Files
meshcore-open/documentation/routing-paths.md
T
zjs81 5ea44d6ce9 Tester feedback: routing, bulk select, notifications, map crash
Routing
- Auto routing picks routes by Thompson sampling over faded delivery
  stats (24 h half-life, capped evidence), weighted by what the radio
  currently hears; retries prefer a different route; last attempt floods.
- Route failures are held until a message's fate is known, so an offline
  contact no longer wrecks its route history; failing routes are only
  deleted after a week without a delivery.
- Direct (zero-hop) path is a rotation candidate (#643) and a new Direct
  mode in the routing sheet; history routes are used for repeater logins.
- Remove the route-weight sliders the new chooser no longer uses.

Chats
- Bulk selection in channel chats (spam / not spam / delete / copy) and
  in the Filtered messages list.
- Tapping a search result jumps to the message.
- Swipe-to-reply ignores vertical scrolls.
- Active conversation follows route lifecycle, so covered chats count
  unread again.

Notifications
- Tapping a notification opens its conversation.
- Per-conversation coalescing and one alert per queued-message sync.
- Monochrome status-bar icon; background service stops when the app is
  swiped away; stale notifications cleared on radio or slot change.

Other
- Online map: opaque links and painted dots for dense areas to avoid an
  Impeller GPU crash; expandable hop list on path maps.
- Contact share: clipboard import of shared contacts, "Add again".
- Android: enableEdgeToEdge and scrolling-screenshot support.
- Translations for all new strings; fix mistranslated routing labels.
2026-11-08 23:19:15 -07:00

5.4 KiB
Raw Blame History

Routing Paths

MeshCore Open supports flood routing, direct paths, and routing overrides. Chat → Routing lets you choose Auto, Flood, Direct (always zero hops), or Manual and inspect recent paths. Regions control the scope of channel floods; they do not replace direct-contact routing.

Hash widths and firmware capability

Each hop is represented by a public-key prefix. Current device-info negotiation clamps mode to 0..2, giving 1–3 bytes per hop. Some parsing/display helpers accept four-byte widths; that is not a claim that current device negotiation offers a four-byte mode.

Bytes per hop Possible prefixes Hops fitting in 64 path bytes
1 256 63 with the current six-bit packed hop count
2 65,536 32
3 16,777,216 21

These are prefix-space sizes, not collision-free network capacities. Longer prefixes reduce collisions and leave less room for hops. Repeaters need compatible firmware to forward multi-byte paths; the UI identifies older-than-v1.14 firmware as incompatible with two- and three-byte IDs.

Encoded paths and model fields

For packed non-flood contact path metadata:

final width = ((pathLenRaw & 0xC0) >> 6) + 1;
final hopCount = pathLenRaw & 0x3F;
final byteLength = hopCount * width;

0xFF is the contact flood/unknown-path sentinel. Handle it before decoding the packed fields.

  • Model/storage pathLength stores hop count, with -1 representing flood/unknown.
  • pathHashWidth stores the bytes per hop for that path.
  • pathBytes (or Contact.path) contains concatenated prefixes.
  • Use the path’s own width when splitting, reversing, matching, and rendering it; do not substitute the device’s current width for historical metadata.

For example, a three-hop path at width 2 has six bytes: [A1,A2,B1,B2,C1,C2], rendered as A1A2 → B1B2 → C1C2, and pathLength=3.

For complete observed paths, hop count is pathBytes.length ~/ width. Some display helpers round up a partial final group; malformed or incomplete bytes must not be interpreted as a valid complete route. Prefer decoded observed path bytes when frame metadata describes a different route.

Where paths come from

Contact response frames place packed path metadata at offset 35 and the 64-byte path area at offsets 36–99, counting the response code as byte 0. See contact frame layout.

Channel text payloads contain sender text, not an encrypted hop chain. Observed channel paths come from the raw packet header/path associated with PUSH_CODE_LOG_RX_DATA. Direct-message frame metadata and path updates likewise need their format-specific decoding.

Short prefixes can match several known nodes. An unresolved or ambiguous hop is not proof of a particular repeater identity. Map positions derived from paths are estimates, not GPS fixes.

Storage and route selection

Contacts/messages persist hop count, path bytes, and width. Legacy contact records without width infer it from consistent hop and byte counts, falling back to one byte. Automatic route selection applies when no override is set and follows the App Settings Auto Route Rotation toggle (autoRouteRotationEnabled, on by default).

Auto route selection

PathHistoryService.selectPathForAttempt picks the route for each send attempt; the logic lives in lib/services/route_chooser.dart.

  • Known routes come from the radio's learned path, every attempt's delivery result, and floods that got through (crediting the route the reply came back on). The direct route — zero hops, no path bytes — is a normal candidate.
  • Faded stats: each route keeps delivery and failure counts that lose half their weight every 24 hours (routeStatsHalfLife), so recent behaviour outweighs old.
  • Thompson sampling: for each attempt, a plausible delivery rate is drawn for every route from its faded counts and the best draw wins. Proven routes win most of the time; untested or recovering routes are still tried occasionally. Speed adjusts the score slightly.
  • Context: MeshCoreConnector.routeContextFor reports what the radio heard in the last two hours (advertisers heard directly and the last hop of relayed adverts). Routes whose first hop was heard — or the direct route, when the contact itself was heard directly — score higher; routes that don't fit score lower. Not hearing a repeater is weak evidence, so nothing is excluded. With nothing heard, context has no effect.
  • Retries halve the score of routes already tried for the same message, so they usually try something different without wasting an attempt on a much weaker route. The final attempt always floods.
  • Offline contacts: MessageRetryService holds a message's route failures until its fate is known. If any attempt is acknowledged, the routes that timed out are blamed; if nothing gets through, the contact was probably unreachable and no route is blamed.
  • Forgetting: a failing route is deleted only after it hasn't delivered for a week, so a repeater that is down for a day is still known when it returns.
  • Route picks and their results are written to the app debug log under the Routing tag.

resolvePathSelection uses a route picked from history even when the radio itself holds no path for the contact, so repeater logins and commands use proven routes instead of flooding.

For message-path actions, trace maps, and retry behavior, see chat, channels, and map.