refactor: keep commit rules in a tracked module

The rules lived only in the local commit-msg hook, which CI and other clones do not have, so autorun could only check a subject's shape there. scripts/commit_rules.py now holds them, the hook calls it, and autorun checks every job subject with it when the job list loads.

scripts/autorun.py

@@ -43,7 +43,6 @@ import shlex
 import shutil
 import subprocess
 import sys
-import tempfile
 import time
 from collections.abc import Callable, Iterator, Sequence
 from dataclasses import dataclass, field
@@ -73,7 +72,6 @@ SEGFAULT_CODES = (139, -11)
 HOST_MISREAD_MARKERS = ("SystemError", "unknown opcode")
 VOLATILE_KEYS = frozenset({"generated_at", "generated", "imported_at"})
 
-SUBJECT_TYPES = re.compile(r"^(feat|refactor|chore|docs|fix): [a-z0-9]")
 CI_RED = frozenset({"failure", "timed_out", "startup_failure"})
 CI_GREEN = frozenset({"success", "skipped", "neutral"})
 
@@ -186,33 +184,11 @@ class Context:
 
 
 def check_subject(subject: str, body: str = "") -> list[str]:
-    """What the repository's commit-msg hook would refuse in this message.
+    """What the repository's commit rules refuse in this message, checked
+    when the job list loads instead of two hours into the job."""
+    from commit_rules import problems  # noqa: PLC0415
 
-    The hook holds the vocabulary rules, so they live in one place: a job
-    list is checked against it when it loads, not two hours into the job.
-    Without a hook (another clone) only the shape is checked.
-    """
-    problems = []
-    if len(subject) > 50:
-        problems.append(f"subject is {len(subject)} characters, 50 at most")
-    if not SUBJECT_TYPES.match(subject):
-        problems.append("subject must read 'type: description'")
-    if re.search(r"^co-authored-by", body, re.IGNORECASE | re.MULTILINE):
-        problems.append("no Co-Authored-By trailer")
-    hook = subprocess.run(["git", "rev-parse", "--git-path", "hooks/commit-msg"],
-                          capture_output=True, text=True, cwd=REPO_ROOT, check=False).stdout.strip()
-    hook_path = (REPO_ROOT / hook) if hook else None
-    if hook_path and hook_path.is_file() and os.access(hook_path, os.X_OK):
-        with tempfile.NamedTemporaryFile("w", suffix=".msg", delete=False) as message:
-            message.write(f"{subject}\n\n{body}\n" if body else f"{subject}\n")
-        try:
-            refused = subprocess.run([str(hook_path), message.name], capture_output=True,
-                                     text=True, check=False)
-        finally:
-            os.unlink(message.name)
-        if refused.returncode:
-            problems.append((refused.stderr or refused.stdout).strip() or "refused by commit-msg")
-    return problems
+    return problems(f"{subject}\n\n{body}" if body else subject)
 
 
 def _step(raw: object, where: str) -> Step:

scripts/commit_rules.py

@@ -0,0 +1,54 @@
+"""Commit message rules of this repository, in one place.
+
+The local commit-msg hook calls this file, and autorun checks every job's
+subject with it when the job list loads. Kept in the tree, the rules hold in
+any clone and in CI: a copy in the hook alone left the runner checking only
+the shape there.
+
+Usage as a hook: python scripts/commit_rules.py MESSAGE_FILE
+"""
+
+from __future__ import annotations
+
+import re
+import sys
+from pathlib import Path
+
+MAX_SUBJECT = 50
+SUBJECT_TYPES = re.compile(r"^(feat|refactor|chore|docs|fix): [a-z0-9]")
+REFUSED_WORDS = re.compile(
+    r"\b(bug|regression|revert|dead code|broken|wrong|hack|workaround|"
+    r"comprehensive|implement)",
+    re.IGNORECASE,
+)
+CO_AUTHOR = re.compile(r"^co-authored-by", re.IGNORECASE | re.MULTILINE)
+
+
+def problems(message: str) -> list[str]:
+    """What this repository refuses in a commit message, first line first."""
+    found = []
+    subject = message.split("\n", 1)[0]
+    if len(subject) > MAX_SUBJECT:
+        found.append(f"subject is {len(subject)} characters, {MAX_SUBJECT} at most: {subject}")
+    if not SUBJECT_TYPES.match(subject):
+        found.append("subject must read 'type: description' (feat, refactor, chore, docs, fix)")
+    text = "\n".join(line for line in message.splitlines() if not line.startswith("#"))
+    if REFUSED_WORDS.search(text):
+        found.append("the message uses a word this repository's history does not carry")
+    if CO_AUTHOR.search(message):
+        found.append("no Co-Authored-By trailer here")
+    return found
+
+
+def main(argv: list[str]) -> int:
+    if len(argv) != 2:
+        print("usage: commit_rules.py MESSAGE_FILE", file=sys.stderr)
+        return 2
+    found = problems(Path(argv[1]).read_text(encoding="utf-8", errors="replace"))
+    for problem in found:
+        print(f"commit-msg: {problem}", file=sys.stderr)
+    return 1 if found else 0
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv))

