Sequencer builder/config parity gaps vs morph-geth

[panos-xyz]

What happened

A pass over morph-geth miner, genesis morph config, and Engine API vs morph-reth shows several sequencer/builder knobs that geth honors and morph-reth either ignores, hard-codes, or only applies on the packing path.

This is the tracking issue. Child issues cover independently fixable slices.

What I expected

A morph-reth sequencer should produce blocks that a morph-geth validator would accept, and operators should have the same runtime levers they have on geth (miner.gaslimit, genesis tx/payload caps, L1-message overflow behavior).

Audit (geth vs morph-reth)

Must-fix (protocol / sequencer liveness)

Item	morph-geth	morph-reth today
Block header gasLimit target	--miner.gaslimit / miner_setGasLimit; each block ramps toward GasCeil by ~1/1024	Sequencer assemble always copies parent. Header is frozen at genesis (30M). --builder.gaslimit / --miner.gaslimit parse but are unused.
Genesis morph.maxTxPerBlock	Mainnet genesis is 100; miner stops packing at that count	Field is dropped when reading genesis. Only --morph.max-tx-per-block exists and defaults to unlimited.
Genesis morph.maxTxPayloadBytesPerBlock	122880; ValidateBody rejects oversized L2 payload	Parsed, and used as a builder packing cap via CLI default 122880. Not checked on import. A morph-reth validator would accept a block a geth validator would reject.
L1 message that does not fit remaining gas	Seal the block with what already fits; leftover L1 messages go to the next block	assembleL2Block fails the whole build. Sequencer can stall.

Should-fix (operator knobs / defaults)

Item	morph-geth	morph-reth today
Block-build time budget	--miner.newblock-timeout; unset sanitizes to 3s	Packing breaker is hard-coded 1s. Comment says use --builder.deadline, but that flag is not wired into the Morph packing breaker (reth default is 12s for a different job).
Miner gas-price floor when packing	--miner.gasprice / miner_setGasPrice, default 0.001 gwei	No equivalent packing floor. Txpool minimum_priority_fee is optional and separate.
miner.maxaccountsnum	Caps how many accounts the miner pulls from the pool	Missing.
miner RPC (setGasLimit, setGasPrice, setExtra)	Present (non-public namespace)	Missing. setExtra is a no-op on Morph anyway because extra-data must be empty.

Not a gap (checked)

• miner.extradata: Morph consensus zeros extra-data; both clients require empty extra.
• Engine API surface (assembleL2Block(V2), newL2Block(V2), validateL2Block, newSafeL2Block, setBlockTags): present.
• Fee vault, base-fee cap (10 gwei), 1/1024 gasLimit validation of imported headers: present.
• --morph.max-tx-payload-bytes packing cap default matches genesis 122880.
• eth_maxPriorityFeePerGas default was already aligned to geth's 0.001 gwei.

Steps to reproduce

1. Compare a morph-geth sequencer started with defaults against morph-reth on the same chain (mainnet genesis gasLimit 30M, maxTxPerBlock 100, maxTxPayloadBytesPerBlock 122880).
2. Try to raise block gasLimit at runtime (miner.setGasLimit) — geth ramps, morph-reth cannot.
3. Assemble a block without --morph.max-tx-per-block 100 — morph-reth may pack more than 100 txs.
4. Import a block whose L2 payload exceeds 122880 bytes — geth rejects, morph-reth accepts.

Additional context

This is not an OP-stack-style SystemConfig.gasLimit. Morph's consensus node copies gasLimit from assembleL2Block; the EL miner owns the target, same as geth. OP/Base put the value on op-node / L1 SystemConfig; Tempo/Scroll wire --builder.gaslimit on the EL. morph-reth currently does neither on the sequencer assemble path.

Child issues:

• header gasLimit target (this conversation)
• genesis maxTxPerBlock ignored
• payload size not consensus-validated
• L1 message gas overflow aborts assemble

[panos-xyz]

Child issues:

• Sequencer cannot change block gasLimit; header stays at parent/genesis #156 — header gasLimit target missing (cannot ramp the network gas limit)
• Genesis morph.maxTxPerBlock is ignored; sequencer may pack more txs than geth #157 — genesis maxTxPerBlock ignored
• assembleL2Block fails the whole block when an L1 message does not fit remaining gas #158 — assembleL2Block aborts when an L1 message does not fit remaining gas
• L2 payload size is not consensus-validated; geth validators would reject oversized blocks #159 — L2 payload size not consensus-validated (mixed-client split vs geth)

Lower-priority items (CLI no-ops, packing time budget, miner gas-price floor, miner RPC) stay on this tracking issue until we decide they need their own tickets.

[panos-xyz]

Follow-up: which geth limits are zkEVM leftovers (do not treat as must-port)

Morph proving is now SP1 zkVM over batches, not Scroll zkEVM circuits. Several geth packing/consensus caps exist because early Morph copied Scroll’s per-block circuit budget.

Do not restore as protocol / default-must-match-geth:

• morph.maxTxPerBlock (mainnet genesis 100) — geth comment: ECDSA signature checks that fit in the zkEVM. Hoodi already omitted it. SP1 shadow-prove caps are per-batch (SHADOW_PROVING_MAX_TXN default 200, SHADOW_PROVING_MAX_BLOCK default 600), not 100 txs/block. morph-reth’s unlimited default (CLI override only) is the right shape; Genesis morph.maxTxPerBlock is ignored; sequencer may pack more txs than geth #157 should not become “hard-code 100”.
• CheckCircuitCapacity — present on geth’s ethconfig, never wired into validation. Dead. Do not port.
• miner.maxaccountsnum — miner mempool-scan perf, not proving. Optional later, not parity-critical.

Keep as operator packing knobs, not zkEVM consensus:

• --morph.max-tx-payload-bytes (120KB) — historical Scroll calldata/circuit size. Today DA is compressed batches into blobs (~127KB usable per blob, often many L2 blocks). A packing cap is still useful for latency / blob filling / mixed geth validators, but 122880 does not need to be re-imposed as a Morph-only consensus rule unless we still mix geth validators that ValidateBody-reject oversize blocks (L2 payload size is not consensus-validated; geth validators would reject oversized blocks #159 is mixed-client, not prover).
• --morph.max-tx-per-block — keep optional. Do not default to genesis 100.

Still worth doing (not prover artifacts):

• Sequencer cannot change block gasLimit; header stays at parent/genesis #156 header gasLimit target (Ethereum-style elasticity, operational).
• assembleL2Block fails the whole block when an L1 message does not fit remaining gas #158 assemble abort when an L1 message does not fit (sequencer liveness).
• Mixed-client payload-size check (L2 payload size is not consensus-validated; geth validators would reject oversized blocks #159) only while geth validators remain; if we raise DA budget, raise it on both clients rather than fossilizing 120KB.