fix(#149): router marker parser 修复(终结 dispatch-gap)

skills/codex-refactor-loop/scripts/phase9_router_daemon.py

@@ -48,13 +48,51 @@
     *LIFECYCLE_PREFIXES,
 )
 MARKER_RE = re.compile(r"\b(?:[A-Z][A-Z0-9_]*_(?:DONE|RESOLVED|BLOCKED)|META_JUDGE_DONE):[^\s`]+")
-# Refactor (iter4/skill-router-fallback-flood-fix): Old pattern: accepted any
-# marker payload, treating regex fragments containing `|` / `\"` / `*` / `r+1`
-# / backslashes as real markers, flooding pending-events for every old log.
-# New principle: real marker prefix and suffix segments contain only
-# [A-Za-z0-9_./\-] plus limited punctuation; reject other special characters as
-# prompt/regex echoes (per 2026-05-26 maintainer-directive).
-VALID_MARKER_PAYLOAD = re.compile(r"^[A-Z][A-Z0-9_]*:[A-Za-z0-9_./\-]+(?::[A-Za-z0-9_./\-]+)*$")
+
+
+class Phase9MarkerGrammar:
+    # Refactor (iter1/issue-149): refactor helper, no behavior change outside existing routes.
+    #   Old pattern: phase9_router_daemon marker parser 不能可靠识别含中文收敛问题/route 后缀的 judge marker → 漏派 triplet judge 与 converge round,controller 被迫 fallback 全部 dispatch(本会话持续 no-gap churn 根因)。
+    #   New principle: 按 .refactor-loop/runs/phase9-issue149-r2-judge.md consensus(structural):route-specific marker-grammar parser fix,正确解析所有 route marker(含中文 body),不引入 Phase9RoundProjection 抽象。使 router 对所有 glob 可见的 3/3 SOLVER_DONE triplet 与 converge 可靠 dispatch。硬约束:不重建 REFERENCE.md;refactor 注释自含 Old/New;不超范围。
+    ROUTE_TOKEN = re.compile(r"^[A-Za-z0-9_./-]+$")
+    VERDICT_TOKEN = re.compile(r"^[A-Za-z0-9_./-]+$")
+    CONVERGE_RE = re.compile(r"^META_JUDGE_DONE:converge:round-(\d+)(?::.*)?$")
+
+    @classmethod
+    def parse_marker_candidate(cls, text: str) -> str | None:
+        if cls.is_solver_done(text) or cls.parse_converge_round(text) is not None or cls.is_stalled_marker(text):
+            return text
+        if cls._is_lifecycle_or_unknown_marker(text):
+            return text
+        return None
+
+    @classmethod
+    def is_solver_done(cls, marker: str) -> bool:
+        parts = marker.split(":", 3)
+        return (
+            len(parts) >= 3
+            and parts[0] == "SOLVER_DONE"
+            and parts[1] in ROLES
+            and bool(cls.VERDICT_TOKEN.match(parts[2]))
+        )
+
+    @classmethod
+    def parse_converge_round(cls, marker: str) -> int | None:
+        match = cls.CONVERGE_RE.match(marker)
+        return int(match.group(1)) if match else None
+
+    @classmethod
+    def is_stalled_marker(cls, marker: str) -> bool:
+        return marker.startswith("META_JUDGE_DONE:escalate:stalled:")
+
+    @classmethod
+    def _is_lifecycle_or_unknown_marker(cls, marker: str) -> bool:
+        parts = marker.split(":")
+        if len(parts) < 2:
+            return False
+        if not re.match(r"^[A-Z][A-Z0-9_]*(?:_DONE|_RESOLVED|_BLOCKED)?$", parts[0]):
+            return False
+        return all(cls.ROUTE_TOKEN.match(part) for part in parts[1:])
 
