ci(atlasbot): emit per-test case result metrics for flaky tracking

scripts/publish_test_metrics.py

@@ -54,6 +54,39 @@ def _load_junit(path: Path) -> dict[str, int]:
     return totals
 
 
+def _load_junit_cases(path: Path) -> list[tuple[str, str]]:
+    if not path.exists():
+        return []
+
+    tree = ET.parse(path)
+    root = tree.getroot()
+    suites: list[ET.Element]
+    if root.tag == "testsuite":
+        suites = [root]
+    elif root.tag == "testsuites":
+        suites = list(root.findall("testsuite"))
+    else:
+        suites = []
+
+    cases: list[tuple[str, str]] = []
+    for suite in suites:
+        for case in suite.findall("testcase"):
+            name = (case.attrib.get("name") or "").strip()
+            classname = (case.attrib.get("classname") or "").strip()
+            if not name:
+                continue
+            test_id = f"{classname}::{name}" if classname else name
+            status = "passed"
+            if case.find("failure") is not None:
+                status = "failed"
+            elif case.find("error") is not None:
+                status = "error"
+            elif case.find("skipped") is not None:
+                status = "skipped"
+            cases.append((test_id, status))
+    return cases
+
+
 def _load_coverage_percent(path: Path) -> float:
     if not path.exists():
         return 0.0
@@ -183,6 +216,7 @@ def main() -> int:
     build_dir = Path(os.getenv("BUILD_DIR", "build"))
 
     totals = _load_junit(junit_path)
+    test_cases = _load_junit_cases(junit_path)
     coverage_pct = _load_coverage_percent(coverage_path)
     gate_rc = _load_gate_rc(gate_rc_path)
     docs_rc = _load_gate_rc(docs_rc_path)
@@ -226,8 +260,13 @@ def main() -> int:
             "# TYPE platform_quality_gate_source_lines_over_500_total gauge",
             f'platform_quality_gate_source_lines_over_500_total{{suite="{suite}"}} {source_lines_over_500}',
             "# TYPE atlasbot_quality_gate_checks_total gauge",
+            "# TYPE platform_quality_gate_test_case_result gauge",
         ]
     ) + "\n"
+    payload += "\n".join(
+        f'platform_quality_gate_test_case_result{{suite="{suite}",test="{_escape_label(test_name)}",status="{_escape_label(test_status)}"}} 1'
+        for test_name, test_status in test_cases
+    ) + "\n"
     payload += "\n".join(
         f'atlasbot_quality_gate_checks_total{{suite="{suite}",check="{check_name}",result="{check_status}"}} 1'
         for check_name, check_status in checks.items()