perf(js-plugins): optimize CFG walker with SoA pattern and custom traverser by re-taro · Pull Request #17423 · oxc-project/oxc

re-taro · 2025-12-28T14:13:29Z

Summary

Optimize walkProgramWithCfg in apps/oxlint/src-js/plugins/cfg.ts by applying multiple performance improvements based on TODO comments in the original code

Close #17232

Replace VisitNodeStep/CallMethodStep classes from @eslint/plugin-kit with plain number constants
Pre-compute type IDs in the first pass to eliminate Map lookups in the second pass
Use Struct of Arrays (SoA) pattern for step storage to improve memory locality and V8 optimization
Replace ESLint's Traverser with a lightweight custom traverseNode function that avoids unnecessary overhead

Changes

Optimization	Before	After
Step encoding	`VisitNodeStep`/`CallMethodStep` classes	Plain constants `STEP_TYPE_*`
Type ID lookup	Map lookup on every step in 2nd pass	Pre-computed in 1st pass
Data structure	Array of Structs (`steps[]`)	Struct of Arrays (`stepTypes[]`, `stepTargets[]`, `stepTypeIds[]`, `stepArgs[]`)
AST traverser	ESLint's `Traverser.traverse()`	Custom `traverseNode()` using `visitorKeys`

Removed dependencies

@eslint/plugin-kit (no longer needed)
ESLint's internal Traverser module

Copilot

Pull request overview

This PR optimizes the walkProgramWithCfg function by replacing ESLint's data structures and traverser with more efficient custom implementations. The optimization focuses on reducing object allocations, improving memory locality, and eliminating redundant lookups.

Key changes:

Replaced class-based step encoding with plain numeric constants and Struct of Arrays (SoA) pattern
Pre-computed type IDs during step building to eliminate Map lookups during AST walking
Implemented a lightweight custom traverseNode function to replace ESLint's Traverser

Reviewed changes

Copilot reviewed 2 out of 3 changed files in this pull request and generated 4 comments.

File	Description
apps/oxlint/src-js/plugins/cfg.ts	Core optimization implementation: replaced step classes with SoA pattern, added custom traverser, pre-computed type IDs
apps/oxlint/package.json	Removed unused `@eslint/plugin-kit` dependency

Files not reviewed (1)

pnpm-lock.yaml: Language not supported

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2025-12-28T14:50:24Z

apps/oxlint/src-js/plugins/cfg.ts

+  if (isArray(node)) {
+    const len = node.length;
+    for (let i = 0; i < len; i++) {
+      traverseNode(node[i], enter, leave);
+    }
+    return;
+  }


The isArray check treats arrays as nodes to traverse recursively, but this will not call enter/leave callbacks on array elements that are actual AST nodes. The function signature accepts Node | null | undefined, not arrays, so this check appears to handle child node arrays incorrectly. Arrays should be iterated and their elements passed to traverseNode, but the type signature should clarify this or the logic should ensure enter/leave are called appropriately for node elements.

apps/oxlint/src-js/plugins/cfg.ts

Copilot · 2025-12-28T14:50:25Z

apps/oxlint/src-js/plugins/cfg.ts

+              for (const [name, id] of NODE_TYPE_IDS_MAP) {
+                if (id === eventTypeId) {
+                  eventNames.push(name);
+                  break;
+                }
+              }


Reverse lookup iterating through the entire NODE_TYPE_IDS_MAP is inefficient in debug builds. Consider creating a reverse map (ID to name) at module initialization for O(1) lookups instead of O(n) iteration for each event.

…verser Apply multiple performance optimizations to `walkProgramWithCfg`: - Replace `VisitNodeStep`/`CallMethodStep` classes with plain number constants - Pre-compute type IDs in first pass to eliminate Map lookups in second pass - Use Struct of Arrays (SoA) pattern for step storage to improve memory locality - Replace ESLint's Traverser with lightweight custom `traverseNode` function - Remove unused `@eslint/plugin-kit` and ESLint Traverser imports

overlookmotel

Thank you for working on this.

I'm on holiday for 10 days, so I'm afraid I won't be able to review properly for some time. Sorry for delay.

But a couple of notes in meantime:

Would it be possible to break this up into multiple PRs, split (roughly) along the lines of the points in the original TODO comment? That'd make this easier to review and get parts of it merged while we discuss other parts. I'm not sure from first glance if AI review suggesting flaws in new traversal impl are valid or not.

You can make a "stack" by basing each PR on the previous PR. I can handle rebasing each onto main before merging.

