mirror of
https://github.com/calibrain/shelfmark.git
synced 2026-11-16 12:38:13 +00:00
Fixes six defects. Each one breaks downloads or deletes user data, and each has a regression test that fails on10710aband on77b80eb(current main). - **Completed-path wait (#1464)**: `_wait_for_completed_path` only returned a directory after two identical signatures of supported formats. - A directory holding only archives (a scene `book.rar` + `.nfo` + `.sfv`) never got a signature and failed with "Completed download has no stable supported files yet". Archives (`.zip`/`.rar`, via `is_archive`) now count toward the signature, so post-processing extracts them again, as in v1.4.0. - With `DOWNLOAD_CLIENT_COMPLETED_PATH_TIMEOUT=0`, `max_attempts` is 1, so every directory failed. One sample with book files or archives is now accepted when no retry is possible. - **Remove & Delete Files (#1401)**: `post_process_cleanup` called `remove(id, delete_files=True)` on torrents matched by `find_existing` that Shelfmark never added, deleting the user's seeding data. `_cleanup_refs` now records `added_by_shelfmark`. A pre-existing torrent (complete, or in progress and joined) is removed but keeps its files. - **Cancelling an unstarted torrent (#1421)**: `_remove_unstarted_torrent` deleted with files whenever `progress == 0`. That included `DownloadStatus.error()`, which always reports 0, and checking states that are re-verifying data already on disk. The torrent must now be QUEUED, PAUSED or DOWNLOADING. rTorrent's `remove(delete_files=True)` always returns False, so a new `DownloadClient.can_delete_files` flag (False on rTorrent) makes it remove without files instead. - **Queue grace overwritten**: when a QUEUED torrent's progress rose on the same poll, the 900 s queue grace was immediately replaced by a shorter stall window, because the orchestrator keeps one deadline. The stall window is now requested only when it is longer than what is left of the queue grace. - **`drop_ignored_files` (#1480)**: the suffix rule `w.endswith("/" + rel)` dropped a kept root `Book.epub` when `Extras/Book.epub` was skipped. Each skipped entry now drops at most one file: the exact path, else the file sharing the most trailing components, else a unique basename, else an AudiobookBay glued `"<folder> <file>"` name (the fallback `match_plan_to_files` uses, matched on folder plus file so chaptered `Book 2 01.mp3` picks the right folder). Ties drop nothing. - **Email and BookLore outputs** ignored `task.ignored_files` and sent every prepared file. Both now apply `drop_ignored_files` once after `prepare_output_files`, and report "No book files found" if the user skipped everything. It is applied in the outputs rather than in `prepare` because the folder output already filters in `transfer_book_files`, and filtering twice could match a kept file on the second pass. - `make python-checks`: ruff, ruff format, basedpyright (backend + tests), vulture all pass. - `uv run pytest tests/ -m "not integration and not e2e"`: 4054 passed on the rebased branch. - New tests run against unfixed77b80eband10710ab: all 18 new regression cases fail, and the 11 existing post-process-cleanup tests also fail there because they now build the 4-tuple cleanup ref.