Simulation methods: fields whose EIP is not active at the target block

[MysticRyuujin]

What happened

geth #35588 added checks to eth_call and friends that reject an accessList before Berlin, blobVersionedHashes before Cancun and an authorizationList before Prague. Erigon ported it in erigontech/erigon#23700 and their mainnet test suite broke on 7 pre-Berlin eth_createAccessList fixtures. That raised the question of what the spec expects. Right now it says nothing.

What clients do today

Measured with hive rpc-compat on the shared test chain at block 20 (pre-Berlin, pre-Cancun, pre-Prague). ethrex has no state before London on that chain, so its numbers are from block 0. Images: ethpandaops/* master/main as of 2026-09-08, plus the Erigon PR branch built from source.

eth_call with a caller-supplied field:

Client	accessList pre-Berlin	blobVersionedHashes pre-Cancun	authorizationList pre-Prague
geth master	rejects, -32000	rejects, -32000	rejects, -32000
reth	rejects, -32003	rejects, -32003	rejects, -32003
Besu	rejects, -32602	rejects, -32602	rejects, -32602
Nethermind	rejects, -32603 "Internal error"	rejects, -32603 "Internal error"	rejects, -32603 "Internal error"
Erigon main	accepts	accepts	rejects, -32000
Erigon #23700	rejects, -32000	rejects, -32000	rejects, -32000
ethrex	accepts	accepts unless maxFeePerBlobGas is set, then -32015	rejects, -32015

An empty accessList: [] behaves the same as a populated one everywhere.

eth_createAccessList pre-Berlin with no accessList in the request:

Client	plain transfer	contract call that touches storage
geth master	fails, -32000	fails, -32000
reth	fails, -32003	fails, -32003
Besu	{"accessList":[],"gasUsed":"0x5208"}	success body with accessList: [], error: "execution reverted", gasUsed: 0x2faf081 (gas cap + 1)
Nethermind	{"accessList":[],"gasUsed":"0x5208"}	{"accessList":[],"gasUsed":"0xbcf4"}
Erigon main and #23700	{"accessList":[],"gasUsed":"0x5208"}	one entry, two storage keys, gasUsed: 0xd52c
ethrex	{"accessList":[],"gasUsed":"0x5208"}	one entry, two storage keys, gasUsed: 0xbf4c

Post-Berlin controls at latest pass on every client with identical results, so the divergence is the fork boundary, not the fixtures.

Proposal

One rule, no special cases. A simulation request that carries a field whose EIP is not active at the target block is invalid and the client rejects it. This applies to eth_call, eth_estimateGas, eth_createAccessList, eth_simulateV1 and debug_traceCall, and to accessList, blobVersionedHashes and authorizationList. It also applies to eth_createAccessList itself on a block where EIP-2930 is not active, since the method has nothing meaningful to return there.

This matches what four of six clients already do for eth_call, and what Erigon does after #23700. Changes needed: ethrex adds the gate, Nethermind turns its internal error into a normal rejection, Besu stops reporting the eth_createAccessList failure as a success, Erigon drops the exception in #23700 and updates its fixtures.

The alternative is to require eth_createAccessList to succeed regardless of fork. That needs fixes in geth, reth and Besu and still leaves three clients disagreeing on the result, so I am not proposing it.

Error codes are out of scope here. They differ across clients today and belong in #784.

Testing

The shared chain already covers this: Berlin activates at block 24, Cancun at 42, Prague at 45. Once the wording lands I will add fixtures targeting block 20 for the three fields and for eth_createAccessList.

Related: #853 and #854 cover a different eth_createAccessList problem, fee defaults, and stay separate.