meshbay.git/packages/meshbay-node/src/meshbay_node/indexer, branch 0.19

perf(node): sample 9 MB with the size above 9 MB, keep known ids

2026-10-07T23:27:58Z

hash_version 3: size + first 4 MB + last 4 MB + 1 MB at the middle, 5.5x faster cold on a USB disk than the 45 MB sample. The cache now serves a hit under whatever version it holds, so no existing id moves. Co-Authored-By: Claude Opus 5.5

fix(node): drop a directory moved out of the root on Windows too

2026-10-07T20:22:51Z

On Windows watchdog cannot tell what a deleted path was -- it is gone -- and reports a directory moved out of the root as a file deleted, so the fix in 3d5a168 never ran there and test_the_watcher_reports_a_directory_moved_out failed. A deleted path the index still holds entries below was a directory. Co-Authored-By: Claude Opus 5.5

fix(node): drop a directory moved out of the root from the index

2026-10-07T10:48:34Z

Watchdog reports such a move as one "directory deleted" event and nothing for the files, which the indexer ignored until the next reconcile. The freeze rules still apply: root live, directory gone, parent present. Co-Authored-By: Claude Opus 5.5

fix(node): plug an auto-ejected removable root back once its files return

2026-10-05T09:10:50Z

A node started with the desktop session runs before the session has mounted its USB drives. The safety net then auto-ejected every removable root and persisted it exactly like an operator's eject, so after each reboot those roots stayed ejected until someone plugged them by hand (seen on a node whose /media drives were mounted a minute after it started). An auto-eject is now stored as such ("auto" in roster.db). At startup and at every reconcile, an auto-ejected root whose path is readable again is checked against a few files the hash cache knows under it, at the same path with the same size and mtime; one found and the root is plugged back and rescanned. An empty mount point or another drive in its place is not recognised and stays ejected. An operator's eject is never undone automatically. Co-Authored-By: Claude Opus 5.5

feat(mnp): sharing a folder is decided on the node's machine only (MNP 6.0)

2026-10-02T08:20:09Z

root_add, root_update and group_attach leave MNP: adding a directory and switching writable/removable go through the loopback API (native dialog in the desktop app) or the CLI. The operator's Settings tab still lists the roots from any browser, read-only. The desktop app refuses to sign those ops; a loopback flag change now reaches open pages (publish_roots). Co-Authored-By: Claude Opus 5.5

refactor: remove the unused GroupIndex.serialize chain

2026-09-28T14:24:12Z

serialize/deserialize had no production caller, and took with them the per-chunk signature, the ChaCha20 cipher variant and the zstandard dependency. Key derivations are unchanged. Docs corrected, including design §4.3's claim that chunks are compressed. Co-Authored-By: Claude Opus 5.5

refactor(node): move per-app enrichment out of NodeDaemon

2026-09-24T16:30:10Z

The _enrich_*/_reenrich_* methods and _on_enriched become EnrichmentMixin in meshbay_node/enrichment.py, with the two directory helpers only they use. APP_DIR_KEYS stays on NodeDaemon. Co-Authored-By: Claude Opus 5.5

Merge origin/main: the tree-wide ruff pass beside the Music reconnect work

2026-09-19T15:58:05Z

fix(node): the Music app's tags survive a restart

2026-09-19T15:42:11Z

media_cache held video_meta, photo_meta and thumbs; the audio tags lived only in the in-memory IndexEntry. So every start re-read every audio file the node serves, and until that pass landed it served an index with no artist on any track — one the Music app cannot group. Over a real 6176-file library the pass costs 27.8s cold and 6.4s from audio_meta. Only what the bytes decided is stored. The filename and folder fallbacks still run live, or a renamed file would get the old name's answer; the sibling-cover scan reads the folder, so it stays live too; and a read that failed is not cached, or one bad read becomes permanent. Co-Authored-By: Claude Opus 5

style: the 98 ruff could not fix, so the linter is a signal again

2026-09-19T12:39:38Z

The pass before this applied ruff's own fixes. These are the ones needing a decision, and the point of doing them is that `ruff check .` now passes: a linter reporting 98 known-acceptable findings reports nothing, because the next real one arrives invisible. **Lines over 100 (70).** Mostly wrapped where they stood. Two exceptions: the aligned trailing comments in `protocol.py`'s message table were shortened rather than wrapped, because wrapping one row of a table breaks the table; and in `models.py` the column comments moved above their columns for the same reason. **Imports below the first statement (14).** `csam.py` kept its FastAPI imports under a section header halfway down the file; two node tests had a constant and a `pytestmark` wedged between two import blocks. Moved, not suppressed. **Bindings nothing reads (4).** Three in tests, where the call stays and only the name goes — `_user(client, "listener")` is there to create the user, not to return one. The fourth was in `revocation.py` and was not a lint finding at all: `_connect_and_listen` opened an httpx stream to the WebSocket URL, did `pass`, and then opened the real connection through the `websockets` library. One pointless request per connect, left over from before that library was used directly. Removed, and `httpx` with it. **`l` as a name (4)**, **semicolons (6)** in the POC spikes, and the rest. 2893 passed, the same count as the two commits before it. `meshbay_node/revocation.py` is worth a decision separately: 154 lines that nothing imports, superseded by `hub_client.maintain_ws`'s `on_revocation`. This commit only stopped it failing the linter. Co-Authored-By: Claude Opus 5