From 8374ec937e6fd868636e468877a9ea8c1dded19d Mon Sep 17 00:00:00 2001
From: StepFun Agent <stepfun-agent@hermes.ai>
Date: Sun, 26 Apr 2026 09:58:37 -0400
Subject: [PATCH] fix(perf-bottleneck): make find_slow_tests_pytest functional;
 unblock pytest collection

- find_slow_tests_pytest: actually run pytest with --durations and parse stdout
  instead of reading stale cache file. Now measures real test durations.
- test_pr_complexity_scorer.py: wrap module-level sys.exit() in
  if __name__ == "__main__" guard. This was crashing pytest collection
  entirely, making the repo un-testable and masking all other performance
  analysis.

Both issues were blocking the Performance Bottleneck Finder (#171) from
working. The script now successfully runs pytest, collects duration data,
and produces accurate reports.

Closes #171
---
 scripts/perf_bottleneck_finder.py    | 53 ++++++++++++++--------------
 scripts/test_pr_complexity_scorer.py |  5 +--
 2 files changed, 30 insertions(+), 28 deletions(-)

diff --git a/scripts/perf_bottleneck_finder.py b/scripts/perf_bottleneck_finder.py
index 6577d51..3597fe9 100644
--- a/scripts/perf_bottleneck_finder.py
+++ b/scripts/perf_bottleneck_finder.py
@@ -70,37 +70,38 @@ class PerfReport:
 # ── Test Analysis ──────────────────────────────────────────────────
 
 def find_slow_tests_pytest(repo_path: str) -> List[Bottleneck]:
-    """Run pytest --durations and parse slow tests."""
+    """Run pytest with --durations and parse slow test output."""
     bottlenecks = []
 
-    # Try to run pytest with durations
     try:
+        # Run pytest to get slowest tests; maxfail=1 avoids hanging on failures
         result = subprocess.run(
-            ["python3", "-m", "pytest", "--co", "-q", "--durations=0"],
-            cwd=repo_path, capture_output=True, text=True, timeout=30
+            ["python3", "-m", "pytest", "-q",
+             f"--durations={PYTEST_DURATIONS_COUNT}", "--tb=no", "--maxfail=1"],
+            cwd=repo_path, capture_output=True, text=True, timeout=60
         )
-        # If tests exist, try to get durations from last run
-        durations_file = os.path.join(repo_path, ".pytest_cache", "v", "cache", "durations")
-        if os.path.exists(durations_file):
-            with open(durations_file) as f:
-                for line in f:
-                    parts = line.strip().split()
-                    if len(parts) >= 2:
-                        try:
-                            duration = float(parts[0])
-                            test_name = " ".join(parts[1:])
-                            if duration > SLOW_TEST_THRESHOLD_S:
-                                severity = "critical" if duration > 10 else "warning"
-                                bottlenecks.append(Bottleneck(
-                                    category="test",
-                                    name=test_name,
-                                    duration_s=duration,
-                                    severity=severity,
-                                    recommendation=f"Test takes {duration:.1f}s. Consider mocking slow I/O, using fixtures, or marking with @pytest.mark.slow."
-                                ))
-                        except ValueError:
-                            continue
-    except (subprocess.TimeoutExpired, FileNotFoundError):
+        # Parse durations from stdout.
+        # Lines look like: "  3.45s call     test_file.py::test_name"
+        for line in result.stdout.splitlines():
+            line = line.strip()
+            m = re.match(r'^(\d+\.?\d*)s\s+(call|setup|teardown)\s+(.+)$', line)
+            if not m:
+                continue
+            try:
+                duration = float(m.group(1))
+                test_name = m.group(3).strip()
+                if duration > SLOW_TEST_THRESHOLD_S:
+                    severity = "critical" if duration > 10 else "warning"
+                    bottlenecks.append(Bottleneck(
+                        category="test",
+                        name=test_name,
+                        duration_s=duration,
+                        severity=severity,
+                        recommendation=f"Test takes {duration:.1f}s. Consider mocking slow I/O, using fixtures, or marking with @pytest.mark.slow."
+                    ))
+            except ValueError:
+                continue
+    except (subprocess.TimeoutExpired, FileNotFoundError, PermissionError):
         pass
 
     return bottlenecks
diff --git a/scripts/test_pr_complexity_scorer.py b/scripts/test_pr_complexity_scorer.py
index 4e98774..c3cafa2 100644
--- a/scripts/test_pr_complexity_scorer.py
+++ b/scripts/test_pr_complexity_scorer.py
@@ -166,5 +166,6 @@ def _():
         assert_true(s in TIME_PER_POINT, f"Missing time for score {s}")
 
 
-print(f"\n=== Results: {PASS} passed, {FAIL} failed ===")
-sys.exit(0 if FAIL == 0 else 1)
+if __name__ == "__main__":
+    print(f"\n=== Results: {PASS} passed, {FAIL} failed ===")
+    sys.exit(0 if FAIL == 0 else 1)