EIP-7702: Set Code for EOAs

Add a new tx type that permanently sets the code for an EOA

Abstract

Add a new EIP-2718 transaction type that allows Externally Owned Accounts (EOAs) to set the code in their account. This is done by attaching a list of authorization tuples – individually formated as [chain_id, address, nonce, y_parity, r, s] – to the transaction. For each tuple, a delegation indicator (0xef0100 || address) is written to the authorizing account’s code. All code executing operations must load and execute the code pointed to by the delegation.

Motivation

Despite great advances in the smart contract wallet ecosystem, EOAs have held back broad adoption of UX improvements across applications. This EIP therefore focuses on adding short-term functionality improvements to EOAs which will allow UX improvements to permeate through the entire application stack. Three particular features this EIP is designed around are:

• Batching: allowing multiple operations from the same user in one atomic transaction. One common example is an ERC-20 approval followed by spending that approval. This is a common workflow in DEXes that requires two transactions today. Advanced use cases of batching occasionally involve dependencies: the output of the first operation is part of the input to the second operation.
• Sponsorship: account X pays for a transaction on behalf of account Y. Account X could be paid in some other ERC-20 for this service, or it could be an application operator including the transactions of its users for free.
• Privilege de-escalation: users can sign sub-keys and give them specific permissions that are much weaker than global access to the account. For example, a permission to spend ERC-20 tokens but not ETH, or to spend up to 1% of the total balance per day, or to interact only with a specific application.

Specification

Parameters

Set code transaction

A new EIP-2718 transaction known as the “set code transaction” is introduced, where the TransactionType is SET_CODE_TX_TYPE and the TransactionPayload is the RLP serialization of the following:

rlp([chain_id, nonce, max_priority_fee_per_gas, max_fee_per_gas, gas_limit,
destination, value, data, access_list, authorization_list, signature_y_parity,
signature_r, signature_s])

authorization_list = [[chain_id, address, nonce, y_parity, r, s], ...]

The fields chain_id, nonce, max_priority_fee_per_gas, max_fee_per_gas, gas_limit, destination, value, data, and access_list of the outer transaction follow the same semantics as EIP-4844. Note, this implies a null destination is not valid.

The signature_y_parity, signature_r, signature_s elements of this transaction represent a secp256k1 signature over keccak256(SET_CODE_TX_TYPE || TransactionPayload).

The authorization_list is a list of tuples that indicate what code the signer of each tuple desires to execute in the context of their EOA. The transaction is considered invalid if the length of authorization_list is zero.

The transaction is also considered invalid when any field in an authorization tuple cannot fit within the following bounds:

assert auth.chain_id < 2**256
assert auth.nonce < 2**64
assert len(auth.address) == 20
assert auth.y_parity < 2**8
assert auth.r < 2**256
assert auth.s < 2**256

The EIP-2718 ReceiptPayload for this transaction is rlp([status, cumulative_transaction_gas_used, logs_bloom, logs]).

Behavior

The authorization list is processed before the execution portion of the transaction begins, but after the sender’s nonce is incremented.

For each [chain_id, address, nonce, y_parity, r, s] tuple, perform the following:

1. Verify the chain ID is 0 or the ID of the current chain.
2. Verify the nonce is less than 2**64 - 1.
3. Let authority = ecrecover(msg, y_parity, r, s).
   • Where msg = keccak(MAGIC || rlp([chain_id, address, nonce])).
   • Verify s is less than or equal to secp256k1n/2, as specified in EIP-2.
4. Add authority to accessed_addresses, as defined in EIP-2929.
5. Verify the code of authority is empty or already delegated.
6. Verify the nonce of authority is equal to nonce.
7. Add PER_EMPTY_ACCOUNT_COST - PER_AUTH_BASE_COST gas to the global refund counter if authority is not empty.
8. Set the code of authority to be 0xef0100 || address. This is a delegation indicator.
   • If address is 0x0000000000000000000000000000000000000000, do not write the delegation indicator. Clear the account’s code by reseting the account’s code hash to the empty code hash 0xc5d2460186f7233c927e7db2dcc703c0e500b653ca82273b7bfad8045d85a470.
9. Increase the nonce of authority by one.

If any step above fails, immediately stop processing the tuple and continue to the next tuple in the list. When multiple tuples from the same authority are present, set the code using the address in the last valid occurrence.

