[icmono]

Problem

With --hl-node-compliant (or ?hl=true), eth_getBlockBy*, eth_getBlockReceipts, eth_getTransactionReceipt, logs and the tx-count methods hide system txs. The by-index lookups were not covered by the compliance layer and still index the full block:

• eth_getTransactionByBlockHashAndIndex
• eth_getTransactionByBlockNumberAndIndex
• eth_getRawTransactionByBlockHashAndIndex
• eth_getRawTransactionByBlockNumberAndIndex

So index N returns a different tx than the one eth_getBlockBy* lists at position N (and than hl-node / rpc.hyperliquid.xyz return), often a system tx.

Example, mainnet block 49,493,297 (0x7543f6df…c2ee3), index 0x1:

	eth_getTransactionByBlockHashAndIndex
hl-node, rpc.hyperliquid.xyz	0x2a350753…622f7e (matches transactions[1] of the block)
nanoreth nb-20261007	0xfa20609f…3d04bf (a system tx)

This breaks Arbitrum Nitro nodes of Orbit chains that settle on HyperEVM: Nitro looks up batch transactions with eth_getTransactionByBlockHashAndIndex and then stalls, retrying forever on

failed to fetch batch mentioned by batch posting report: L1 client returned unexpected transaction hash 0xfa20… when looking up block 0x7543… transaction 1 with expected hash 0x2a35…

Fix

Add the four methods to EthBlockApi, following the same pattern as the other compliance overrides:

• shift the requested index by the block's system_tx_count before the lookup, and shift transactionIndex in the response back;
• for the by-number variants, resolve the tag (latest, …) to a concrete block number first, so the system tx count and the lookup refer to the same block;
• non-compliant mode is unchanged.

Testing

• Unit tests for the index mapping.
• On a mainnet node running this patch: 1,104 by-hash / by-number lookups over 150 random blocks from the last ~150k blocks compared against hl-node, 0 mismatches; out-of-range indexes return null like hl-node.
• A Nitro (Orbit) node that was stuck on the error above synced past it after switching to the patched build.
• cargo test passes. cargo clippy --workspace --tests --all-features with -D warnings on current stable (1.99) fails with clippy::double_must_use on every #[rpc] trait in the crate, including ones this PR doesn't touch (e.g. call_forwarder.rs, tx_forwarder.rs, sync_server.rs); no other warnings.

Related: #160 (another system-tx difference vs hl-node).

[gjwang]

Resolving latest once handles head advancement. Could the two by-number handlers also pin the resolved block hash across the lookup? At 5de251b, the helper returns a height and system-tx count, then each handler performs a separate number-based lookup.

If the two lookups can observe different canonical snapshots during a reorg, consider this regression case: header A at height N has one system tx; N is replaced by B with two before the lookup. Compliant index 0 is mapped using A’s count to full index 1, which on B is still a system tx. Both the structured and raw number-based methods need to keep the count and transaction tied to the same block. Carrying the resolved hash into a by-hash lookup, or rejecting/retrying a changed identity, would cover that case.

This is a source-review scenario, not a reproduced node reorg or a claim about the live validation reported above.

[icmono]

Good point, thanks. HyperEVM blocks are final, so I don’t expect this to happen in practice, but it’s cheap to rule out. In d46f5cb the by-number handlers resolve the block once to its hash plus system tx count (sealed_header_by_id), then do the lookup by hash. The count and the transaction now always come from the same block, and the by-number methods go through the same path as the by-hash ones.

Re-checked on a mainnet node running this commit: 177 lookups on 59 random recent blocks. By-number, by-hash and rpc.hyperliquid.xyz agree on all of them, and the raw variants agree too. Out-of-range indexes and unknown blocks return null.

[gjwang]

Thanks, that pins both indexed lookup paths to the same sealed header as the system-tx count. Agreed on finality—the example was a defensive consistency case, not an observed HyperEVM reorg. Appreciate you checking the raw variants too.

[sprites0]

Thanks! Landed on node-builder as 6aa2085 and f0cad31 (authorship preserved) and released in nb-20261110. Closing since the branch predates the CI clippy fix (#165).