iroh-gossip-rendezvous/CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file. The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

[0.2.1] — 2026-05-11

Changed

• Fast recovery from DhtStatus::PublishFailing. The write loop previously slept the full write_period (default 5 min) between every iteration and still applied the 1/(n+1) probability gate, so a node stuck in PublishFailing could wait many minutes — even half an hour with a populated active view — before retrying. The loop now checks dht_status before sleeping: while failing, it uses the new failure_retry_period (default 15 s) and bypasses the gate, so every tick attempts a write until one succeeds. On success the steady-state cadence and gate are restored.

Added

• Builder::failure_retry_period(Duration) — override the recovery cadence (additive, non-breaking). Public constant DEFAULT_FAILURE_RETRY_PERIOD = 15 s.

Rationale

The probability gate exists to keep N healthy publishers from storming the DHT with redundant writes — but once our publish has failed, redundancy is irrelevant; we need to land one successful write. Skipping the gate while failing turns the recovery time into a function of failure_retry_period, not of (n + 1) × write_period.

[0.2.0] — 2026-05-11

Changed

• Breaking: bumped iroh to =1.0.0-rc.0 and iroh-gossip to 0.99. Public API exposes iroh types (Endpoint, SecretKey, PublicKey, iroh_gossip::api::*), so consumers must move to the iroh 1.0-rc.0 ecosystem in lockstep.
• Breaking: bumped ed25519-dalek to =3.0.0-pre.7 to match iroh-base 1.0.0-rc.0. The version is pinned with = because iroh-base itself hard-pins the pre-release; drop the = when iroh-base relaxes.
• Loosened the sha2 pin from =0.11.0-rc.5 to ^0.11. iroh-base 1.0.0-rc.0 no longer hard-pins the rc, so we follow the same caret-range convention as the rest of the crypto stack.
• Bumped transitive crypto deps via cargo update: mainline 6.0.1 → 6.1.2, plus the usual sweep of tokio/hyper/rustls patches.

Notes

• No source changes were required: the iroh/gossip APIs we touch (Endpoint::builder, endpoint::presets::N0, SecretKey::generate, PublicKey, iroh_gossip::api::{Event, GossipSender}) survived the bump verbatim.
• All 96 lib tests, property tests, and integration tests pass against the new stack.

0.1.0 — 2026-04-22

First public release. Passphrase-based peer rendezvous for iroh-gossip swarms over the public Mainline DHT, keyed from a shared passphrase + an application salt — no central bootstrap server and no prior knowledge of peer addresses required.

Algorithm

• K-sharded BEP 44 mutable-item slots (default K = 3), each with its own HKDF-derived Ed25519 signing keypair and per-slot AEAD wrapper key.
• Logical aging with random-tie eviction; vouching of up to V neighbors at age 1; two-layer ChaCha20-Poly1305 AEAD envelope over fixed-width entries.
• Seq-ratcheted per-epoch wrapper keys; one-shot body key per record.
• Clock-free: no dependence on wall-time across nodes.
• Full specification lives at PROTOCOL.md.

Public API

• Rendezvous::join(passphrase, app_salt) — one-call entry point.
• Rendezvous::builder() — full Builder for endpoint/secret_key/ parameter overrides.
• Rendezvous::{sender, subscribe, state, endpoint, node_id, topic_id, broadcast, shutdown} — handle surface.
• iroh_gossip types re-exported where useful; otherwise thin wrappers.

Feature flags

• sim — discrete-event simulator + test-support exports.
• kani — Kani proof harnesses (off by default).
• miri-compat — skip Miri-incompatible tests.
• mainline-live — integration tests against the real public Mainline DHT.
• test-support — Builder::dht_backend for injecting custom DhtSlots.

Correctness evidence shipping with the crate

• 119 tests across unit, property, integration, and doctest layers; all green.
• Property tests (proptest) on wire format roundtrip, merge invariants (|R'| ≤ B, age bound, self at age 0), crypto roundtrip across epoch boundaries.
• Integration tests over real iroh endpoints with an injected in-memory DHT: two-node bootstrap, multiple-subscriber fan-out, drop cancels background tasks.
• Mutation testing (cargo-mutants) — 100% catch rate on testable mutations (119 caught, 27 unviable, 0 missed). See .cargo/mutants.toml for the exclusions and their rationale.
• Fuzz harnesses (cargo-fuzz) for decode_entries.
• Bounded model checking (Kani) of undefined-behavior / panic absence on crypto::epoch_of (full i64 × u64 ≥ 1 space) and wire::decode_entries (arbitrary byte input). Structural invariants are covered by the stronger proptest + Stateright combo.
• Distributed-system model checking (Stateright): safety invariants S1 (record size ≤ B) and S2 (age < A_max) proved exhaustively for N ≤ 6, K ≤ 3; liveness witnesses under weak fairness; randomized large-N exploration available via just model-random.
• Miri — UB-free on the synchronous modules (keys, wire, crypto, merge).
• Simulator baseline (examples/simulator.rs) — cold-start (8 nodes) and symmetric partition (5/5) scenarios both heal in 1 round at 100%. Asymmetric 1:100 with default K = 3 is a documented limitation in PROTOCOL.md §8.

Supply chain

• MIT-licensed; MSRV 1.93.0; all crypto dependencies from the audited RustCrypto stack.
• cargo-deny + cargo-audit green on every push (transitive unmaintained advisory from paste via iroh → portmapper → netwatch documented in deny.toml).
• [package.metadata.docs.rs] configured for all-feature docs builds.