Various performance baselines for typescript-eslint.
You'll need hyperfine installed locally, such as with brew install hyperfine or winget install hyperfine.
See sharkdp/hyperfine#installation.
npm install
npm run generate
npm run measureYou can manually measure individual cases by running hyperfine ../../node_modules/eslint/bin/eslint.js --ignore-failure --warmup 1.
The caseEntries values in src/data.ts can be modified to test:
files: roughly how many generated files should be lintedlayout: what rough shape of imports those files exhibit:"even": a single root-levelindex.tsimporting from roughly an even triangle shape of files"references": a single root-leveltsconfig.jsonwith project references to a few projects"wide": one root-levelindex.tsimporting from all files in the project
singleRun: whether to enable single-run inference as a performance boosttypes: whether to useparserOptions.projectorparserOptions.projectServicefor typed linting
Right now, parserOptions.project with single-run inference outperforms parserOptions.projectService.
This is a performance issue and we are investigating it as a critical bug for v8.
┌───────┬───────────────────────┬───────────────────────┐
│ files │ project (even layout) │ service (even layout) │
┼───────┼───────────────────────┼───────────────────────┤
│ 1024 │ '1.750 s ± 0.008 s' │ '2.473 s ± 0.011 s' │
┴───────┴───────────────────────┴───────────────────────┘
See typescript-eslint/typescript-eslint#9571 Performance: parserOptions.projectService no longer outperforms parserOptions.project in typescript-eslint. Also see the 📌 pinned issues later in this file.
- Example measurements taken on an M1 Max Mac Studio with Node.js 22.4.1
- These results are similar across TypeScript versions: 5.0.4, 5.4.5, and 5.5.3
The traces/ directory contains more specific traces for investigations.
✨ You might consider using 0x for nice flamegraph visuals.
All comparisons were run on a common shape of linting: 1024 files with the "even" (triangle-shaped) imports layout.
📌 Filed on typescript-eslint as ⚡ Performance: Overhead of populateGlobalsFromLib in scope-manager.
This trace shows the impact of @typescript-eslint/scope-manager's populateGlobalsFromLib.
See traces/globals-scope-manager/:
baseline.cpuprofile: Baseline measurement with no changesskipping.cpuprofile: Commenting out the contents ofpopulateGlobalsFromLib
They were generated with:
cd files-1024-layout-even-singlerun-true-types-service
node --cpu-prof --cpu-prof-interval=100 --cpu-prof-name=baseline.cpuprofile ../../node_modules/eslint/bin/eslint.js
# clear ../../node_modules/@typescript-eslint/scope-manager/dist/referencer/Referencer.js > populateGlobalsFromLib
node --cpu-prof --cpu-prof-interval=100 --cpu-prof-name=skipping.cpuprofile ../../node_modules/eslint/bin/eslint.jsHyperfine measurements show a ~20% improvement in lint time:
| Variant | Measurement | User Time |
|---|---|---|
| Baseline | 3.137 s ± 0.024 s | 4.417 s |
| Skipping | 2.477 s ± 0.014 s | 3.501 s |
This is a preliminary trace to start debugging their differences.
See traces/Project 1 - Service 2.cpuprofile.
- Trace #1:
parserOptions.project - Trace #2:
parserOptions.projectService
It was generated with:
cd cases/files-1024-layout-even-singlerun-true-types-project
node --inspect-brk ../../node_modules/eslint/bin/eslint.js
cd ../files-1024-layout-even-singlerun-true-types-service
node --inspect-brk ../../node_modules/eslint/bin/eslint.jsComparing equivalent code paths:
| Code Path | Project | Service |
|---|---|---|
All verifyTexts |
2040ms | 2859ms |
parseForESLint |
993ms | 1090ms |
📌 Filed on TypeScript as ⚡ Performance: Project service spends excess time cleaning client files when called synchronously.
This comparison shows the cost of the TypeScript project service calling cleanupProjectsAndScriptInfos.
See traces/service-file-cleanup/:
baseline.cpuprofile: Baseline measurement with no changesskipping.cpuprofile: Commenting out the contents of TypeScript'scleanupProjectsAndScriptInfos
They were generated with:
cd files-1024-layout-even-singlerun-true-types-service
node --cpu-prof --cpu-prof-interval=100 --cpu-prof-name=baseline.cpuprofile ../../node_modules/eslint/bin/eslint.js
# clear ../../node_modules/typescript/lib/typescript.js > cleanupProjectsAndScriptInfos
node --cpu-prof --cpu-prof-interval=100 --cpu-prof-name=skipping.cpuprofile ../../node_modules/eslint/bin/eslint.jsHyperfine measurements show a ~15-20% improvement in lint time:
| Variant | Measurement | User Time |
|---|---|---|
| Baseline | 3.215 s ± 0.041 s | 4.483 s |
| Skipping | 2.501 s ± 0.017 s | 3.758 s |
📌 Filed on TypeScript as ⚡ Performance: Project service doesn't cache all fs.statSync.
