 linter: Treat `TSGlobalDeclaration` as ambient in `has_ambient_typescript_ancestor` (#22577) (camc314) ### ⚡ Performance - c22938d linter/no-async-endpoint-handlers: Populate node types (#22601) (camc314) - 607486e linter/no-negated-condition: Populate node types (#22602) (camc314) - 4dcaa59 linter/consistent-type-imports: Populate node types (#22600) (camc314) - 5bd3b25 linter/no-unused-vars: Avoid cloned ancestor iterator (#22598) (camc314) - 97fe9ba linter/no-extra-non-null-assertion: Reduce node matches (#22588) (camc314) - ae98296 linter/consistent-indexed-object-style: Populate node types (#22578) (camc314) # Oxfmt ### 🚀 Features - 16b8058 oxfmt: Support `vite-plus/resolveConfig` for vite.config.ts (#22454) (leaysgur) ### 🐛 Bug Fixes - 5a26479 formatter: Preserve import phases (#22692) (Cameron) ### ⚡ Performance - 78cf83f formatter: Pre-size output buffer using source text length (#22594) (Dunqing) Co-authored-by: Dunqing <29533304+Dunqing@users.noreply.github.com>
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
Printerconstructs its underlyingCodeBuffer(aVec<u8>) viaCodeBuffer::with_indent, which starts the buffer atcap = 0. For multi-MB outputs, that's ~log2(output_size)growth reallocations of a multi-MB buffer — each one allocates a bigger block, copies all existing bytes, and frees the old one.Add a
Printer::with_capacity(capacity, options, sorted_tailwind_classes)constructor and routeFormatted::print/print_with_indentthrough it withsource_text.len()as the hint. Formatted output is typically within a small factor of the input size, so a single pre-allocation eliminates the growth reallocs in one shot.Printer::newis kept (delegates towith_capacity(0, ...)) for back-compat with internal callers.Allocation impact
cargo allocs,allocs_formatter.snap:checker.tscal.com.tsxRadixUIAdoptionSection.jsxpdf.mjsantd.jsbinder.tsSys alloc counts unchanged. The 9-19 reductions per file match
log2(file_size)— exactly the doubling-growth pattern of the output buffer.Timing
Local
cargo bench --bench formattershows small wins (~3-7%) on tightly-measured benches (errors.ts, Search.tsx, core.js, handle-comments.js). Two files (next.ts, index.tsx) have wide Criterion intervals that swallow any signal in either direction. App.tsx (the biggest at 5.97 ms) shows essentially no change — modern allocators handle largeVec<u8>reallocations via mmap remap rather than physical copy, so the wins on big files turn out smaller than the realloc count suggests.CodSpeed (instruction-count-based) will give the deterministic answer.
Test Plan
cargo test -p oxc_formatter --lib --tests— 364 tests passcargo allocs— formatter snapshot shows clean reallocation reductionscargo bench --bench formatter— no regressions outside noiseSide fix
The commit hook flagged a pre-existing
clippy::redundant_pub_cratewarning incrates/oxc_parser/src/lexer/mod.rs:38(pub(crate) useinside a private module). One-character fix topub useincluded so the hook passes.AI disclosure: drafted with Claude Code, reviewed manually.