 
 @dataclass(frozen=True)
@@ -212,6 +250,9 @@ def _collect_markers(self) -> list[Marker]:
         return markers
 
     def _extract_marker(self, line: str) -> str | None:
+        # Refactor (iter1/issue-149):
+        #   Old pattern: phase9_router_daemon marker parser 不能可靠识别含中文收敛问题/route 后缀的 judge marker → 漏派 triplet judge 与 converge round,controller 被迫 fallback 全部 dispatch(本会话持续 no-gap churn 根因)。
+        #   New principle: 按 .refactor-loop/runs/phase9-issue149-r2-judge.md consensus(structural):route-specific marker-grammar parser fix,正确解析所有 route marker(含中文 body),不引入 Phase9RoundProjection 抽象。使 router 对所有 glob 可见的 3/3 SOLVER_DONE triplet 与 converge 可靠 dispatch。硬约束:不重建 REFERENCE.md;refactor 注释自含 Old/New;不超范围。
         stripped = line.strip().strip("`")
         if self._is_placeholder_or_echo(stripped):
             return None
@@ -228,9 +269,7 @@ def _extract_marker(self, line: str) -> str | None:
                 candidate = match.group(0).rstrip("`.,);:|\"\\")
         if candidate is None:
             return None
-        if not VALID_MARKER_PAYLOAD.match(candidate):
-            return None
-        return candidate
+        return Phase9MarkerGrammar.parse_marker_candidate(candidate)
 
     def _is_placeholder_or_echo(self, text: str) -> bool:
         if "<" in text and ">" in text:
@@ -395,8 +434,7 @@ def _directly_handled(self, marker: Marker, ledger: set[str]) -> bool:
         return False
 
     def _round_from_converge(self, marker: str) -> int | None:
-        match = re.match(r"META_JUDGE_DONE:converge:round-(\d+):", marker)
-        return int(match.group(1)) if match else None
+        return Phase9MarkerGrammar.parse_converge_round(marker)
 
     def _stalled_predicate_holds(self, issue: str, round_no: int) -> bool:
         if round_no < 3:

skills/codex-refactor-loop/scripts/test_phase9_router_daemon.py

@@ -165,6 +165,44 @@ def test_phase9_router_solver_triplet_dispatches_meta_judge_once(self) -> None:
         self.assertIn(str((self.repo / ".refactor-loop" / "logs" / "phase9-issue37-r4-judge.log").resolve()), command)
         self.assertEqual(self.ledger_entries()[0]["key"], "37-4-judge")
 
+    def test_phase9_router_solver_triplet_accepts_non_ascii_summary(self) -> None:
+        # Refactor (iter1/issue-149):
+        #   Old pattern: phase9_router_daemon marker parser 不能可靠识别含中文收敛问题/route 后缀的 judge marker → 漏派 triplet judge 与 converge round,controller 被迫 fallback 全部 dispatch(本会话持续 no-gap churn 根因)。
+        #   New principle: 按 .refactor-loop/runs/phase9-issue149-r2-judge.md consensus(structural):route-specific marker-grammar parser fix,正确解析所有 route marker(含中文 body),不引入 Phase9RoundProjection 抽象。使 router 对所有 glob 可见的 3/3 SOLVER_DONE triplet 与 converge 可靠 dispatch。硬约束:不重建 REFERENCE.md;refactor 注释自含 Old/New;不超范围。
+        for role in ("minimal", "structural", "delete"):
+            self.write_log(
+                f"phase9-issue149-r1-{role}.log",
+                f"SOLVER_DONE:{role}:propose:中文摘要-继续收敛",
+            )
+
+        self.router.tick()
+
+        self.assertEqual(len(self.commands), 1)
+        self.assertIn("phase9-issue149-r1-judge.log", " ".join(self.commands[0]))
+        self.assertEqual([entry["key"] for entry in self.ledger_entries()], ["149-1-judge"])
+
+    def test_phase9_router_controller_dispatched_triplet_across_ticks(self) -> None:
+        for role in ("minimal", "structural"):
+            self.write_log(
+                f"phase9-issue149-r2-{role}.log",
+                f"SOLVER_DONE:{role}:propose:中文摘要-等待第三路",
+            )
+
+        self.router.tick()
+        self.assertEqual(self.commands, [])
+        self.assertEqual(self.ledger_entries(), [])
+
+        self.write_log(
+            "phase9-issue149-r2-delete.log",
+            "SOLVER_DONE:delete:propose:中文摘要-第三路完成",
+        )
+        self.router.tick()
+        self.router.tick()
+
+        self.assertEqual(len(self.commands), 1)
+        self.assertIn("phase9-issue149-r2-judge.log", " ".join(self.commands[0]))
+        self.assertEqual([entry["key"] for entry in self.ledger_entries()], ["149-2-judge"])
+
     def test_phase9_router_accepts_solver_issue_logs_for_triplet(self) -> None:
         for role in ("minimal", "structural", "delete"):
             self.write_log(
@@ -238,6 +276,24 @@ def test_phase9_router_converge_dispatches_next_round_solvers(self) -> None:
             ["37-5-delete", "37-5-minimal", "37-5-structural"],
         )
 
