fix(api): survive Windows errno 13 on the job progress file

tests/remote/unit/test_download_manager.py

@@ -7,6 +7,7 @@ cleanup/serialize); they do not exercise the actual subprocess download path.
 from __future__ import annotations
 
 from datetime import datetime, timedelta
+from pathlib import Path
 
 import pytest
 
@@ -141,3 +142,30 @@ def test_normalize_sub_format_preserves_original_sentinel() -> None:
     # A real codec still maps to its enum member; unknowns collapse to None.
     assert normalize_sub_format("srt") is Subtitle.Codec.SubRip
     assert normalize_sub_format("nonsense") is None
+
+
+def test_worker_write_result_retries_denied_replace(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
+    """Windows denies os.replace while the parent holds the progress file open, so the write must retry.
+
+    Without the retry a denied replace loses that update. A value the worker writes once, such as a
+    phase change, then stays invisible until the next write.
+    """
+    import json
+    import os
+
+    from unshackle.core.api import download_worker
+
+    target = tmp_path / "progress.json"
+    real_replace = os.replace
+    attempts = []
+
+    def flaky_replace(src: Path, dst: Path) -> None:
+        attempts.append(dst)
+        if len(attempts) == 1:
+            raise PermissionError(13, "Permission denied", str(dst))
+        real_replace(src, dst)
+
+    monkeypatch.setattr(download_worker.os, "replace", flaky_replace)
+    download_worker.write_result(target, {"progress": 42.0})
+    assert json.loads(target.read_text(encoding="utf-8")) == {"progress": 42.0}
+    assert len(attempts) == 2

unshackle/core/api/download_manager.py

@@ -1004,7 +1004,6 @@ class DownloadQueueManager:
                     stat = os.stat(progress_path)
                     stat_key = (stat.st_mtime_ns, stat.st_size)
                     if stat_key != last_progress_stat:
-                        last_progress_stat = stat_key
                         with open(progress_path, "r", encoding="utf-8") as handle:
                             progress_data = json.load(handle)
                             if progress_data.get("phase") and progress_data["phase"] != job.phase:
@@ -1041,7 +1040,8 @@ class DownloadQueueManager:
                             if progress_data != last_published:
                                 last_published = progress_data
                                 self.publish(job, "progress")
-                except (FileNotFoundError, json.JSONDecodeError, ValueError) as e:
+                        last_progress_stat = stat_key
+                except (OSError, json.JSONDecodeError, ValueError) as e:
                     log.debug(f"Could not read progress for job {job.job_id}: {e}")
 
                 if job.cancel_event.is_set() or job.status == JobStatus.CANCELLED:

unshackle/core/api/download_worker.py

@@ -6,6 +6,7 @@ import json
 import logging
 import os
 import sys
+import time
 import traceback
 from pathlib import Path
 from typing import Any, Dict
@@ -21,10 +22,19 @@ def read_payload(path: Path) -> Dict[str, Any]:
 
 
 def write_result(path: Path, payload: Dict[str, Any]) -> None:
-    # Atomic replace: the parent polls this file while it is being rewritten.
+    """Write the payload with an atomic replace, because the parent polls this file during the write.
+
+    Windows denies the replace while the parent holds the destination open, so retry with a backoff.
+    """
     path.parent.mkdir(parents=True, exist_ok=True)
     tmp = path.with_name(path.name + ".tmp")
     tmp.write_text(json.dumps(payload), encoding="utf-8")
+    for attempt in range(5):
+        try:
+            os.replace(tmp, path)
+            return
+        except PermissionError:
+            time.sleep(0.02 * (attempt + 1))
     os.replace(tmp, path)