Roadmap

Where sharerr is going next. Nothing below is a release commitment; the ordering is by size, not a schedule. An item is removed the moment it ships. An idea that gets declined instead moves to docs/COMPATIBILITY.md, with the reason attached.

Open work

Smallest first, by how much each item touches, with what is blocked upstream last:

  • IPv6 link-local and IPv4-mapped addresses in is_private_ip. Its doc says link-local, but the IPv6 branch covers only loopback and fc00::/7, so fe80::/10 reads as public; and it never canonicalises, so with server.bind = "[::]:8477" an IPv4 neighbour arrives as ::ffff:a.b.c.d and reads as public too. Three callers would change: the gluetun refresh hook (commands/serve.rs’s refuse_public) would admit those neighbours, the advertised-host check in web/settings would refuse a typed fe80:: address, and an announce from fe80:: would take the peer’s declared IP. Widens who may call the local hooks, so it needs a decision first; the code is a line plus a test case.
  • A public lighthouse. The software and a Linode module to host it (docker/deploy/lighthouse/terraform/linode/) are done; what is missing is a running public instance, with its URL in docs/LIGHTHOUSE.md, for a friend group that would rather not run their own.
  • Unraid and TrueNAS app listings. Where much of the *arr audience installs from; a Community Applications template and a TrueNAS app.
  • Mutation survivors in sharerr, in phases. scripts/check.sh mutants exists, and every other crate, plus sharerr’s web/auth.rs and totp.rs, has had its survivors worth killing killed (what is left and why: docs/TESTING.md). A sharerr mutant costs about 19 seconds at -j 4 (16CPU/60GB), because each one runs the crate’s whole suite, so a 30-minute sitting covers about 95 mutants. The rest of the crate is 2,173 mutants, about 11.5 hours or 23 sittings, below by phase with counts from cargo mutants --list -p sharerr on 2026-11-06 (they move with the code; re-list a phase before starting it, then split it by file into sittings of about 95). Each phase records its survivors in docs/TESTING.md and kills only the ones whose behaviour matters; MUTANTS_ARGS takes -p for a crate, --file for a file, and --iterate to re-test only the last run’s survivors after adding tests.
    1. Sync, 340 (4 sittings): sync/mod.rs (128), sync/seed.rs (92), library.rs (83), tracker.rs (37).
    2. Items and settings, 350 (4): web/items/mod.rs (164), web/settings/mod.rs (134), web/config_io/mod.rs (52).
    3. Friends, 415 (5): web/topology/mod.rs (203), web/peers.rs (121), gossip.rs (48), pairing.rs (33), catalog.rs (10).
    4. Feeds and diagnostics, 350 (4): torznab/mod.rs (119), web/diagnostics.rs (79), checks.rs (58), web/composition.rs (55), api.rs (27), jackett.rs (12).
    5. Commands and state, 435 (5): commands/doctor/mod.rs (91), state.rs (82), commands/preview.rs (71), notify.rs (61), commands/serve.rs (54), metrics.rs (43), gluetun.rs (33).
    6. The rest, 283 (3): the 21 files under 35 mutants each.
  • Season packs and albums as one torrent. One media file per torrent means a friend’s season search returns single episodes; a multi-file torrent over the existing folder still moves nothing, and seeding.sidecars already builds one. seeding.hybrid and joining a friend’s swarm take single-file torrents only, so each needs its multi-file case too.
  • Request flow. Discovery is one-way today. An inbound request queue with an approve step is the other half of the original idea. Seerr (the merged Overseerr and Jellyseerr) is the request app most friends already run, so its request model is the one to fit.
  • A Prowlarr catalog listing. Blocked upstream. A per-host definition friends install by hand exists (docs/prowlarr/); one listing in Prowlarr’s own catalog, so a friend picks “sharerr” and types one key, needs Prowlarr to accept a custom base URL for Cardigann indexers. That doc says why.
This site uses Just the Docs, a documentation theme for Jekyll.