From 83a4ac98cfd4639cff5090a93e88fa71238929e3 Mon Sep 17 00:00:00 2001
From: alexeybe1kin <210597588+alexeybe1kin@users.noreply.github.com>
Date: Fri, 13 Nov 2026 11:15:06 +0200
Subject: [PATCH] fix(drill): the stale-decision mutant changes behaviour
 instead of crashing

Its replacement called control_plane.get from approvals.py, which does not import it, so the
test failed on NameError and the drill printed "Caught" without proving that a stale read is
detected (toolgate-5 F1). The mutant now reads the row outside the write transaction, and the
drill fails as BROKEN MUTANT whenever a mutant's run shows NameError, ImportError or SyntaxError.
---
 toolgate/scripts/mutation_check.py    | 15 +++++++++++--
 toolgate/tests/test_mutation_check.py | 32 +++++++++++++++++++++++++++
 2 files changed, 45 insertions(+), 2 deletions(-)

diff --git a/toolgate/scripts/mutation_check.py b/toolgate/scripts/mutation_check.py
index 1225358..99f41f2 100644
--- a/toolgate/scripts/mutation_check.py
+++ b/toolgate/scripts/mutation_check.py
@@ -16,6 +16,8 @@
 MCP = "toolgate/tests/test_mcp_adapter.py::"
 ARCHIVE = "toolgate/tests/test_execution_boundary.py::"
 INTEGRITY = "toolgate/tests/test_approval_integrity.py::"
+# Failures from a mutant that cannot run, not from the behaviour it changes.
+BROKEN_MUTANT = ("NameError", "ImportError", "SyntaxError", "IndentationError")
 
 # A mutant counts as caught only when its named behavioral test fails, never
 # when collection fails or the subprocess cannot start.
@@ -93,9 +95,12 @@
         if guard is not None:
             guard(conn, record)
         if record.get("status") != "pending":""",
-        """    record = get("request", request_id)
-    if not record:
+        # A stale read: the row is read outside the write transaction, then decided on.
+        """    with _conn() as read:
+        row = read.execute("SELECT * FROM v2_objects WHERE kind='request' AND id=?", (request_id,)).fetchone()
+    if not row:
         return None
+    record = _row(row)
     with _conn() as conn:
         if guard is not None:
             guard(conn, record)
@@ -472,6 +477,12 @@ def main() -> int:
                 return 1
             finally:
                 path.write_text(original, encoding="utf-8")
+            output = result.stdout + result.stderr
+            if any(crash in output for crash in BROKEN_MUTANT):
+                # A test failing on the mutant's own crash proves nothing about the behaviour.
+                print(output)
+                print(f"BROKEN MUTANT (it crashes instead of changing behaviour): {label}")
+                return 1
             if result.returncode != 1 or f"FAILED {test}" not in result.stdout:
                 print(result.stdout + result.stderr)
                 print(f"NOT CAUGHT: {label}")
diff --git a/toolgate/tests/test_mutation_check.py b/toolgate/tests/test_mutation_check.py
index dd6c2dd..876174f 100644
--- a/toolgate/tests/test_mutation_check.py
+++ b/toolgate/tests/test_mutation_check.py
@@ -65,3 +65,35 @@ def fake_run_tests(root, tests, name):
     assert drill.main() == 1
     assert calls == ["baseline", "mutant-0"]
     assert "TIMED OUT: drill anchor mutated" in capsys.readouterr().out
+
+
+def test_a_mutant_that_crashes_is_reported_broken_not_caught(drill, monkeypatch, capsys):
+    crash = "E   NameError: name 'get' is not defined\nFAILED some/test - NameError"
+
+    def fake_run_tests(root, tests, name):
+        if name == "baseline":
+            return subprocess.CompletedProcess([], 0, stdout="", stderr="")
+        return subprocess.CompletedProcess([], 1, stdout=crash, stderr="")
+
+    monkeypatch.setattr(drill, "run_tests", fake_run_tests)
+    monkeypatch.setattr(
+        drill,
+        "MUTANTS",
+        [("crashing mutant", ANCHOR[1], ANCHOR[0], ANCHOR[0] + "  # mutated", "some/test")],
+    )
+    assert drill.main() == 1
+    out = capsys.readouterr().out
+    assert "BROKEN MUTANT (it crashes instead of changing behaviour): crashing mutant" in out
+    assert "Caught: crashing mutant" not in out
+
+
+def test_the_stale_decision_mutant_reads_the_row_itself():
+    # The stale-decision mutant once called control_plane.get from approvals.py, which does not
+    # import it: the test failed on NameError and the drill printed "Caught".
+    spec = importlib.util.spec_from_file_location("mutation_check_names", SCRIPT)
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    label, _file, _before, after, _test = next(
+        m for m in module.MUTANTS if m[0] == "stale decision overwrites consumption"
+    )
+    assert 'get("request"' not in after, label