prizmkit 1.1.140 → 1.1.142

package/bundled/dev-pipeline/prizmkit_runtime/test_result.py

@@ -0,0 +1,160 @@
+"""Validation for prizmkit-test terminal result and report artifacts."""
+
+from __future__ import annotations
+
+import json
+import re
+from pathlib import Path
+from typing import Any
+
+TEST_RESULTS = {"TEST_PASS", "TEST_NEEDS_FIXES", "TEST_BLOCKED"}
+INDEPENDENT_REVIEW_STATUSES = {"completed", "downgraded", "not_applicable"}
+_RESULT_LINE_PATTERN = re.compile(
+    r"(?m)^- Result:\s*(TEST_PASS|TEST_NEEDS_FIXES|TEST_BLOCKED)\s*$"
+)
+_BARE_RESULT_LINE_PATTERN = re.compile(
+    r"(?m)^(TEST_PASS|TEST_NEEDS_FIXES|TEST_BLOCKED)\s*$"
+)
+_RESULT_KEYS = {
+    "schema_version",
+    "result",
+    "report",
+    "main_review_rounds",
+    "independent_review",
+    "repair_rounds",
+    "production_changed",
+    "review_required",
+    "review_scope",
+    "unresolved_items",
+}
+_INDEPENDENT_REVIEW_KEYS = {
+    "status",
+    "responses",
+    "downgrade_reason",
+    "final_state_rechecked",
+}
+
+
+def load_test_result(path: str | Path) -> dict[str, Any] | None:
+    """Load one valid terminal projection, or return None."""
+    try:
+        value = json.loads(Path(path).read_text(encoding="utf-8"))
+    except (OSError, ValueError, TypeError):
+        return None
+    if not isinstance(value, dict):
+        return None
+    if set(value) != _RESULT_KEYS or value.get("schema_version") != 1:
+        return None
+    if value.get("result") not in TEST_RESULTS or value.get("report") != "test-report.md":
+        return None
+
+    rounds = value.get("main_review_rounds")
+    repairs = value.get("repair_rounds")
+    if not _integer_in_range(rounds, 0, 10) or not _integer_in_range(repairs, 0, 3):
+        return None
+
+    independent = value.get("independent_review")
+    if not isinstance(independent, dict) or set(independent) != _INDEPENDENT_REVIEW_KEYS:
+        return None
+    status = independent.get("status")
+    responses = independent.get("responses")
+    if status not in INDEPENDENT_REVIEW_STATUSES or not _integer_in_range(responses, 0, 5):
+        return None
+    if not isinstance(independent.get("final_state_rechecked"), bool):
+        return None
+    downgrade_reason = independent.get("downgrade_reason")
+    if status == "downgraded":
+        if not isinstance(downgrade_reason, str) or not downgrade_reason.strip():
+            return None
+        if responses != 0 or independent["final_state_rechecked"]:
+            return None
+    elif downgrade_reason is not None:
+        return None
+    if status == "completed":
+        if responses < 1 or not independent["final_state_rechecked"]:
+            return None
+    if status == "not_applicable":
+        if responses != 0 or independent["final_state_rechecked"]:
+            return None
+
+    if not isinstance(value.get("production_changed"), bool):
+        return None
+    if not isinstance(value.get("review_required"), bool):
+        return None
+    expected_scope = "delta" if value["review_required"] else None
+    if value.get("review_scope") != expected_scope:
+        return None
+    unresolved = value.get("unresolved_items")
+    if not isinstance(unresolved, list) or any(
+        not isinstance(item, str) or not item.strip() for item in unresolved
+    ):
+        return None
+
+    if value["result"] == "TEST_PASS":
+        if value["review_required"] or unresolved or rounds < 1:
+            return None
+    return value
+
+
+def valid_test_result(path: str | Path, *, required_result: str | None = None) -> bool:
+    """Return whether the terminal projection is valid and has the required result."""
+    value = load_test_result(path)
+    return value is not None and (
+        required_result is None or value.get("result") == required_result
+    )
+
+
+def valid_test_artifact_pair(
+    report_path: str | Path,
+    result_path: str | Path,
+    *,
+    required_result: str | None = None,
+) -> bool:
+    """Validate one sibling report/result pair and their terminal result."""
+    report = Path(report_path)
+    result = Path(result_path)
+    if not report.is_file() or not result.is_file():
+        return False
+    if report.resolve().parent != result.resolve().parent:
+        return False
+    projection = load_test_result(result)
+    if projection is None:
+        return False
+    if required_result is not None and projection.get("result") != required_result:
+        return False
+    try:
+        report_text = report.read_text(encoding="utf-8")
+    except OSError:
+        return False
+    final_result = _report_final_result(report_text)
+    return (
+        report_text.startswith("# Test Report")
+        and final_result == projection["result"]
+    )
+
+
+def _report_final_result(report: str) -> str | None:
+    """Extract one unambiguous result from a supported terminal section."""
+    declared_results: list[str] = []
+    for match in re.finditer(
+        r"(?ms)^## (?P<title>Final Result|Terminal Result|Final Decision)\s*$\n"
+        r"(?P<body>.*?)(?=^## |\Z)",
+        report,
+    ):
+        body = match.group("body")
+        result_lines = _RESULT_LINE_PATTERN.findall(body)
+        if match.group("title") == "Final Decision":
+            result_lines.extend(_BARE_RESULT_LINE_PATTERN.findall(body))
+        declared_results.extend(result_lines)
+
+    if not declared_results or len(set(declared_results)) != 1:
+        return None
+    return declared_results[0]
+
+
+def _integer_in_range(value: object, minimum: int, maximum: int) -> bool:
+    return (
+        isinstance(value, int)
+        and not isinstance(value, bool)
+        and minimum <= value <= maximum
+    )

package/bundled/dev-pipeline/scripts/check-session-status.py

@@ -150,7 +150,9 @@ def determine_status(data):
     """Determine the single-line status string from the parsed data.
 
     Returns one of: success, partial_resumable, partial_not_resumable,
-    failed, commit_missing, docs_missing, merge_conflict, context_overflow.
+    failed, crashed, timed_out, infra_error, commit_missing, docs_missing,
+    merge_conflict, context_overflow, stalled_context_continuation,
+    workflow_blocked, workflow_skipped.
     """
     status = data.get("status", "")
 
@@ -174,12 +176,20 @@ def determine_status(data):
         return "crashed"
     elif status == "timed_out":
         return "timed_out"
+    elif status == "infra_error":
+        return "infra_error"
+    elif status == "stalled_context_continuation":
+        return "stalled_context_continuation"
     elif status == "commit_missing":
         return "commit_missing"
     elif status == "docs_missing":
         return "docs_missing"
     elif status == "merge_conflict":
         return "merge_conflict"
+    elif status == "workflow_blocked":
+        return "workflow_blocked"
+    elif status == "workflow_skipped":
+        return "workflow_skipped"
     else:
         # Unknown status value — treat as crashed
         return "crashed"