🟡 Heimdall Review Status

Review Summary

The changes are well-structured and correct. Both fixes address real side-effect bugs in the 2D nonce pool:

Channelized path (two_d_nonce_pool.rs): Moving the stale-nonce and replacement-underpriced checks before sender_id_or_create / lane mutation is the right approach. The stale nonce check against state_nonce directly is semantically equivalent to the old lane.next_nonce check (since lane.next_nonce was always set to state_nonce before the check). The replacement price lookup via self.lanes.get(&lane_id) correctly returns None for non-existent lanes, avoiding false positives.

Nonce-free path (two_d_nonce_pool.rs): Same pattern — the underpriced check already preceded sender_id_or_create in the old code, but the sender_id_or_create + TransactionId assignment was happening before the check. Now both are correctly deferred until after validation passes.

Sidecar listeners (pool.rs): Changing from try_send().is_ok() to !matches!(try_send(), Err(TrySendError::Closed(_))) correctly distinguishes "channel full" (keep subscriber) from "channel closed" (remove subscriber). This prevents spurious unsubscription under backpressure.

Tests: Good coverage of the key invariants — stale nonce rejection leaves lane cursor, hashes, and sender IDs untouched; underpriced replacement preserves lane state; full listener channels stay subscribed while closed channels get cleaned up.

No issues found.

Base Std historical fork tests

View run