[code-infra] Port the benchmarks to new library

.circleci/config.yml

@@ -249,37 +249,40 @@ jobs:
     executor:
       name: code-infra/mui-node-browser
       playwright-img-version: v1.64.1-noble
+    # Building both sides is most of the setup time, and more logical CPUs leave the browser less to
+    # contend with while sampling. The `benchmark` CLI warns below 3.
+    resource_class: large
     steps:
       - checkout
       - code-infra/install-deps
       - run:
-          name: Build packages
-          command: pnpm release:build
-      - run:
-          name: Run baseline benchmarks at merge-base
-          environment:
-            PLAYWRIGHT_BROWSERS_PATH: /tmp/base-playwright-browsers
+          name: Resolve the baseline commit
+          # On a pull request the fork point, on master the previous commit.
           command: |
             set -e
-            if [ "$CIRCLE_BRANCH" = "master" ]; then
-              echo "On master branch; no baseline to compare against. Skipping baseline."
-              exit 0
-            fi
-            git fetch --no-tags origin master:refs/remotes/origin/master
-            BASE_SHA=$(git merge-base HEAD origin/master)
+            BASE_SHA=$(pnpm code-infra baseline --base-branch master)
             echo "Baseline commit: $BASE_SHA"
-            git worktree add /tmp/base-worktree "$BASE_SHA"
-            cd /tmp/base-worktree
-            pnpm install --frozen-lockfile
-            pnpm exec playwright install chromium
-            pnpm release:build
-            BENCHMARK_BRANCH=master pnpm test:benchmark
-            echo 'export BENCHMARK_BASELINE_PATH=/tmp/base-worktree/test/performance/benchmarks/results.json' >> "$BASH_ENV"
+            echo "$BASE_SHA" > .benchmark-baseline-sha
+      - restore_cache:
+          name: Restore the packed-baseline cache
+          keys:
+            - benchmark-packed-v1-{{ checksum ".benchmark-baseline-sha" }}
       - run:
-          name: Run head benchmarks
-          environment:
-            BENCHMARK_UPLOAD: 'true'
-          command: pnpm test:benchmark
+          name: Run benchmarks
+          no_output_timeout: 30m
+          # Only the smoke benchmarks, to keep a pull request's check short; the other benchmark files
+          # are left to a fuller run.
+          command: pnpm test:benchmark smoke --baseline "$(cat .benchmark-baseline-sha)"
+      - run:
+          name: Upload the benchmark report
+          # Also after a failed run, so the comment says what happened.
+          when: always
+          command: pnpm test:benchmark upload
+      - save_cache:
+          name: Save the packed-baseline cache
+          key: benchmark-packed-v1-{{ checksum ".benchmark-baseline-sha" }}
+          paths:
+            - test/benchmark/.benchmark/packed
 
 workflows:
   pipeline:

.gitignore

@@ -46,5 +46,6 @@ next-env.d.ts
 # Testing
 test-results/
 
-# Benchmark results
-test/performance/benchmarks/results.json
+# `benchmark` CLI output and generated pages
+.benchmark/
+test/benchmark/src/__bench__/

.lintignore

@@ -14,4 +14,6 @@ routeTree.gen.ts
 
 # TODO: fix lint violations and remove these ignores
 /CHANGELOG.old.md
-/CLAUDE.md
+/CLAUDE.md
+/test/benchmark/src/__bench__
+.benchmark

package.json

@@ -62,7 +62,7 @@
     "typescript": "tsc -b tsconfig.json",
     "types:check": "pnpm --filter @base-ui/react build && pnpm --filter @base-ui/react types:check",
     "inline-scripts": "tsx ./scripts/inlineScripts.mts",
-    "test:benchmark": "pnpm -F ./test/performance benchmark",
+    "test:benchmark": "pnpm -F ./test/benchmark benchmark",
     "use-react-18": "code-infra set-version-overrides --pkg react@18"
   },
   "devDependencies": {