Glamsterdam devnet 6 eips

[Marchhill]

Merge without squashing

Changes

Implements glamsterdam devnet-6. Constituent PRs (already reviewed and merged into this branch):

• EIP-8246: Remove SELFDESTRUCT burn (#12213)
• EIP repricings & existing-EIP updates (2780/8037/8038 + fixes) (#12214)
• EIP-8282: Builder execution requests (#12215)
• targetGasLimit in PayloadAttributesV4 (#12216)
• Stateless executor: tests-zkevm@v0.5.0 alignment (#12217)
• fix(specs): BPO3 inherits BPO2, not Amsterdam; run BPO transition pyspec fixtures (#12312)
• Fix state gas spill with negative reservoir (#12338)

Also updates the EF fixture pins (tests-glamsterdam-devnet@v6.1.1, tests-zkevm@v0.5.0) and merges latest master + fixes.

Types of changes

What types of changes does your code introduce?

• New feature (a non-breaking change that adds functionality)
• Bugfix (a non-breaking change that fixes an issue)

Testing

Requires testing

• Yes

If yes, did you write tests?

• Yes

Notes on testing

Covered by the EF tests-glamsterdam-devnet v6.1.1 fixtures (state/blockchain/engine) and unit tests; stateless execution covered by the tests-zkevm v0.5.0 fixtures.

Documentation

Requires documentation update

• No

Requires explanation in Release Notes

• No

References

• glamsterdam-devnet-6 spec notes

[claude]

Claude finished @Marchhill's task in 1m 14s —— View job

---

Review

• Gather context and understand the change
• Analyze CI status
• Post review

---

The change is a straightforward two-line constant update, bumping the pyspec test fixture reference from bal@v7.2.0 to glamsterdam-devnet@v6.0.0. The logic in LoadPyspecTestsStrategy.cs that consumes these constants is correct and unchanged — no issues there.

---

Medium — Three test CI failures on the June 22 commit

The previous CI run (triggered on June 22 before the upstream merge) shows three failures:

Workflow	Result
Nethermind/Ethereum tests	❌ failure
Nethermind extra test variants	❌ failure
Nethermind tests (Flat DB)	❌ failure

These are exactly the suites that exercise pyspec/EF tests. The June 24 re-run (triggered when the PR was opened after merging upstream/master) has all three queued but not yet complete. Please confirm the current CI run passes, or add a brief note if failures are expected (e.g., incomplete EIP implementations in this devnet branch).

---

Low — PR description is unfilled

The PR body is the blank template. Even a one-liner ("Updates pyspec fixture constants to glamsterdam-devnet v6.0.0") would help reviewers and the changelog.

---

No code issues found with the change itself — the constant values follow the established naming pattern and the constructed download URL has the correct shape (https://github.com/ethereum/execution-specs/releases/download/tests-glamsterdam-devnet@v6.0.0/fixtures_glamsterdam-devnet.tar.gz).

[github-actions]

EVM Opcode Benchmark Diff

Aggregated runs: base=3, pr=3
Noisy rerun opcodes: CREATE, EXTCODECOPY, EXTCODEHASH, MCOPY

Regressions (2)

Opcode	Base Median (ns)	PR Median (ns)	Delta	Abs Δ (ns)	Base CV	PR CV	Threshold	Uncertainty	Effective
DIV	296.950	338.390	+13.96%	41.440	2.0%	1.8%	±5.0%	±2.8%	±5.0%
MULMOD	232.890	1049.460	+350.62%	816.570	2.3%	1.1%	±5.0%	±4.8%	±5.0%

Improvements (2)

Opcode	Base Median (ns)	PR Median (ns)	Delta	Abs Δ (ns)	Base CV	PR CV	Threshold	Uncertainty	Effective
EXTCODECOPY	346.340	227.500	-34.31%	118.840	1.3%	3.4%	±5.0%	±2.5%	±6.8%
EXTCODEHASH	530.200	410.900	-22.50%	119.300	2.6%	1.4%	±5.0%	±2.5%	±5.2%

[claude]

Claude finished @Marchhill's task in 11m 6s —— View job

---

PR Review — Glamsterdam devnet-6 EIPs

• Gather context and review diff vs origin/master
• Review EVM / gas-policy changes (EIP-8037/8038/2780)
• Review execution requests & payload attributes (EIP-8282, targetGasLimit)
• Review SELFDESTRUCT burn removal (EIP-8246)
• Review spec wiring & stateless executor
• Post review summary

---

Reviewed the devnet-6 EIP implementation across the EVM two-dimensional gas model (EIP-8037/8038), repricings (EIP-2780), SELFDESTRUCT burn removal (EIP-8246), builder execution requests (EIP-8282), and targetGasLimit plumbing. The constituent PRs were individually reviewed/merged and this branch is EEST-fixture-pinned (v6.1.1 / zkevm v0.5.0). Overall the change is coherent and internally consistent; findings below are concentrated in the state-gas "reservoir" accounting, which is the highest-residual-risk area.

