feat(core): implement infosync

[iamquang95]

Fix #173

Stack on: #489

File	Change
crates/infosync/src/lib.rs (+506)	Component: new / trigger / protocols(slot) / proposals(slot), a ResultStore (dedup, history cap, slot-selection), and the priority subscribe callback.
crates/infosync/tests/infosync_integration.rs (+356)	3-host libp2p integration test (ports TestInfoSync): all nodes converge on the same agreed priorities.
crates/core/src/types.rs (+89)	ProposalType → open enum (Unknown(String)), as_str / From / serde-via-String; drops Copy.

1. crates/infosync/src/lib.rs — self-contained. Check the store dedup/cap (>= ⇒ caps at 99, matches charon), protocols/proposals slot-selection + defaults (local / [Full]), handle_results topic routing, build_request (3 topics + InfoSync duty).
2. crates/core/src/types.rs — the only shared-type change. Design call to confirm: ProposalType open enum + loss of Copy, to preserve forward-compat values from newer peers (charon's type is an open string). Only users are core's def + infosync.
3. integration test — note it runs at the infosync+priority layer, not app.Run (infosync isn't wired into the app yet).

[iamquang95]

@claude review this PR

[claude]

Claude finished @iamquang95's task in 2m 5s —— View job

---

Review: feat(core): implement infosync

I read the new infosync crate, the ProposalType change, and the integration test, and compared against charon/core/infosync/infosync.go. This is a faithful, well-documented port with strong test coverage. Findings below — mostly observations; nothing blocking.

Tasks

• Read changed files
• Compare against charon Go source
• Review correctness (store dedup/cap, slot-selection, topic routing, request build)
• Review ProposalType open-enum change
• Review integration test

---

✅ Parity with charon — confirmed

• add_result dedup-vs-last + >= MAX_RESULTS cap (effective 99) matches addResult exactly, including the off-by-one — and it's documented on MAX_RESULTS. 👍
• protocols(slot) / proposals(slot) slot-selection + defaults (local / [Full]) match Protocols/Proposals.
• handle_results topic routing + "store only if a version was agreed" (!res.versions.is_empty()) matches.
• build_request (3 topics in version/protocol/proposal order + InfoSync duty) and the mock-alpha augmentation at new() match.
• ProposalType open enum maps cleanly onto charon's type ProposalType string — full/builder/synthetic known, everything else preserved as Unknown. Removing Copy is safe: the only users are core/types.rs + infosync (verified by grep).

Observations / minor suggestions

1. Slot-ordering assumption is now implicit (inherited from charon).
Both getters break on result.slot > slot and add_result dedups only against results.last() (lib.rs:109, 131, 150). Both rely on results being appended in non-decreasing slot order. If trigger is ever called out of slot order (a late/retried slot), selection silently returns the wrong entry and the dedup misses. Charon has the identical assumption, so this is parity-preserving, not a regression — but in Rust it'd be cheap to make the invariant explicit (a debug_assert!(results.last().map_or(true, |r| r.slot <= result.slot)) in add_result, or a doc note). Low severity since infosync is triggered once per epoch in order.

2. Poisoned-mutex behavior in the subscribe callback (lib.rs:103-119).
handle_results → add_result use .expect("...poisoned"). The priority fan-out invokes this callback; if any lock holder ever panics, the lock stays poisoned and every subsequent decided duty panics the fan-out path. Given the critical sections are tiny and panic-free this is unlikely, but consider lock().unwrap_or_else(|e| e.into_inner()) so one bad duty can't permanently wedge infosync. Minor.

3. Debug-log fidelity (lib.rs:93). Charon logs each topic separately (z.Any(result.Topic, result.Priorities)); the port logs ?results once. Cosmetic — the info is still there.

4. Forward note (not this PR). When infosync is wired into the app, downstream consumers of proposals(slot) must handle ProposalType::Unknown — that's the intended trade-off of the open enum, just worth tracking so an Unknown doesn't get silently treated as Full.

Tests

Unit tests cover dedup, cap, slot-selection, topic routing, unknown-proposal preservation, and the no-version-discard path — thorough. The 3-host integration test is well-constructed: the capture subscriber registered after infosync's guarantees the store is updated before assertion, and the TestConsensus claims the decision under-lock before fan-out to avoid double-notify. Nice.

---

Verdict: Looks good to merge. The slot-ordering assumption (#1) is the only item I'd consider hardening with a debug_assert, and it's optional given charon does the same.
· feat/infosync

[iamquang95]

For 1, keep the current implementation for parity, and it's low enhancement