SOCIAL_MEDIA +152.4%

HIGHLY_CONNECTED    +56.6%
PHARMA              +54.6%
MEDICAL_RECORDS      -2.0%
ENGINEERING         -26.7%
TRAIN               -54.6%
LIBRARY             -55.0%
ELECTRICAL_GRID     -57.9%

Author: hmottestad

.codex/skills/jmh-benchmark-compare/scripts/jmh_compare_core.py

@@ -17,6 +17,11 @@
     re.IGNORECASE,
 )
 METRIC_COLUMNS = {"Score", "Error", "Cnt"}
+JMH_MODES = {"thrpt", "avgt", "sample", "ss", "all"}
+STRICT_NUM_RE = re.compile(
+    r"[-+]?(?:(?:\d+(?:,\d{3})*|\d+)(?:\.\d+)?|\.\d+)(?:[eE][-+]?\d+)?|[-+]?(?:inf|nan)",
+    re.IGNORECASE,
+)
 DATE_PATTERNS = (
     re.compile(
         r"(20\d{2})[-_]?([01]\d)[-_]?([0-3]\d)[Tt _-]?([0-2]\d)[-_:]?([0-5]\d)(?:[-_:]?([0-5]\d))?"
@@ -123,17 +128,26 @@ def is_int_token(text: str) -> bool:
     return bool(re.fullmatch(r"[+-]?\d+", text or ""))
 
 
+def is_numeric_metric_token(text: str) -> bool:
+    value = (text or "").strip()
+    if value.endswith("±"):
+        value = value[:-1].strip()
+    return bool(STRICT_NUM_RE.fullmatch(value))
+
+
 def has_valid_metric_values(row: Dict[str, str], columns: Sequence[str]) -> bool:
+    if "Mode" in columns and row.get("Mode", "").strip().lower() not in JMH_MODES:
+        return False
     for col in columns:
         value = row.get(col, "")
         if col == "Score":
-            if extract_numeric(value) is None:
+            if not is_numeric_metric_token(value):
                 return False
         elif col == "Cnt" and value:
             if not is_int_token(value):
                 return False
         elif col == "Error" and value:
-            if extract_numeric(value) is None:
+            if not is_numeric_metric_token(value):
                 return False
     return True
 
@@ -227,8 +241,6 @@ def parse_file(path: Path, label: str, id_columns: Optional[str], timestamp_sour
     for line in lines[header_idx + 1 :]:
         stripped = line.strip()
         if not stripped:
-            if saw_data:
-                break
             continue
         if stripped.startswith("#"):
             continue
@@ -244,6 +256,10 @@ def parse_file(path: Path, label: str, id_columns: Optional[str], timestamp_sour
             if saw_data:
                 break
             continue
+        if not has_valid_metric_values(row, columns):
+            if saw_data:
+                break
+            continue
         score = extract_numeric(row.get("Score", ""))
         if score is None:
             if saw_data and (stripped.startswith("Result") or stripped.startswith("Secondary result")):

.codex/skills/jmh-benchmark-compare/scripts/test_jmh_compare_core.py

@@ -20,7 +20,7 @@ def test_missing_cnt_and_error_values_do_not_shift_score(self) -> None:
         repo_root = SCRIPT_DIR.parents[3]
         result_file = (
             repo_root
-            / "core/sail/lmdb/src/test/java/org/eclipse/rdf4j/sail/lmdb/benchmark/results-2026-03-01.md"
+            / "core/sail/lmdb/src/test/java/org/eclipse/rdf4j/sail/lmdb/benchmark/theme-query-benchmark-results/results-2026-03-01.md"
         )
 
         parsed = core.parse_file(result_file, "results-2026-03-01", None, "mtime")
@@ -33,7 +33,7 @@ def test_plus_minus_error_rows_keep_score_numeric(self) -> None:
         repo_root = SCRIPT_DIR.parents[3]
         result_file = (
             repo_root
-            / "core/sail/lmdb/src/test/java/org/eclipse/rdf4j/sail/lmdb/benchmark/results-2026-03-04.md"
+            / "core/sail/lmdb/src/test/java/org/eclipse/rdf4j/sail/lmdb/benchmark/theme-query-benchmark-results/results-2026-03-04.md"
         )
 
         parsed = core.parse_file(result_file, "results-2026-03-04", None, "mtime")
@@ -74,6 +74,58 @@ def test_compare_uses_column_names_when_key_order_differs(self) -> None:
             self.assertAlmostEqual(row["Score [right]"], 20.0, places=3)
             self.assertAlmostEqual(row["Diff % [right - left]"], 100.0, places=3)
 
+    def test_blank_lines_between_jmh_rows_do_not_end_table(self) -> None:
+        results = "\n".join(
+            [
+                "Benchmark  (themeName)  (z_queryIndex)  Mode  Score  Units",
+                "ThemeQueryBenchmark.executeQuery  MEDICAL_RECORDS  0  avgt  10.0  ms/op",
+                "",
+                "ThemeQueryBenchmark.executeQuery  SOCIAL_MEDIA  8  avgt  20.0  ms/op",
+            ]
+        )
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            result_file = Path(tmpdir) / "results.txt"
+            result_file.write_text(results, encoding="utf-8")
+
+            parsed = core.parse_file(result_file, "results", None, "mtime")
+
+            self.assertEqual(len(parsed.rows), 2)
+            key = ("ThemeQueryBenchmark.executeQuery", "SOCIAL_MEDIA", "8", "avgt", "ms/op")
+            self.assertIn(key, parsed.score_by_key)
+            self.assertAlmostEqual(parsed.score_by_key[key], 20.0, places=3)
+
+    def test_non_jmh_text_after_blank_does_not_parse_as_rows(self) -> None:
+        results = "\n".join(
+            [
+                "Benchmark                              (themeName)  (z_queryIndex)  Mode  Cnt      Score   Error  Units",
+                "ThemeQueryBenchmark.executeQuery   MEDICAL_RECORDS               0  avgt          10.0          ms/op",
+                "",
+                "ThemeQueryBenchmark.executeQuery      SOCIAL_MEDIA               8  avgt          20.0          ms/op",
+                "",
+                "Initializing state: k=64, subjectBuckets=4096, predicateBuckets=64, "
+                "objectBuckets=4096, contextBuckets=16, contextPairSketchesEnabled=false",
+                "Projection (resultSizeActual=1, hasNextCallCountActual=2)",
+            ]
+        )
+
+        with tempfile.TemporaryDirectory() as tmpdir:
+            result_file = Path(tmpdir) / "results.txt"
+            result_file.write_text(results, encoding="utf-8")
+
+            parsed = core.parse_file(result_file, "results", None, "mtime")
+
+            self.assertEqual(
+                [
+                    row["Benchmark"]
+                    for row in parsed.rows
+                ],
+                [
+                    "ThemeQueryBenchmark.executeQuery",
+                    "ThemeQueryBenchmark.executeQuery",
+                ],
+            )
+
 
 if __name__ == "__main__":
     unittest.main()