[aelmanaa]

Relates to #33.

Problem

Safe.getMultiSendCallOnly(client, chainId) reverts with MultiSendCallOnlyNotFound for every chain the official Safe Transaction Service does not serve. So every batch path (proposeTransactions, getExecTransactionsData, simulateTransactionsNoSign, simulateTransactionsMultiSigNoSign, ...) is unusable on those chains. The canonical MultiSendCallOnly contracts (v1.4.1 and v1.3.0, deployed through the singleton factory) are often present there anyway. For example, one project targets 5 Safe testnets, and 3 of them (0G Galileo, Ink Sepolia, Plume testnet) have no transaction service.

Solution

getMultiSendCallOnly now resolves in this order:

1. An explicit address set on the client with the new safe.withMultiSendCallOnly(address). Passing address(0) clears it.
2. The existing table for served chains, unchanged.
3. On the current chain only (chainId == block.chainid), the canonical v1.4.1 address if it has code, then v1.3.0.
4. Otherwise it reverts with MultiSendCallOnlyNotFound, as before.

Notes:

• getMultiSendCallOnly changes from pure to view because it now reads the client and checks code.
• Client gains a new multiSendCallOnly field. Existing call sites compile unchanged.
• The code check only runs for the current chain. Code on the forked chain says nothing about a different chainId argument.
• The README batch section documents the fallback and the override.

Testing

• Added 9 tests:
  • SafeMultiSendFallbackTest in test/Safe.t.sol forks Base, copies the real v1.4.1 and v1.3.0 bytecode, clears both addresses, sets vm.chainId(763373) (Ink Sepolia) and etches the code back as each case needs. The cases are: v1.4.1 only, v1.3.0 only, both (v1.4.1 wins), neither (reverts), a non-current chainId (reverts), a served chain (Base still resolves to v1.3.0), the override taking precedence, and the override cleared with address(0).
  • test/SafeSimulation.t.sol has one end-to-end batch simulation on an unserved chain id.
• forge test (forge 1.8.7): 38 passed, 0 failed. It was 29 before.
• Negative check: with the new resolution logic removed from src/Safe.sol, 6 of the new tests fail with MultiSendCallOnlyNotFound(763373). The other 3 are guards that pass either way. With all of src/ reverted, the tests do not compile because withMultiSendCallOnly does not exist.
• forge fmt --check passes.

🤖 Generated with Claude Code

[aviggiano]

Thanks for this. Looks good to merge once CI passes. I’ll add a check for custom MultiSend addresses with no code in a follow-up, since those can report success without executing the batch. I’ll also clarify that the override must point to a trusted contract.