fix(read): translate gossip→peer-RPC port in selector-registry pool feed (#1590, #836 DATA leg)

[MichaelTaylor3d]

What / why

Closes the #836 read-leg DATA gate. Root cause (decider plan #1586): the connected-pool fetch feed (spawn_selector_registry_feed) seeded the download-side ConnectedPool from raw gossip PoolEvent addresses carrying the gossip port (:9445). PoolProviderLocator::find_providers then emitted :9445 candidates, so the Tier-2 dig.fetchRange dialed the gossip listener instead of the peer-RPC listener → InvalidContentType → silent Tier-2 miss → Tier-3 §21 → 404. Same class of bug #1575 fixed for the DHT routing feed, recurring in the selector-registry/pool feed.

Fix

Apply dht_addr_from_gossip_addr() (existing, #1575; gossip 9445 → peer-RPC 9444) at BOTH feed sites in spawn_selector_registry_feed: the snapshot-seed loop AND the live PeerAdded branch (map_gossip_pool_event). on_pool_event stays addr-agnostic — translation happens at the gossip boundary, mirroring #1575 exactly.

Observability (the gap that hid this 6 iterations)

Added INFO/DEBUG tracing to peer_serve_plaintext, fetch_resource, and the pool locator (dial targets incl. port).

TDD (RED→GREEN, through the REAL feed)

1. selector_pool_feed_candidate_uses_peer_rpc_port_not_gossip_port — drives map_gossip_pool_event→on_pool_event→PoolProviderLocator; asserts the candidate for a resource carries :9444 (RED = :9445).
2. serve_content_plaintext_reaches_pool_holder_at_peer_rpc_port — full Node tiering: Tier1-miss → Tier2 with a port-gated transport (serves only at :9444), holder fed through the real gossip feed at :9445, unroutable upstream; asserts Served{source: Peer}.

Blast radius (gitnexus disabled — socraticode + ripgrep + read)

• spawn_selector_registry_feed / map_gossip_pool_event (peer.rs) — sole caller is the feed; map_gossip_pool_event now pub(crate) for tests.
• ConnectedPool / PoolProviderLocator unchanged in behaviour (still reads stored addrs); only the addr STORED changes.
• No wire/API/format change → patch (0.58.0 → 0.58.1).

Version

patch 0.58.0 → 0.58.1 (behaviour fix, no API/wire change), Cargo.toml + Cargo.lock in agreement.

DRAFT until CI green.

🤖 Generated with Claude Code

[MichaelTaylor3d]

Independent CORRECTNESS review (fresh context) — verdict: CHANGES-REQUIRED (recorded as a COMMENT review; self-review cannot REQUEST_CHANGES) (one gating finding on the hot read path; the port-translation fix itself is correct and well-tested).

What is correct (verified against a479417):

• Both selector-feed sites translate exactly once: snapshot-seed dht_addr_from_gossip_addr(addr) (peer.rs:1289) and the live PeerAdded branch via map_gossip_pool_event (peer.rs:1392). The helper is reused, not duplicated. on_pool_event and PoolProviderLocator stay addr-agnostic (no port math) — no double-translation.
• The DHT routing feed's translation (peer.rs:1336/1358) is unchanged.
• Both tests drive the REAL feed (map_gossip_pool_event→on_pool_event→locator / full Node tiering with a :9444-port-gated transport) and gate on the port, so they go RED on the unfixed code (candidate = :9445). They close the exact gap (mocks that ignored the port) that hid this 6 iterations.
• verify-then-decrypt fail-closed path unchanged (the Err arm still returns None and falls through; only tracing added). New tracing logs public identifiers (store/root/rk hex, byte lengths, dial targets, provider counts) — no keys/salt/plaintext.
• patch bump 0.58.0→0.58.1 correct (no API/wire change); Cargo.toml + Cargo.lock agree. SPEC §5.3 note + DEVELOPMENT_LOG updated. Comments are WHY-focused and read cleanly.
• CI: Rustfmt/Clippy/CodeQL/version-increment/commitlint all green; Test+coverage pending (will validate the two new tests).

Gating finding — see inline on download.rs:712.