+    def test_phase9_router_converge_accepts_non_ascii_reason(self) -> None:
+        self.write_log(
+            "phase9-issue149-r2-judge.log",
+            "META_JUDGE_DONE:converge:round-3:中文收敛问题-继续三路判断",
+        )
+
+        self.router.tick()
+
+        self.assertEqual(len(self.commands), 3)
+        logs = " ".join(" ".join(command) for command in self.commands)
+        self.assertIn("phase9-issue149-r3-minimal.log", logs)
+        self.assertIn("phase9-issue149-r3-structural.log", logs)
+        self.assertIn("phase9-issue149-r3-delete.log", logs)
+        self.assertEqual(
+            sorted(entry["key"] for entry in self.ledger_entries()),
+            ["149-3-delete", "149-3-minimal", "149-3-structural"],
+        )
+
     def test_phase9_router_accepts_meta_judge_issue_log_for_converge(self) -> None:
         self.write_log("meta-judge-issue100-r2.log", "META_JUDGE_DONE:converge:round-3:need-more")
 
@@ -567,6 +623,7 @@ def test_phase9_router_persists_fallback_dedup_across_restart(self) -> None:
 
     def test_phase9_router_rejects_junk_markers_with_regex_special_chars(self) -> None:
         """Markers containing pipe/quote/backslash/template chars are prompt/regex echoes, not real markers."""
+        self.write_log("phase9-issue41-r1-judge.log", "META_JUDGE_DONE:converge:round-2:中文收敛问题-合法")
         junk_lines = [
             'grep "META_JUDGE_DONE:converge:r+1" log',
             'pattern META_JUDGE_DONE:converge:round-2:With|round-3|Choose|minimal\\""',
@@ -581,7 +638,8 @@ def test_phase9_router_rejects_junk_markers_with_regex_special_chars(self) -> No
         self.router.tick()
 
         events = self.pending_events()
-        self.assertEqual(self.commands, [])
+        self.assertEqual(len(self.commands), 3)
+        self.assertTrue(all("phase9-issue41-r2-" in " ".join(command) for command in self.commands))
         for forbidden_token in ("r+1", "round-3|Choose", "no-op;", "stalled:*", "CANONICAL_HUMAN_LABELS"):
             with self.subTest(forbidden_token=forbidden_token):
                 self.assertNotIn(
@@ -611,6 +669,17 @@ def test_phase9_router_source_does_not_introduce_forbidden_abstractions(self) ->
                     f"phase9_router_daemon.py must not introduce forbidden boundary token: {forbidden}",
                 )
 
+    def test_phase9_router_marker_grammar_is_route_specific_not_ascii_payload_gate(self) -> None:
+        src = PHASE9_ROUTER.read_text(encoding="utf-8")
+
+        self.assertIn("class Phase9MarkerGrammar", src)
+        self.assertIn("def parse_marker_candidate", src)
+        self.assertIn("def parse_converge_round", src)
+        self.assertIn("def is_stalled_marker", src)
+        self.assertNotIn("class Phase9RoundProjection", src)
+        self.assertNotIn("Phase9RoundProjection(", src)
+        self.assertNotIn("VALID_MARKER_PAYLOAD.match(candidate)", src)
+
     def test_main_once_dispatches_via_temp_repo_root(self) -> None:
         self.solver_triplet(issue=37, round_no=4)
         commands: list[list[str]] = []