Note, if transaction execution results in failure (e.g. any exceptional condition or code reverting), the processed delegation indicators is not rolled back.

Delegation indicator

Delegation indicators use the banned opcode 0xef, defined in EIP-3541, to indicate that the code must be handled differently than regular code. The delegation forces all code executing operations to follow the address pointer to obtain the code to execute. For example, CALL loads the code at address and executes it in the context of authority.

The affected executing operations are:

• CALL
• CALLCODE
• DELEGATECALL
• STATICCALL
• any transaction where destination points to an address with a delegation indicator present

For code reading, only CODESIZE and CODECOPY instructions are affected reading. They operate directly on the executing code instead of the delegation. For example, when executing a delegated account EXTCODESIZE returns 23 (the size of 0xef0100 || address) whereas CODESIZE returns the size of the code residing at address.

Note, this means during delegated execution CODESIZE and CODECOPY produce a different result compared to calling EXTCODESIZE and EXTCODECOPY on the authority.

Precompiles

When a precompile address is the target of a delegation, the retrieved code is considered empty and CALL, CALLCODE, STATICCALL, DELEGATECALL instructions targeting this account will execute empty code, and therefore succeed with no execution when given enough gas to initiate the call.

Loops

In case a delegation indicator points to another delegation, creating a potential chain or loop of delegations, clients must retrieve only the first code and then stop following the delegation chain.

Gas Costs

The intrinsic cost of the new transaction is inherited from EIP-2930, specifically 21000 + 16 * non-zero calldata bytes + 4 * zero calldata bytes + 1900 * access list storage key count + 2400 * access list address count. Additionally, add a cost of PER_EMPTY_ACCOUNT_COST * authorization list length.

The transaction sender will pay for all authorization tuples, regardless of validity or duplication.

If a code reading instruction accesses a cold account during the resolution of delegated code, add an additional EIP-2929 COLD_ACCOUNT_READ_COST cost of 2600 gas to the normal cost and add the account to accessed_addresses. Otherwise, assess a WARM_STORAGE_READ_COST cost of 100.

Transaction origination

Modify the restriction put in place by EIP-3607 to allow EOAs whose code is a valid delegation indicator, i.e. 0xef0100 || address, to originate transactions. Accounts with any other code values may not originate transactions.

Additionally, if a transaction’s destination has a delegation indicator, add the target of the delegation to accessed_addresses.

Rationale

Below is the rationale for both general design directions of the EIP, as well as specific technical choices.

General design philosphy

Persistence of code delegation

The first draft of this proposal had a clever idea to avoid disagreement on whether in-protocol revocation was needed or not. The idea was to temporarily set code in the account with the authorization. After the transaction finished, the code would be completely cleared. This was a new design space for enriching EOA functionality.

Even this approach was not without its flaws. Fundamentally, there was not much friction for users including set code authorizations. This meant that some users and applications would opt to treat the extension as more of a scripting facility, rather than a full-fledged upgrade to a smart contract wallet. The outcome of this would be two somewhat competing workstreams for UX improvements: smart contract wallets and EOA scripts.

Previous proposals had been met with similar criticisms. To counteract this, persistent delegations were introduced. They create enough friction in deployment that users will not deploy new, unique ones regularly. This will hopefully unify the workstreams and minimize fragmentation in UX developments.

No initcode

Running initcode is not desirable for many reasons. It creates a new mode of execution that needs extensive testing, and may be used for purposes not possible with standard smart contract wallets. It also forces developers to perform initialization as a standard call to the EOA after delegation. The lack of atomicity in these operations is another factor that will push users to complete smart contract wallet solutions, instead of EOA scripts.

Additionally, initcode tends to be propagated inside the transaction calldata. This means it would need to be included in the authorization tuple and signed over. The minimum initcode is around 15 bytes – it would simply copy the contract code from an external address. The total cost would be 16 * 15 = 240 calldata cost, plus the EIP-3860 cost of 2 * 15 = 30, plus the runtime costs of around 150. So nearly 500 additional gas would be spent preparing the account. Even more likely, 1200+ gas if not copying from an external account.

Creation by template

Initcode or not, there is a question of how users should specify the code they intend to run in their account. The two main options are to specify the bytecode directly in the transaction or to specify a pointer to the code. The simplest pointer would just be the address of code deployed on-chain.

