[Tracking] Lantern bridge-off parity for stock Curio (maxboom / PR #1311)

[Reiers]

Goal

Make Lantern serve the full PDP read+write+watch surface with NO --vm-bridge-rpc and no Glif on the hot path, so it works for both consumers:

• curio-core (us) — uses Lantern via the VMBridge seam; already mainnet-proven zero-Glif. No change required.
• stock Curio / maxboom (upstream PR Only PDP filecoin-project/curio#1311) — embeds Lantern over plain JSON-RPC, no bridge. Every method that still forwards to the bridge is a hard failure here. This is the gap.

This is why a stock-Curio node hit the Glif 429 (#71) that curio-core never saw: maxboom exercises Lantern RPC methods directly that curio-core routed through the bridge.

Background

PR #1311 ("make skiff"/maxboom, review-APPROVED) embeds Reiers/lantern/pkg/daemon + /wallet, configures lanternd.Config{EmbeddedMode:true, Network, FEVMPrefetchAddrs} with no VMBridgeRPC, and talks to it as FULLNODE_API_INFO. So it needs local-only:
read (call/code/logs/storage/block), write (sendRaw/nonce/gas/receipt), and event-watch (logs/search).

Already local (proven bridge-off)

eth_call (#43/#44), eth_getTransactionCount, eth_estimateGas, eth_gasPrice, eth_getBalance, eth_sendRawTransaction (#45 S4), eth_getTransactionReceipt (#45 S5 + #49), eth_getTransactionByHash (local-first), eth_getBlockByNumber, head tracking via gossipsub.

Remaining work (checklist)

• Standalone daemon polls Glif ChainHead every 6s even when gossipsub is live → 429 under stock Curio + PDP #71 / fix(#71): gossip-aware head-poll skip + standalone 30s relaxation (stop Glif 429 under stock Curio) #72 — ChainHead redundant Glif poll under gossip (the 429). DONE, PR fix(#71): gossip-aware head-poll skip + standalone 30s relaxation (stop Glif 429 under stock Curio) #72 in review.
• Local eth_getLogs (FEVM log/event source) so stock Curio PDP watchers work bridge-off #73 — local eth_getLogs (FilecoinPay rails + PDP event watchers). P1, biggest gap.
• Serve eth_getCode from local state (contract-presence checks) bridge-off #74 — local eth_getCode (contract-presence checks). P1.
• Local eth_getStorageAt + eth_getBlockByHash (read parity) bridge-off #75 — local eth_getStorageAt + eth_getBlockByHash. P2.
• Message-block availability: StateSearchMsg fetches message blocks via bridge, stalls bridge-off #50 — StateSearchMsg message-block availability bridge-off (receipt/confirm path). P2.
• mpool: pending-tx confirm + rebroadcast loop (published-but-unmined messages silently stall) #47 — mpool pending-tx confirm + rebroadcast (published-but-unmined stall). P2.
• eth_subscribe(logs) — currently hard-errors w/o bridge; provide pollable fallback or local sub once Local eth_getLogs (FEVM log/event source) so stock Curio PDP watchers work bridge-off #73 lands. P3.

Acceptance gate

A maxboom node (stock Curio, PR #1311 build, no --vm-bridge-rpc) runs PDP end-to-end — add-piece -> prove -> settle, with FilecoinPay rail + dataset watchers advancing — using embedded Lantern with zero Glif on the hot path (Glif only as a cold-gossip catch-up fallback). curio-core's mainnet-proven path stays unchanged.

Action on PR #1311 side

They pin an older Lantern than our HEAD and don't yet have these fixes. Once #73/#74 land + a release, recommend they bump the pin. (Keep pkg/daemon + wallet public API stable — it's now a frozen contract for upstream.)

[Reiers]

Update — all sub-issues resolved (PR #72 + PR #77, HEAD 4462987)

The full bridge-off PDP read+write+event surface is now local-capable:

• Standalone daemon polls Glif ChainHead every 6s even when gossipsub is live → 429 under stock Curio + PDP #71/fix(#71): gossip-aware head-poll skip + standalone 30s relaxation (stop Glif 429 under stock Curio) #72 — ChainHead redundant Glif poll under gossip → fixed (gossip-aware poll skip + standalone 30s relaxation).
• Local eth_getLogs (FEVM log/event source) so stock Curio PDP watchers work bridge-off #73 — local eth_getLogs (per-receipt event-AMT decode → ETH logs; FilecoinPay rails + PDP watchers).
• Serve eth_getCode from local state (contract-presence checks) bridge-off #74 — local eth_getCode (verified bytecode from EVM actor state).
• Local eth_getStorageAt + eth_getBlockByHash (read parity) bridge-off #75 — local eth_getStorageAt + eth_getBlockByHash.
• Message-block availability: StateSearchMsg fetches message blocks via bridge, stalls bridge-off #50 — message/receipt block availability (retrying getter + reconcile warms in-flight tx blocks).
• mpool: pending-tx confirm + rebroadcast loop (published-but-unmined messages silently stall) #47 — mpool pending→confirm→rebroadcast (no more silently-stuck sends).
• eth_subscribe(logs) — still bridge-gated; now cheaply backable by the Local eth_getLogs (FEVM log/event source) so stock Curio PDP watchers work bridge-off #73 local logs as a pollable source. P3, leave open under this umbrella.

Remaining acceptance step (not code): a live bridge-off PDP end-to-end soak on a maxboom-shaped node (add-piece → prove → settle with watchers advancing, zero Glif on the hot path). That's a deploy/verify task, gated on a supervised window — not a Lantern code gap.

Shipping these in v1.8.3. Recommend upstream PR #1311 bump their Lantern pin to v1.8.3 so maxboom nodes get bridge-off behavior. pkg/daemon + wallet public API unchanged.

[Reiers]

v1.8.3 released: https://github.com/Reiers/lantern/releases/tag/v1.8.3 — all parity sub-issues (#71 #73 #74 #75 #47 #50) shipped, plus colorized logging and the non-blocking-reconcile audit fix. latest now resolves here. Leaving this umbrella open only for the P3 eth_subscribe(logs) follow-up + the operational bridge-off PDP soak.