Ethereum’s Next Upgrade Could Let Callers Skip Payments

Simulations under Ethereum's proposed Glamsterdam gas rules show a withdrawal succeeding while a 2.5% payment inside it is quietly dropped, and the caller's gas limit decides which happens.
Key Takeaways
The findings come from tests of EIP-8037 and EIP-8038, two gas-repricing proposals that led AllCoreDevs to pause the Hoodi testnet fork, according to the Ethereum Foundation’s November 6 update.
A Withdrawal That Pays Everyone but One
The clearest case comes from a proof-of-concept repository published under the GitHub account mejango and cited in the developer discussion. Its vault sends 2.5% of each withdrawal to a separate component contract inside a try block. If that payment reverts, the vault keeps the share and the user is paid anyway, a common way to stop one broken component from freezing every withdrawal.
The author ran the same call at gas limits from 30,000 to 900,000 using eth_simulateV1 on public RPC endpoints, once against Ethereum mainnet and once against Sepolia, which the repository says already applies Glamsterdam gas rules. Nothing was deployed. On mainnet, every limit either reverted the whole withdrawal or paid the component. Under Glamsterdam rules, a band roughly 112,000 gas wide completed the withdrawal and skipped the payment.
These are the author’s simulations. Coindoo has not reproduced them, and they do not describe an exploit of a live contract.
The Caller Chooses Which Outcome Happens
The outer gas limit is set by whoever calls the withdrawal. According to the repository, that turns the band into a lever: a caller can pick a limit inside it on purpose, making the component payment fail while the withdrawal itself goes through.
The vault has no way to respond. Its contracts are immutable, so logic written to tolerate a broken component now also tolerates a deliberately underfunded one. Under current mainnet rules the same attempt fails, because a limit too low to fund the component also leaves too little gas to finish, and the whole transaction reverts.
Why the Payment Fails but the Fallback Survives
Two proposals produce that result together. EIP-8038 raises the gas cost of reading and writing state, so a budget that used to cover the component’s work can now run out sooner. EIP-8037 separates the cost of creating new state from ordinary execution gas and raises it. Coindoo has covered what that means for onboarding costs.
The second change is what keeps the fallback alive. ChainSecurity’s October 13 analysis explains that state-creation charges are drawn from a separate pool that is restored when a call fails, and that a failed call’s state charges paid from regular gas go back to its caller. A contract can therefore have more gas left after a failure than it would today.
ChainSecurity’s worked example uses a write to a fresh storage slot. Under current rules it costs 22,100 gas, so a failure handler with about 15,000 gas left runs out and the transaction reverts. Under the reviewed Glamsterdam specification, the same write costs 12,100 execution gas, while a 97,920-gas state charge comes from the separate pool, leaving room for the write and a short continuation. The write is not cheaper overall; a different pool pays for the expensive part.
The vault scenario needs both effects: the component’s call fails earlier because work costs more, and the vault’s fallback completes because it no longer runs out of gas. A contract that counted on running out of gas to force a rollback can end in a different state. ChainSecurity limits that conclusion to specific call patterns, and most findings reported so far stem from the cost increases rather than the separate pool, according to the summary of the November 16 testing call.
When the Default Gas Estimate Lands in the Gap
A higher fee per unit of gas changes nothing here, because price does not add units to an internal call. A higher gas limit does help in the vault example, since larger limits pay the component, but only when the software setting the limit chooses a large enough number. The repository’s second test shows what happens when it uses the standard estimate.
That test models an OP Stack withdrawal being finalized on Ethereum, using the real OptimismPortal2 and L1CrossDomainMessenger contracts. The message being delivered requests a fixed 375,000 gas. On mainnet, finalizing at the eth_estimateGas result of 931,416 delivered the message. Under Glamsterdam rules on Sepolia, finalizing at the estimate of 1,517,344 succeeded, but the messenger recorded the delivery as failed and the funds were not credited. Doubling the limit delivered it.
The repository presents that estimate as the one a wallet or relayer would send, so the user would have done nothing unusual. Recovery remains possible because the messenger lets anyone replay a failed message, but it takes a third, manual transaction and another gas payment. The two runs also used different messenger versions, 2.11.1 on mainnet and 2.13.0 on Sepolia, a difference the comparison does not isolate.
The two gas sweeps side by side
Vault withdrawal – current mainnet rules
- 30,000–106,000: whole withdrawal reverts
- 108,000–898,000: component paid
Vault withdrawal – Glamsterdam rules
- 30,000–180,000: whole withdrawal reverts
- 182,000–294,000: withdrawal succeeds, payment skipped
- 296,000: reverts
- 298,000–898,000: component paid
OP Stack finalization – mainnet
- 400,000–900,000: reverts, retry later
- 925,000–10,000,000: message delivered
OP Stack finalization – Glamsterdam rules
- 400,000–1,000,000: reverts
- 1,025,000–1,600,000: finalized, delivery recorded as failed
- 1,625,000–10,000,000: message delivered
Glamsterdam results were simulated on Sepolia. OP Stack runs were pinned to a single L1 block.
What a Success Receipt Does Not Show
In both tests, the transaction receipt reports success. A receipt’s status covers only the outer call. It says nothing about whether a call caught inside a try block, or a message handed to a bridge messenger, did what it was meant to do.
Confirming that means checking the specific evidence: a balance change or event from the component in the vault case and, for OP Stack withdrawals, whether the messenger emitted RelayedMessage or FailedRelayedMessage. A bridge interface that marks a withdrawal “finalized” from the receipt alone can show a user a completed withdrawal before any funds arrive.
10,000-15,000 Contracts, and Who Can Fix Them
At the November 16 testing call, the Ethereum Foundation’s Marius estimated that around 10,000–15,000 out of roughly 90 million contracts would be affected, according to the meeting’s auto-generated summary. The summary does not explain how that figure, or a separate value-at-risk estimate of “a few hundred thousand dollars,” was produced. A contract count is not an application count, and an affected contract does not necessarily put funds at risk.
Auditors brought their own tallies to the same call. ChainSecurity reported 152 findings, rising to 200 when combined with the Foundation’s internal search. The summary describes most findings as easy to mitigate, with some requiring contract upgrades or redeployment.
Who can apply a fix depends on where the outdated gas assumption sits:
- Wallets and RPC providers can change how they estimate gas, the layer behind the OP Stack result.
- Applications with adjustable gas settings can raise those values and retest them.
- Upgradeable contracts can ship a new implementation, usually after a governance vote.
- Immutable contracts cannot be patched. Their options are a supported retry, an external workaround, moving users to a replacement contract, or a change to the protocol rules themselves.
That last group is why the issue sits with core developers rather than app teams alone, a compatibility question Coindoo examined in its earlier look at how the upgrade could change old contracts. The testing-call summary also assigns follow-up work with affected app teams, L2s and OP Stack chains, including testnet replays to find contracts not yet identified.
What the November 19 Call Can Settle
Core developers take up the findings at the November 19 execution-layer call. The Foundation says no decision is pre-committed for that meeting, while the testing-call summary lists deciding whether to proceed with EIP-8037 and EIP-8038 as the action item.
Options raised on November 16 included proceeding as planned, dropping the proposal, waiting for contract-size limit increases, and delaying. Those favouring the current plan pointed to the post-quantum roadmap, and one developer warned that dropping the proposal would prevent large contracts from being deployed. The Foundation’s update adds that several specification revisions are being prototyped, each documented for the problems it fixes and the other invariants it could break.
Hoodi stays unforked until that is resolved. The gas bands above describe the proposal as tested, and any revision to the specifications would change the rules those simulations ran against.









