
Here's my email to galsterdam@ethreum.org describing @juiceboxETH and @revnets Glam implications. Hi, We simulated Juicebox V6 against Glamsterdam (EIP-8037/8038) on Sepolia chains (Ethereum, Base, Optimism, Arbitrum). The pattern to watch is a fail-open try/catch around an external call. With a tuned gas limit, the callee can run out of gas while the outer call succeeds and treats that as an ordinary failure. BACKGROUND Under EIP-150, each call level keeps 1/64 of its gas. When a deep call runs out, the levels above revert and hand that gas back. Spent gas is gone, but the outer call can finish on what was never spent. Contracts can forward less, but can't turn this off. This happens on mainnet today, and contracts can be tuned to today's gas constraints so these accidental revert-then-continues don't happen. If the constraints change, so does the execution, and immutable contracts can't follow. WHAT GLAMSTERDAM CHANGES Two changes can let the outer call finish where it couldn't before: 1. New slots go from about 22k to about 98k gas, which mostly hits callees, not the work after a catch. 2. State gas spent by a reverting frame can return mid-transaction, which can't happen today. FINDING 1: OP STACK L1 MESSENGER (NOT OUR CODE) L1CrossDomainMessenger reserves 40k gas to record a failed relay, which is less than a new slot now costs. If the target runs out of gas, the failure may go unrecorded. The portal finalizes the withdrawal with no replay, and the ETH stays in the L1 bridge. github.com/ethereum-optimism… Our first L2 to L1 root is lost at 1.06M to 2.00M gas on OP and 1.02M to 1.97M on Base. Our 375k message limit is too low, and an unbuffered eth_estimateGas hit that range. With an all-gas-burning target, it's lost up to 7.7M. Ethereum mainnet records the failure from about 960k. github.com/Bananapus/nana-su… These contracts are on L1, so exposure starts when L1 activates Glamsterdam, whether or not the L2s adopt it. We haven't contacted OP or Base yet. FINDINGS 2 TO 5: JUICEBOX V6 Our core and hook try/catch calls set no gas limit or reserve, so they keep only the 1/64. The native-token paths below showed no window on the mainnets we swept, because their callees were too cheap to starve. All of them have a window on Glamsterdam Sepolia. 2. Payouts: anyone can call sendPayoutsOf. At 878k to 1.03M gas, a split to another project fails, the transaction succeeds, and the money returns to the project with the payout limit used. On mainnet, a simple test hook needs 3.2M+ gas. Routed token splits weren't swept. github.com/Bananapus/nana-co… 3. NFTs: at 2.85M to 3.11M gas, a buyer keeps the NFT while the tier's split payment fails. The recipient gets nothing, the money goes to the project instead, and there's no retry. github.com/Bananapus/nana-72… 4. Reserved tokens: at 488k to 815k gas, a split meant to pay another project fails, so the beneficiary gets the raw tokens instead. The destination project never receives the payment. github.com/Bananapus/nana-co… 5. Fees: at 678k to 819k gas, the 2.5% cash-out fee call fails and the fee is forgiven. The holder's reclaim doesn't change, and the fee stays in the project as fee-free surplus. github.com/Bananapus/nana-co… METHOD We used eth_simulateV1 on fresh test projects, so real projects will differ. The bridge test used the real portal with an overridden proof and a synthetic message. Nothing was sent on Sepolia. The messenger cause is inferred from the code and the numbers, not traced. QUESTION Fail-open try/catch assumes a callee with a reasonable gas budget won't starve. If the parent can finish after the callee runs out of gas, the caller chooses the outcome. Our V6 contracts are immutable, so a fix means redeploying. In practice, the latest version of the protocol, deployed in June, currently holds about $140k, all of which looses its value prop given a faulty protocol under Glam. The greater threat is the uncertainty Glam creates for protocol clients going forward. It's impossible to trust an immutable protocol like Juicebox and Revnets if failures caused by changes to the underlying network are fair game. Earlier versions of the protocol have facilitated over $200m of ETH volume, including for significant moments like ConstitutionDAO, AssangeDAO and the Tornado Cash legal defense fund. Those versions no longer facilitate new projects, but they carry Glamsterdam risks similar to V6's. What is our course of action? We can share the simulation scripts and the exact eth_simulateV1 requests if they'd help. Thanks, Jango Juicebox
Bruhva. I’ve read this. Less passive aggressive way to respond would be just to affirm the email is in fact the best way… and best service would be to translate the twitter postings into formal tickets for me. Same team. I’ll send the email.















