.gitea/workflows/ci.yml

@@ -389,7 +389,12 @@ jobs:
             - name: Run ASV benchmark regression check via nox
               run: |
                   mkdir -p build
-                  nox -s benchmark_regression 2>&1 | tee build/nox-benchmark-output.log
+                  # --quick: skip per-benchmark calibration/repeat/process
+                  # averaging (asv default). This is a fast, informational
+                  # trend check (continue-on-error), not the accurate,
+                  # fully-calibrated run used for real performance
+                  # investigation (nox -s benchmark, no --quick).
+                  nox -s benchmark_regression -- --quick 2>&1 | tee build/nox-benchmark-output.log
 
             - name: Upload benchmark log artifact
               if: always()

asv.conf.json

@@ -11,13 +11,16 @@
     "env_dir": ".asv/env",
     "results_dir": ".asv/results",
     "html_dir": ".asv/html",
+    "matrix": {
+        "uv": [""]
+    },
     "build_command": [
-        "python -m build --wheel -o {build_cache_dir} {build_dir}"
+        "uv build --python {env_dir}/bin/python --wheel -o {build_cache_dir} {build_dir}"
     ],
     "install_command": [
-        "python -mpip install {wheel_file}"
+        "uv pip install --python {env_dir}/bin/python {wheel_file} --force-reinstall"
     ],
     "uninstall_command": [
-        "return-code=any python -mpip uninstall -y {project}"
+        "return-code=any uv pip uninstall --python {env_dir}/bin/python {project}"
     ]
 }

noxfile.py

@@ -364,7 +364,21 @@ def benchmark_regression(session: nox.Session) -> None:
     base = os.environ.get("BENCHMARK_BASE", "origin/master")
     session.run("asv", "machine", "--yes")
     session.run(
-        "asv", "continuous", base, "HEAD", "--show-stderr", success_codes=[0, 2]
+        "asv",
+        "continuous",
+        base,
+        "HEAD",
+        "--show-stderr",
+        *session.posargs,
+        # asv continuous returns the boolean `worsened` as its exit code:
+        # 0 = no significant regression, 1 = a benchmark measurably
+        # worsened past --factor. This session should fail on 1 so the
+        # regression is visible (the CI job's continue-on-error: true is
+        # what keeps this from blocking the PR, not this success list).
+        # There is no exit code 2 in this asv version; the prior [0, 2]
+        # never matched the real "worsened" code and made this session
+        # fail unconditionally on any regression signal.
+        success_codes=[0],
     )