fix: use native Windows process liveness checks

plugins/review-suite/scripts/review_suite_local.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import ctypes
 import html
 import json
 import math
@@ -128,6 +129,22 @@
     ("CODEX_THREAD_ID", "codex_thread_id"),
 )
 
+_PROCESS_SYNCHRONIZE = 0x00100000
+_WAIT_TIMEOUT = 0x00000102
+_ERROR_ACCESS_DENIED = 5
+
+if os.name == "nt":
+    _kernel32 = ctypes.WinDLL("kernel32", use_last_error=True)
+    _open_process = _kernel32.OpenProcess
+    _open_process.argtypes = (ctypes.c_ulong, ctypes.c_int, ctypes.c_ulong)
+    _open_process.restype = ctypes.c_void_p
+    _wait_for_single_object = _kernel32.WaitForSingleObject
+    _wait_for_single_object.argtypes = (ctypes.c_void_p, ctypes.c_ulong)
+    _wait_for_single_object.restype = ctypes.c_ulong
+    _close_handle = _kernel32.CloseHandle
+    _close_handle.argtypes = (ctypes.c_void_p,)
+    _close_handle.restype = ctypes.c_int
+
 
 @dataclass(frozen=True)
 class LocalReviewRequest:
@@ -3817,26 +3834,23 @@ def launch_round(
 
 
 def _process_is_running(pid: int | None) -> bool:
-    if not pid:
+    try:
+        normalized_pid = int(pid) if pid is not None else 0
+    except TypeError, ValueError, OverflowError:
+        return False
+    if normalized_pid <= 0 or normalized_pid > 0xFFFFFFFF:
         return False
     if os.name == "nt":
-        proc = subprocess.run(
-            ["tasklist", "/FI", f"PID eq {int(pid)}", "/FO", "CSV", "/NH"],
-            capture_output=True,
-            text=True,
-            encoding="utf-8",
-            errors="replace",
-            check=False,
-        )
-        if proc.returncode != 0:
-            return False
-        output = (proc.stdout or "").strip()
-        if not output or output.startswith("INFO:"):
-            return False
-        return f'"{int(pid)}"' in output or f",{int(pid)}," in output
+        handle = _open_process(_PROCESS_SYNCHRONIZE, False, normalized_pid)
+        if not handle:
+            return ctypes.get_last_error() == _ERROR_ACCESS_DENIED
+        try:
+            return _wait_for_single_object(handle, 0) == _WAIT_TIMEOUT
+        finally:
+            _close_handle(handle)
     try:
         with open(
-            f"/proc/{int(pid)}/status", encoding="utf-8", errors="replace"
+            f"/proc/{normalized_pid}/status", encoding="utf-8", errors="replace"
         ) as handle:
             for line in handle:
                 if not line.startswith("State:"):
@@ -3849,7 +3863,7 @@ def _process_is_running(pid: int | None) -> bool:
     except OSError:
         pass
     try:
-        os.kill(int(pid), 0)
+        os.kill(normalized_pid, 0)
     except ProcessLookupError:
         return False
     except OSError:

plugins/review-suite/tests/test_review_suite_local.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import json
+import os
 import subprocess
 import sys
 from datetime import datetime, timezone
@@ -24,6 +25,7 @@
     _print_live_completed_run,
     _print_stall_warnings,
     _print_transport_events,
+    _process_is_running,
     _reviewer_wait_line,
     _reroll_candidate_variants,
     _running_status_line,
@@ -61,6 +63,60 @@
 )
 
 
+@pytest.mark.skipif(os.name != "nt", reason="Windows-native process check")
+def test_process_is_running_uses_no_subprocess(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    monkeypatch.setattr(
+        review_suite_local.subprocess,
+        "run",
+        lambda *_args, **_kwargs: pytest.fail("process check started a subprocess"),
+    )
+
+    assert _process_is_running(os.getpid()) is True
+
+
+@pytest.mark.skipif(os.name != "nt", reason="Windows-native process check")
+def test_process_is_running_tracks_child_exit() -> None:
+    child = subprocess.Popen(
+        [sys.executable, "-c", "import time; time.sleep(30)"],
+    )
+    try:
+        assert _process_is_running(child.pid) is True
+        child.terminate()
+        child.wait(timeout=10)
+        assert _process_is_running(child.pid) is False
+    finally:
+        if child.poll() is None:
+            child.kill()
+            child.wait(timeout=10)
+
+
+@pytest.mark.parametrize("pid", [None, 0, -1, "not-a-pid"])
+def test_process_is_running_rejects_invalid_pid(pid: object) -> None:
+    assert _process_is_running(pid) is False  # type: ignore[arg-type]
+
+
+@pytest.mark.skipif(os.name != "nt", reason="Windows-native process check")
+def test_process_is_running_handles_missing_and_native_failures(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    assert _process_is_running(0xFFFFFFFC) is False
+
+    monkeypatch.setattr(review_suite_local, "_open_process", lambda *_args: None)
+    monkeypatch.setattr(review_suite_local.ctypes, "get_last_error", lambda: 5)
+    assert _process_is_running(123) is True
+
+    closed: list[int] = []
+    monkeypatch.setattr(review_suite_local, "_open_process", lambda *_args: 123)
+    monkeypatch.setattr(
+        review_suite_local, "_wait_for_single_object", lambda *_args: 0xFFFFFFFF
+    )
+    monkeypatch.setattr(review_suite_local, "_close_handle", closed.append)
+    assert _process_is_running(123) is False
+    assert closed == [123]
+
+
 def test_reviewer_wait_line_uses_actual_count() -> None:
     assert _reviewer_wait_line({"runs": [{"slot": "alpha"}]}) == (
         "[review-suite] waiting for 1 reviewer; wrapper is active as long as output streams, do not stop it prematurely"