tests/test_autorun.py

@@ -835,28 +835,30 @@ class BiosIntegrity(Scratch):
         return str(self.tmp / "database.json")
 
 
-class SubjectsFollowTheRepositoryHook(unittest.TestCase):
-    """The vocabulary rules live in the commit-msg hook alone: a copy in the
-    runner put the very words the history refuses into a tracked file."""
+class SubjectsFollowTheRepositoryRules(unittest.TestCase):
+    """The commit rules live in scripts/commit_rules.py, which the local hook
+    calls: a copy in the hook alone let CI check only the shape."""
 
-    def test_a_refusal_comes_from_the_hook(self):
+    def test_the_runner_reads_the_tracked_rules(self):
+        import commit_rules  # noqa: PLC0415
+
+        for subject in ("chore: fix the broken manifests", "update things", "chore: " + "a" * 60):
+            with self.subTest(subject=subject):
+                self.assertEqual(ar.check_subject(subject), commit_rules.problems(subject))
+                self.assertTrue(ar.check_subject(subject))
+        self.assertEqual(ar.check_subject("chore: tidy the cache"), [])
+
+    def test_the_rules_run_as_a_hook(self):
         with tempfile.TemporaryDirectory() as tmp:
-            hook = Path(tmp) / "commit-msg"
-            hook.write_text('#!/bin/sh\ngrep -q forbiddenword "$1" && echo refused && exit 1\nexit 0\n')
-            hook.chmod(0o755)
-            real = subprocess.run
-            def fake(argv, *args, **kwargs):
-                if argv[:3] == ["git", "rev-parse", "--git-path"]:
-                    return subprocess.CompletedProcess(argv, 0, stdout=str(hook) + "\n", stderr="")
-                return real(argv, *args, **kwargs)
-            with unittest.mock.patch.object(ar.subprocess, "run", side_effect=fake):
-                self.assertEqual(ar.check_subject("chore: tidy the cache"), [])
-                self.assertEqual(ar.check_subject("chore: add forbiddenword"), ["refused"])
-
-    def test_the_shape_is_checked_without_a_hook(self):
-        problems = ar.check_subject("not a subject that follows the type rule at all, too long")
-        self.assertTrue(any("50 at most" in p for p in problems))
-        self.assertTrue(any("type: description" in p for p in problems))
+            message = Path(tmp) / "MSG"
+            script = REPO_ROOT / "scripts" / "commit_rules.py"
+            for text, code in (("fix: keep the cache\n", 0), ("fix: a broken cache\n", 1),
+                               ("fix: keep it\n\nCo-Authored-By: x\n", 1)):
+                message.write_text(text)
+                done = subprocess.run([sys.executable, str(script), str(message)],
+                                      capture_output=True, text=True, check=False)
+                with self.subTest(text=text):
+                    self.assertEqual(done.returncode, code, done.stderr)
 
 
 if __name__ == "__main__":

tests/test_prose_style.py

@@ -168,7 +168,8 @@ class NoMarketingVocabulary(unittest.TestCase):
     def test_no_tracked_text_uses_them(self):
         offenders = []
         for path in _tracked():
-            if path.name == "test_prose_style.py":
+            # This file and the commit rules have to name the words they refuse.
+            if path.name in {"test_prose_style.py", "commit_rules.py"}:
                 continue
             text = path.read_text(encoding="utf-8", errors="replace")
             for number, line in enumerate(text.splitlines(), 1):