I think the SOA arrays can probably be cut down from 4 to 2:

stepTargets and stepArgs could be combined. When step type is "enter node" or "exit node", the array element would be the node object. When it's a CFG event, the array element would be an array of args to call the visitor function with.
typeId and step type could be combined by using typeId unchanged for "enter node" and typeId + N for "exit node" (where N is some number greater than maximum type ID e.g. 256).

overlookmotel · 2026-01-20T14:51:54Z

@re-taro I'm back from being away. Wondered if you have an interest in continuing with this?

re-taro · 2026-01-20T15:15:03Z

@overlookmotel

Yes! I am interested!! However, I am quite busy myself and it seems I can respond starting this weekend.

overlookmotel · 2026-01-20T18:46:05Z

That's great to hear! Please ping me when you've had time to make some progress.

re-taro · 2026-01-25T16:15:29Z

Hi @overlookmotel, thanks for your feedback!

I've split the original PR into 3 separate PRs as you suggested, following the TODO comments:

perf(linter/plugins): replace ESLint step classes with plain objects #18527 - Replace ESLint step classes with plain objects + merge kind/phase into single type
perf(linter/plugins): use SoA pattern with 2 arrays for CFG steps #18528 - Use SoA pattern with 2 arrays (as you suggested: stepTypeIds + stepData)
perf(linter/plugins): replace ESLint Traverser with lightweight traverseNode #18529 - Replace ESLint Traverser with lightweight traverseNode

Since GitHub doesn't support PR stacks across forks, I created them as independent PRs targeting main. They should be merged in order (1 → 2 → 3). I'll rebase each one after the previous is merged.

I also incorporated your suggestion to reduce the SoA arrays from 4 to 2:

Combined stepTargets and stepArgs into stepData
Combined typeId and step type using offset encoding (enter = typeId, exit = typeId + 256)

Let me know if you'd like any changes!

overlookmotel · 2026-01-26T14:22:31Z

Superceded by the 3 PRs listed above.

github-actions bot added A-linter Area - Linter A-cli Area - CLI A-linter-plugins Area - Linter JS plugins C-performance Category - Solution not expected to change functional behavior, only performance labels Dec 28, 2025

re-taro force-pushed the refactor/linter-cfg branch from 19b0bc3 to 71a255d Compare December 28, 2025 14:48

re-taro marked this pull request as ready for review December 28, 2025 14:49

re-taro requested review from camc314 and overlookmotel as code owners December 28, 2025 14:49

Copilot AI review requested due to automatic review settings December 28, 2025 14:49

Copilot AI reviewed Dec 28, 2025

View reviewed changes

re-taro force-pushed the refactor/linter-cfg branch 2 times, most recently from c8a6284 to 31dc842 Compare December 28, 2025 15:06

re-taro force-pushed the refactor/linter-cfg branch from 31dc842 to cf1fc5f Compare December 28, 2025 15:17

overlookmotel reviewed Jan 4, 2026

View reviewed changes

overlookmotel self-assigned this Jan 20, 2026

Merge branch 'oxc-project:main' into refactor/linter-cfg

add4a3c

overlookmotel closed this Jan 26, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

perf(js-plugins): optimize CFG walker with SoA pattern and custom traverser#17423

perf(js-plugins): optimize CFG walker with SoA pattern and custom traverser#17423
re-taro wants to merge 2 commits intooxc-project:mainfrom
re-taro:refactor/linter-cfg

re-taro commented Dec 28, 2025 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Dec 28, 2025

Uh oh!

Uh oh!

Uh oh!

Copilot AI Dec 28, 2025

Uh oh!

overlookmotel left a comment •

edited

Loading

Uh oh!

overlookmotel commented Jan 20, 2026

Uh oh!

re-taro commented Jan 20, 2026

Uh oh!

overlookmotel commented Jan 20, 2026

Uh oh!

re-taro commented Jan 25, 2026

Uh oh!

overlookmotel commented Jan 26, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

re-taro commented Dec 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes

Removed dependencies

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Dec 28, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Copilot AI Dec 28, 2025

Choose a reason for hiding this comment

Uh oh!

overlookmotel left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

overlookmotel commented Jan 20, 2026

Uh oh!

re-taro commented Jan 20, 2026

Uh oh!

overlookmotel commented Jan 20, 2026

Uh oh!

re-taro commented Jan 25, 2026

Uh oh!

overlookmotel commented Jan 26, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

re-taro commented Dec 28, 2025 •

edited

Loading

overlookmotel left a comment •

edited

Loading