Note: I couldn't compute a git merge-base for a precise diff on this shallow checkout, so this review is based on reading the changed files and their call sites directly.

---

Medium — Asymmetric clamp in RemoveStateGasRefundFromReservoir with a negative reservoir

src/Nethermind/Nethermind.Evm/GasPolicy/EthereumGasPolicy.cs:399-409

long fromReservoir = Math.Min(amount, gas.StateReservoir);
gas.StateReservoir -= fromReservoir;
amount -= fromReservoir;
if (amount > 0) { gas.StateGasUsed -= Math.Min(amount, gas.StateGasUsed); }

amount is always positive (caller RemoveAdvancedStateGasRefund guards StateGasRefundAdvanced > 0), but gas.StateReservoir can legitimately be negative on a net-spill (set via ConsumeStateGas's Math.Min(0, reservoir) at :143, and propagated through Refund/RestoreChildStateGas). When the reservoir is negative, Math.Min(amount, negativeReservoir) returns the negative value, so:

• gas.StateReservoir -= fromReservoir pushes the reservoir up toward 0 (spuriously reclaiming spilled state gas), and
• amount -= fromReservoir increases the residual, over-deducting StateGasUsed by |reservoir|.

The sibling routine DiscardStateGas (:376) clamps its equivalent with Math.Max(0, …); this one does not. Both review passes independently landed on this asymmetry as the top concern.

Reachability is narrow: a sub-frame that both spilled state gas (reservoir negative) and advanced a speculative state-gas refund (CreditStateGasRefund → AddStateGasRefundToReservoir), then reverts/exceptionally-halts. PopAndRestoreParentState (VirtualMachine.cs:646) → RemoveAdvancedStateGasRefund (:720-724) runs before RestoreChildStateGasOnHalt merges the child into the parent, so a corrupted StateReservoir/StateGasUsed would flow into the tx's committed block state gas — a consensus-relevant quantity feeding the 2D block-gas check (BlockAccessListManager.Validation.cs:84,89).

I could not fully confirm the intended semantics from the code alone (the reservoir model is intricate and this exact nesting may simply not arise), so treat this as plausible, needs confirmation rather than a proven break. Suggested action: add a targeted unit test for "sub-frame spills state gas + advances a refund + halts/reverts" and confirm the resulting block state gas matches the reference. If the current behavior is intentional, a one-line comment explaining why the clamp differs from DiscardStateGas would close the gap.

Fix this →

---

Low — SELFDESTRUCT new-account charge has no EIP-2780 branch (unlike CALL)

src/Nethermind/Nethermind.Evm/Instructions/EvmInstructions.ControlFlow.cs:244-254

The CALL path special-cases EIP-2780/8038 when deciding the dead-beneficiary NEW_ACCOUNT charge (EvmInstructions.Call.cs:174-180: spec.IsEip8038Enabled ? … : !spec.IsEip2780Enabled && (…)). The SELFDESTRUCT path computes chargesNewAccount purely from spec.ClearEmptyAccountWhenTouched with no 2780 branch. On the real glamsterdam fork EIP-8038 is co-activated (so SELFDESTRUCT takes the IsEip8038Enabled branch and the fixtures cover it), which is why this is Low — but a 2780-without-8038 spec configuration would charge a legacy NEW_ACCOUNT cost that CALL suppresses. Worth a confirming comment that the two paths are intended to diverge for that config, or aligning them.

---

Low — ConsumeStateGas doesn't set the OutOfGas flag on failure

src/Nethermind/Nethermind.Evm/GasPolicy/EthereumGasPolicy.cs:137-141

On the spill path, a failed TryConsume returns false but (unlike UpdateGas/Consume) leaves gas.Value untouched and never sets gas.OutOfGas. Every current caller propagates the boolean into an explicit goto OutOfGas/SetOutOfGas, so no live path is affected, but a future caller that checks IsOutOfGas instead of the return value would mis-handle a state-gas OOG. Recommend either setting the flag defensively or documenting the return-value contract on the method.