The cost analysis makes the answer clear. The smallest proxy would be around 50 bytes and an address is 20 bytes. The 30 byte difference provides no useful additional functionality and will be inefficiently replicated billions of times.

Furthermore, specifying code directly would again make it possible for EOAs to have a new, unique ability to execute arbitrary code specified in the transaction calldata. It is for these reasons that creation by template is chosen.

Interaction with applications and wallets

While this EIP provides a lot of flexibility to applications and EOAs, there are incorrect ways of using it. Applications must not expect that they can suggest the user sign an authorization, and therefore it is the duty of the wallet to not provide an interface to do so.

There is no safe way to provide this interface. The code specified by an authorization has unrestricted access to the account and must always be closely audited by the wallet. Few users have the level of sophistication to reasonably verify the code they are delegating to.

It is also not possible to implement a system of permissions at this level to minimize the risk. If applications require custom wallet functionality, they must use standardized extension / module systems built on top of the delegated code that correctly implements permissions.

Forward-compatibility with future account abstraction

This EIP is designed to be forward-compatible with endgame account abstraction, without over-enshrining any fine-grained details of ERC-4337 or RIP-7560.

To start, the address that users sign could directly point to existing ERC-4337 wallet code. This essentially requires the “code pathways” that are used are code pathways that would, in most cases, continue to make sense in a pure-smart-contract-wallet world. Hence, it avoids the problem of creating two separate UX workstreams because, to a large extent, they would be the same ecosystem.

There will be some workflows that require kludges under this solution that would be better done in some different “more native” under “endgame AA”, but this is relatively a small subset. The EIP does not require adding any opcodes, that would become dangling and useless in a post-EOA world, and it allows EOAs to masquerade as contracts to be included in ERC-4337 bundles, in a way that’s compatible with the existing EntryPoint.

Self-sponsoring: allowing tx.origin to set code

Allowing tx.origin to set code and execute its own delegated code enables what is called self-sponsoring. It allows users to take advantage of EIP-7702 without relying on any third party infrastructure.

However, that means the EIP breaks the invariant that msg.sender == tx.origin only happens in the topmost execution frame of a transaction. This will affect smart contracts containing require(msg.sender == tx.origin) style checks. This check is used for at least three purposes:

1. Ensuring that msg.sender is an EOA (given that tx.origin always has to be an EOA). This invariant does not depend on the execution layer depth and, therefore, is not affected.
2. Protecting against atomic sandwich attacks like flash loans, which rely on the ability to modify state before and after the execution of the target contract as part of the same atomic transaction. This protection would be broken by this EIP. However, relying on tx.origin in this way is considered bad practice, and can already be circumvented by miners conditionally including transactions in a block.
3. Preventing reentrancy.

Examples of (1) and (2) can be found in contracts deployed on Ethereum mainnet, with (1) being more common (and unaffected by this proposal). On the other hand, use case (3) is more severely affected by this proposal, but the authors of this EIP did not find any examples of this form of reentrancy protection, though the search was non-exhaustive.

This distribution of occurrences—many (1), some (2), and no (3)—is exactly what the authors of this EIP expect because:

• Determining if msg.sender is an EOA without tx.origin is difficult, if not impossible.
• The only execution context which is safe from atomic sandwich attacks is the topmost context, and tx.origin == msg.sender is the only way to detect that context.
• In contrast, there are many direct and flexible ways of preventing reentrancy (e.g., using a transient storage variable). Since msg.sender == tx.origin is only true in the topmost context, it would make an obscure tool for preventing reentrancy, rather than other more common approaches.

There are other approaches to mitigate this restriction which do not break the invariant:

• Set tx.origin to a constant ENTRY_POINT address when using the CALL* instruction in the context of an EOA.
• Set tx.origin to a special address derived from the sender or signer addresses.
• Disallow tx.origin from setting code. This would make the simple batching use cases impossible, but could be relaxed in the future.

Rationale for technical details

Cost of delegation

The PER_AUTH_BASE_COST is the cost to process the authorization tuple and set the delegation destination. To compute a fair cost for this operation, the authors review its impact on the system:

• ferry 101 bytes of calldata = 101 * non-zero cost (16) = 1616
• recovering the authority address = 3000
• reading the nonce and code of authority = 2600
• storing values in already warm account = 200
• cost to deploy code = 200 * 23 = 4600

The impact-based assessment identifies 12016 gas of comparable computation for the operation. It is rounded up to 12500 to account for miscellaneous costs associated with shuttling data around the state transition.

