MeshBay — read-only public mirror https://git.meshbay.org/meshbay.git/atom?h=v0.9.0 2026-08-24T21:34:57Z 2026-08-24T21:34:57Z Christophe Besson cbesson@gmail.com 2026-08-24T21:34:57Z urn:sha1:aeeaf0d537a1942010caf222b72333841b80b7f8 Every existing audio_root test either called ops.set_audio_root directly or mocked out _issue_admin_challenge — none of them exercised real signature verification, _do_admin_response, or the shared groups_ctx/ roster wiring _run_op depends on. Worth ruling out a break somewhere in that real path specifically: a report described a save that looked like it worked (the Music tab showed content right after) not surviving a reload. Drives the real _do_audio_root -> admin_challenge -> sign -> _do_admin_response -> _admin_exec_audio_root path with a genuine Ed25519 operator key, then opens a *separate* Roster instance against the same db file — the direct question a "worked, then reverted" report raises: does the value actually land durably, in a form any later connection reads back correctly. It does; this passes. The one thing missing from the session fixture to get this far was peer-registry self-registration (a real session adds itself on handshake completion — without it, the final ack has nowhere to go, including back to the requester). 2026-08-24T20:24:52Z Christophe Besson cbesson@gmail.com 2026-08-24T20:24:52Z urn:sha1:665fb2004e55b72ea483aa50fea81f6a6fd9c322 musicbay.md's original call — Music needs no root, tag reads are cheap so just cover the whole shared tree — didn't hold up against a real messy library: everything under every shared folder got mixed together with no way to scope Music down to an actual music collection. This adds an audio_root setting, symmetric to video_root in every respect: signed operator op (audio_root/audio_root_ack, MNP bumped to 0.10), validated against a real directory in the group's own roots before a signature is even asked for, gates tag/cover enrichment exactly like video_root gates ffprobe/TMDB (nothing runs until it's set, only files under it once it is), and a set/change fires a one-off sweep of whatever the folder already contains. The old trigger — sweep everything the instant "music" joins enabled_apps — is gone along with the root-less design it belonged to; setting audio_root is now the trigger, mirroring set_video_root's enrich_video_root_fn exactly. Test coverage mirrors the video_root suite: policy (refuse before a signature round trip, accept/store correctly) and the enrichment gate itself (nothing without a root, only files under it, sweep on set).