---

Low — Release-mode silent underflow in CreateAvailableFromIntrinsic

src/Nethermind/Nethermind.Evm/GasPolicy/EthereumGasPolicy.cs:550

Eip7825Constants.DefaultTxGasLimitCap - intrinsicGas.Value wraps (to a huge ulong) if intrinsic regular gas exceeds the 16 MiB cap; guarded only by a Debug.Assert plus the upstream EIP-7825 ExceedsCap inclusion check. Defense-in-depth, not a live bug — but a SaturatingSub here would match the saturating arithmetic used elsewhere in the file.

---

Low — Duplicated destroy-finalization blocks in TransactionProcessor

src/Nethermind/Nethermind.Evm/TransactionProcessing/TransactionProcessor.cs:373-393 and :1381-1401

The deferred (EIP-8037+7708) and non-deferred destroy-finalization paths duplicate the DeleteAccount → CreateAccount(addr, balance) delete/recreate logic under RemoveSelfdestructBurn, differing only in log type (CreateBurn vs CreateSelfDestruct). No behavioral bug (balance is 0 at finalization since the opcode subtracts first), but the duplication is a maintenance hazard — a fix to one block can silently miss the other. Consider extracting a shared helper.

---

Verified correct (no action)

• EIP-8246 SELFDESTRUCT burn removal — self-target with RemoveSelfdestructBurn correctly goto Stop (no log, no SubtractFromBalance); the finalization delete+recreate preserves balance with no double-subtract/double-log (ControlFlow.cs:268-276).
• The Fix state gas spill with negative reservoir #12338 negative-reservoir spill fix in ConsumeStateGas/CalculateStateGasSpill is internally consistent (reservoir + spill portions sum exactly to stateGasCost).
• Eip8037BlockGasInclusionCheck.Validate early > blockGasLimit rejections and strict-> comparisons are correct.
• EIP-8282 wiring: builder requests are dequeued after withdrawal/consolidation so the flat encoding stays in ascending type order (ExecutionRequestsProcessor.cs:103-113), MaxRequestsCount bumped to 5, builder predeploys added to IsSystemContract for BAL validation, and the stateless path correctly no-ops RecordRequests.
• PayloadAttributes V4 targetGasLimit is included in the payload-id hash and field validation; TargetAdjustedGasLimitCalculator uses saturating arithmetic for the bound-divisor step.
• Gas constant arithmetic (GasCostOf.cs, Eip803*Constants.cs) checks out with no long overflow for realistic inputs.

---

Recommendation: The Medium finding is plausible-but-unconfirmed and sits in a consensus-relevant hot path; I'd ask for either a targeted regression test proving the negative-reservoir + advanced-refund + halt path is handled, or an explicit acknowledgement that the asymmetry with DiscardStateGas is intentional, before merge. Everything else is Low.

[Marchhill]

@claude will adress comments in follow up PR

[claude]

Claude finished @Marchhill's task in 19s —— View job

---

Understood — noted that the review findings will be addressed in a follow-up PR.

For reference, the open items from the earlier review, in priority order:

• Medium — Asymmetric clamp in RemoveStateGasRefundFromReservoir with a negative reservoir (EthereumGasPolicy.cs:399-409). The Math.Min(amount, gas.StateReservoir) goes wrong when StateReservoir is negative (net spill), unlike the Math.Max(0, …) clamp in the sibling DiscardStateGas. Worth a targeted regression test for "sub-frame spills state gas + advances a refund + halts/reverts" to confirm whether it reproduces.
• Low — SELFDESTRUCT new-account charge has no EIP-2780 branch (unlike CALL) (EvmInstructions.ControlFlow.cs:244-254).
• Low — ConsumeStateGas doesn't set the OutOfGas flag on failure (EthereumGasPolicy.cs:137-141).
• Low — Release-mode silent underflow in CreateAvailableFromIntrinsic (EthereumGasPolicy.cs:550).
• Low — Duplicated destroy-finalization blocks in TransactionProcessor (TransactionProcessor.cs:373-393 and :1381-1401).

This PR is already approved by @flcl42, so no action needed here from me — ping me on the follow-up PR if you'd like a re-review of the fixes.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 0.00%. Comparing base (1aae065) to head (148a794).
⚠️ Report is 9 commits behind head on master.

Additional details and impacted files

@@             Coverage Diff             @@
##           master   #12117       +/-   ##
===========================================
- Coverage   77.94%        0   -77.95%     
===========================================
  Files        2972        0     -2972     
  Lines      126482        0   -126482     
  Branches    19079        0    -19079     
===========================================
- Hits        98581        0    -98581     
+ Misses      24565        0    -24565     
+ Partials     3336        0     -3336     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.