Glamsterdam devnet 8 specific commits

[flcl42]

Resolves #12599

Changes

• Update Glamsterdam devnet tests to v8.1.0
• feat(eip-2780): fold EIP-7708 transfer log cost into TX_VALUE_COST
• Apply final EIP-8038 state-access repricing

[claude]

Claude finished @flcl42's task in 13m 32s —— View job

---

Code review

• Gather context (diff of the 3 PR commits)
• Review production changes (Eip8038Constants, GasCostOf, EthereumGasPolicy, IntrinsicGasCalculator)
• Cross-check every updated test expectation against the new constants
• Check CI signal (this PR + the source PR Apply final EIP-8038 state-access repricing #12633)
• Post findings

---

Summary

Three devnet-8 commits: EEST fixture bump to v8.1.0, the EIP-2780 TX_VALUE_COST fold, and the final EIP-8038 repricing (COLD_STORAGE_ACCESS 3000→2100, ACCOUNT_WRITE 8000→9000, access-list entries → cold − warm, CREATE_ACCESS → ACCOUNT_WRITE + COLD_ACCOUNT_ACCESS).

I verified the arithmetic of essentially every updated expectation, and it holds up well:

• TxValueCostEip2780 = 6000 is exactly 4244 + 1756, so non-create value transfers are unchanged — the fold is a pure refactor there, and the 20600 / 14600 test names still hold.
• The Eip7981Tests tie retune is exact: with 115 zero bytes + 1 address + 2 keys, standard = 15000 + 460 + (6900 + 16×336) = 27736 and floor = 15000 + (460+336)×16 = 27736. A genuine tie, and it needed retuning because the old 50-byte/1-address case stops tying at addressCost = 2900.
• Every Eip8037* delta reconciles: −900 (cold storage), +1000 (CREATE_ACCESS / ACCOUNT_WRITE), +864 (SClear refund), −984 (63/64 of the +1000 CALL_VALUE bump), and the ×4/5 scaling where the EIP-3529 refund cap binds.
• PerAuthBaseExecution correctly stays at 7816 (ColdAccountAccess unchanged), and no dangling references to the removed TransferLogEip2780 remain.

Findings are inline; nothing blocks on correctness of the implementation — the open questions are all "does this match the devnet-8 spec text", which the code can't answer on its own.

#	Severity	Finding
1	Medium	Create-with-value: the EIP-7708 transfer log is now unpriced, and the code comment justifies the change with the wrong component
2	Medium	ColdStorageAccess silently reprices SClearEip8038 12480 → 11616 via an undocumented 4800/5000 ratio that disagrees with the EIP-3529 structural derivation
3	Medium	The v8.1.0 fixture bump is not green — see below
4	Low	Access-list cold − warm departs from EIP-2930's cold − 2×warm; no EIP citation
5	Low	Duplicated rationale in Eip2780ExtraGas

3. The fixture bump is not green (Medium)

CI on this PR is currently useless — nearly every nethtest shard failed with Internal Server Error occurred while resolving "actions/checkout@v6" etc., i.e. a GitHub Actions infra outage, not test failures. So I went to the run for #12633, which carries the same tree (glamsterdam-devnet-8, run 31107564873):

• txTest: 15 failures, all forks (Frontier through Amsterdam), all test_tx_nonce_overflow: Expected 'TransactionException.NONCE_OVERFLOW' but got 'NonceTooHigh: Nonce exceeds max nonce'. This is an exception-mapping gap surfaced by the v8.1.0 bump, not a gas issue — but it reproduces on all three variants (checked, flat, no-intrinsics) and should be fixed or explicitly tracked as a known-failure before the branch is promoted.
• engineTest: 1 failure in each of 4 shards (15664/15665 passing) plus zkevmTest failing. I couldn't extract the individual case names from the artifacts in this sandbox, so these need a look — if any of them is an Amsterdam gas vector it's directly relevant to findings 1 and 2.

Good news: Nethermind.Merge.Plugin.Test, Nethermind.Merge.AuRa.Test and Nethermind.Evm.Test all passed on that run, which confirms the updated engine-module hashes are correct (including the eip8037Enabled: true cases whose expectations are unchanged — worth a mental note that those cases appear insensitive to this repricing, so the repriced paths are covered only by the eip8037Enabled: false cases there).

Note on test design

Eip8038IntrinsicGasTests expresses its expectations in terms of the constants under test, so it can't catch a wrong constant — Eip8038ConstantsTests.Derived_parameters_match_the_eip8038_derivations (which pins 2900 / 2000 / 11616 / 12000 as literals) is the only real guard. That's fine as-is, but it means the EEST fixtures are doing all the real validation, which is why finding 3 matters.

Verdict

The implementation is internally consistent and the test updates are correct. Blocking items are the three Medium findings — two need a spec citation (cheap to resolve), and one needs the v8.1.0 failures triaged. DO NOT MERGE is already on the PR, so this is guidance for promoting the devnet branch rather than for merging to master.
· branch glamsterdam-devnet-8-consensus

[github-actions]

EVM Opcode Benchmark Diff

Aggregated runs: base=3, pr=3
Noisy rerun opcodes: EXTCODESIZE, JUMPDEST

No significant regressions or improvements detected.