This comparison shows the cost uncached fs.statSync calls inside the project service.
See traces/service-uncached-stats/:
baseline.cpuprofile: Baseline measurement with no changescaching.cpuprofile: Adding a cachingMapto TypeScript'sstatSync
They were generated with:
cd files-1024-layout-even-singlerun-true-types-service
node --cpu-prof --cpu-prof-interval=100 --cpu-prof-name=baseline.cpuprofile ../../node_modules/eslint/bin/eslint.js
# edit ../../node_modules/typescript/lib/typescript.js > statSync (see diff below)
node --cpu-prof --cpu-prof-interval=100 --cpu-prof-name=caching.cpuprofile ../../node_modules/eslint/bin/eslint.jsdiff patch to switch to the Caching variant...
diff --git a/node_modules/typescript/lib/typescript.js b/node_modules/typescript/lib/typescript.js
index 4baad59..44639d5 100644
--- a/node_modules/typescript/lib/typescript.js
+++ b/node_modules/typescript/lib/typescript.js
@@ -8546,9 +8546,15 @@ var sys = (() => {
}
}
};
+ const statCache = new Map();
return nodeSystem;
function statSync(path) {
- return _fs.statSync(path, { throwIfNoEntry: false });
+ if (statCache.has(path)) {
+ return statCache.get(path);
+ }
+ const result = _fs.statSync(path, { throwIfNoEntry: false });
+ statCache.set(path, result);
+ return result;
}
function enableCPUProfiler(path, cb) {
if (activeSession) {Hyperfine measurements show a ~7-12% improvement in lint time:
| Variant | Measurement | User Time |
|---|---|---|
| Baseline | 3.112 s ± 0.033 s | 4.382 |
| Caching | 2.740 s ± 0.030 s | 4.032 |
📌 Filed on TypeScript as ⚡ Performance: Project service doesn't cache all fs.realpath.
This comparison shows the cost uncached fs.realpath calls inside the project service.
See traces/service-uncached-realpaths/:
baseline.cpuprofile: Baseline measurement with no changescaching.cpuprofile: Adding a cachingMapto TypeScript'srealpath
They were generated with:
cd files-1024-layout-even-singlerun-true-types-service
node --cpu-prof --cpu-prof-interval=100 --cpu-prof-name=baseline.cpuprofile ../../node_modules/eslint/bin/eslint.js
# edit ../../node_modules/typescript/lib/typescript.js > realpath (see diff below)
node --cpu-prof --cpu-prof-interval=100 --cpu-prof-name=caching.cpuprofile ../../node_modules/eslint/bin/eslint.jsdiff patch to switch to the Caching variant...
diff --git a/node_modules/typescript/lib/typescript.js b/node_modules/typescript/lib/typescript.js
index 4baad59..e53476d 100644
--- a/node_modules/typescript/lib/typescript.js
+++ b/node_modules/typescript/lib/typescript.js
@@ -13,6 +13,8 @@ See the Apache Version 2.0 License for specific language governing permissions
and limitations under the License.
***************************************************************************** */
+var realpathCache = new Map();
+
var ts = {}; ((module) => {
"use strict";
var __defProp = Object.defineProperty;
@@ -8798,6 +8800,15 @@ var sys = (() => {
return path.length < 260 ? _fs.realpathSync.native(path) : _fs.realpathSync(path);
}
function realpath(path) {
+ const cached = realpathCache.get(path);
+ if (cached) {
+ return cached;
+ }
+ const result = realpathWorker(path);
+ realpathCache.set(path, result);
+ return result;
+ }
+ function realpathWorker(path) {
try {
return fsRealpath(path);
} catch {Hyperfine measurements with --runs 50 show a ~0.5-2.5% improvement in lint time:
| Variant | Measurement | User Time |
|---|---|---|
| Baseline | 3.153 s ± 0.039 s | 4.403 s |
| Caching | 3.073 s ± 0.048 s | 4.377 s |
 Jake Bailey 🤔 |
 Josh Goldberg ✨ 🤔 🚇 🚧 📆 🔧 |