broadcast_queue_shim_for_ndk 0.8.2

Offline-first wrapper around the ndk package's broadcast use case. Persists Nostr events to a local sembast queue and retries until targeted relays acknowledge delivery, return terminal NIP-01 rejecti [...]

0.8.2

• Accept url_set_for_ndk 0.2.0 alongside 0.1, so the queue can be used with sync_engine_shim_for_ndk 0.9.0. Nothing changed in the code.

0.8.1

• Widen the ndk constraint to >=0.10.0-dev.7 <0.12.0 so the shim can be used alongside ndk 0.11.x. The lower bound matches url_set_for_ndk.

0.8.0

• Breaking: RelaySet moves to the url_set_for_ndk package as UrlSet, re-exported by this one. Rename RelaySet to UrlSet, and the related types to match: RelayListFn to UrlListFn, RelayLookup to UrlListLookup (RelayListFound.relays becomes UrlListFound.urls), RelayListKind to UrlListKind, ndkRelayListFn to ndkUrlListFn, OutboxRelays to OutboxUrls, etc. The hide RelaySet on the ndk import is no longer needed.
• Breaking: UrlSet.inbox and UrlSet.dm take one pubkey. Replace RelaySet.dm(pubkeys) with UrlSet.union([for (final p in pubkeys) UrlSet.dm(p)]). Queues stored by earlier versions load unchanged.
• broadcast rejects UrlSet.blossom leaves, which name Blossom servers.

0.7.0

• Breaking: BroadcastStatus gains superseded and cancelled. Exhaustive switch statements on it need new cases.
• Queuing a replaceable or addressable event supersedes the pending records of its older versions: they stop retrying, become BroadcastStatus.superseded and get QueuedBroadcast.supersededBy. An older version queued after a newer one is never sent.
• Add cancel(id, {pubkey}): stops the retries of a pending record, which becomes BroadcastStatus.cancelled and gets QueuedBroadcast.cancelledAt. Queuing it again with broadcast reopens it.

0.6.0

• Breaking: require ndk: ^0.10.0-dev.1.
• Breaking: broadcast takes required RelaySet relaySet instead of relays. Migrate with relaySet: RelaySet.explicit([...]). Files importing package:ndk/ndk.dart need hide RelaySet.
• Add RelaySet: explicit, outbox, inbox, nip65, dm (kind 10050), private (NIP-37, kind 10013), union and fallback. Sets are persisted, resolved by the worker (offline included) and frozen at the first successful resolution. dm and private lists are searched on the NIP-65 write relays.
• A lookup is found, not found (EOSE) or unavailable. Unavailable ones retry with backoff and never make fallback fall through. An entry resolving to no relay is failed.
• Relay lists are queried on defaultIndexerRelays. Private lists are decrypted once through ndk.decryptedEventPayloads, never timed out.
• New options: relayListFn, relayListDiscoveryRelays, relayListQueryTimeout. New QueuedBroadcast fields: pendingRelaySet, resolutionAttempts, resolutionError.

0.5.1

• Widen the ndk constraint to >=0.9.0 <0.11.0 so the shim can be used alongside ndk 0.10.x.

0.5.0

• Require ndk: ^0.9.0.

0.4.0

• Add an optional pubkey argument to broadcast, get, watch, and rebroadcast that attributes a queued entry to a local account. Records are now keyed by the pair (event.id, pubkey), so the same event queued under two accounts is two independent records. QueuedBroadcast gains a pubkey field and key / keyFor helpers.
• Add clearLocalAccountData({required String pubkey}) to remove every entry queued under an account (for logout) and clearAllLocalData() to empty the store. pubkey is a plain label, not event.pubKey: gift wraps (kind 1059) carry an ephemeral event pubkey, so pass the real sending account for NIP-17 DMs to be clearable.
• Entries queued without a pubkey stay unattributed and keep the bare event.id as their sembast key, so databases written before 0.4.0 keep working. Unattributed entries are never removed by clearLocalAccountData.

0.3.0

• Stop retrying a relay/event pair when the relay returns NIP-01 OK false with a terminal machine-readable prefix: pow, blocked, invalid, restricted, or error.
• Add BroadcastStatus.failed, QueuedBroadcast.terminalErrors, and QueuedBroadcast.failedAt for entries where every relay is either acked or terminally rejected.
• Keep retrying non-terminal failures such as rate-limited, unknown prefixes, timeouts, no response, and transport errors.
• Stop retrying a relay/event pair after 12 consecutive online attempts where the relay is inaccessible.

0.2.1

• Fix int64 overflow in computeBackoff on native platforms (Android, iOS, Linux, macOS, Windows). After ~52 failed attempts (with the default 5s initial delay), pow(2, attempts) wrapped to a negative integer, causing double.clamp to throw Invalid argument(s). The calculation now uses double arithmetic (pow(2.0, ...)) to avoid the overflow.

0.2.0

• Connectivity-aware periodic retries. OfflineBroadcast.withNdk() now subscribes to ndk.connectivity.relayConnectivityChanges and pauses the internal timer while no public relay is connected. The offline -> online edge triggers an immediate replay. retryNow() always bypasses the check.
• Local/private hosts (loopback, RFC1918 IPv4, ULA/link-local IPv6, mDNS .local) are filtered out when deriving the online signal, so a connected dev relay on localhost cannot mask a real outage.
• The default constructor accepts an optional Stream<bool> onlineSignal for callers who want to wire their own connectivity source.

0.1.0

Initial release.

• OfflineBroadcast wraps ndk.broadcast and persists every event to a caller-provided sembast database before attempting to send it.
• Required relays list at broadcast time. No gossip-based auto-selection.
• Per-relay acks are tracked monotonically. An entry is considered delivered only when every targeted relay has acknowledged the event with broadcastSuccessful: true; once set, the delivery timestamp is never cleared by the shim.
• Records are kept after delivery. They are never auto-deleted.
• rebroadcast(id) queues a one-shot push to every relay without touching the historical ack set.
• rebroadcast(id, relay: ...) queues a one-shot push to a specific relay, adding it to the entry's target list if absent.
• Retries are driven by an internal periodic timer started via start(), plus a public retryNow() callers can wire to connectivity signals.
• Backoff is exponential with full jitter, bounded by initialBackoff and maxBackoff.
• The shim never signs events. Signing remains NDK's responsibility.