Clearing delegation indicators

A general design goal in state transition changes is to minimize the number of special cases an EIP has. In early iterations, this EIP resisted a special case for clearing an account’s delegation indicator.

For most intents and purposes, an account delegated to 0x0 is indistinguishable from a true EOA. However, one particular unfortunate case is unavoidable. Even if a user has a zeroed out delegation indicator, most operations that interact with that account will incur an additional COLD_ACCOUNT_READ_COST upon the first touch caused by attempting to load the code at 0x0.

For this reason, the authors have opted to include a special case which allow users to restore their EOA to its original purity.

Lack of instruction prohibition

Consistency is a valuable property in the EVM, both from an implementation perspective and a user-understanding-perspective. Despite considering bans on several families of instructions in the context of EOAs, the authors feel there is not a compelling reason to do so, as it would cause smart contract wallets and EOA smart contract wallets to proceed down distinct UX workstreams.

The main instruction families where a ban was considered were storage related and contract creation related. The decision to not ban storage instructions hinged mostly on their importance to smart contract wallets. Although it’s possible to have an external storage contract that the smart contract wallet calls into, it is unnecessarily complicated and inefficient. In the future, new state schemes may allow substantially cheaper access to certain storage slots within an account. This is something smart contract wallets will want to take advantage of that a storage contract wouldn’t support.

Creation instructions were considered for a ban as well on other similar EIPs, however because this EIP allows EOAs to spend value intra-transaction, the concern with bumping the nonce intra-transaction and invalidating pending transactions is not significant.

Protection from mallebility cross-chain

One consideration when signing a code pointer is what code that address points to on another chain. While it is possible to create a deterministic deployment, i.e. via Nick’s method, verifying such a deployment may not always be desirable. In such situations, the chain ID can be set to reduce the scope of the authorization. When universal deployment is preferred, simply set chain ID to 0.

An alternative to adding chain ID could be to substitute in the actual code for the address in the signature. This seems to have the benefit of both minimizing the on-chain size of auth tuples, by continuing to serialize only the address, while retaining specificity of the actual code running in the account, by pulling in the code for the signature. One unfortunate issue of this format, though, is that it imposes a database lookup to determine the signer of each auth tuple. This imposition itself seems to create enough complexity in transaction propagation that it is decided to avoid and simply sign over the address directly.

Delegation of code execution only

Other code retrieving operations like EXTCODEHASH do not automatically follow delegations, they operate on the delegation indicator itself. If instead delegations were followed, an account would be able to temporarily masquerade as having a particular codehash, which would break contracts that rely on codehashes as a definition of possible account behavior. A change of behavior in a contract is currently only possible if its code explicitly allows it (in particular via DELEGATECALL), and a change of codehash is only possible in the presence of SELFDESTRUCT (which, as of Cancun, only applies in the same transaction as contract creation), so choosing to follow delegations in EXTCODE* opcodes would have created a new type of account breaking prior assumptions.

Charge maximum cost upfront

While computing the intrinsic gas cost, the transaction is charged the worst-case cost for each delegation. Later, while processing the authorization list, a refund is issued if the account already exists in state. This mechanism is designed to avoid state lookups for each authorization when computing the intrinsic gas and can quickly determine the validity of the transaction with only a state lookup on the sender’s account.

No blobs, no contract creation

Transactions should be thought of as specialized tools and not necessarily a one-type-does-all solution. EIP-4844 is treated differently at the p2p level due to burden blobs place on a node’s bandwidth. EIP-7702 has different implications on transaction gossiping and there is no need to complicate those rules unneccesarily by making it a superset of all possible functionality. The authors ultimately do not expect there to be much demand for atomic delegation and blob submission.

Contract creation is another specialized use case that has been grandfathered into several transaction types. It adds complexity to testing, because it is a new distinct branch of execution that needs to be tested when any change to the EVM occurs and verify the change works as expected in that context.

For these reasons, the authors have chosen to keep the scope of the EIP focused on improving UX.

Disallow delegation to precompiles

Precompiles are themselves edge cases, so allowing delegations to precompiles or not requires some focus in implementation. Considering the fact that precompiles technically do not have code associated with their accounts, the authors decided it would be marginally simpler to not execute the precompile logic when a user delegates to one. This is somewhat unintuitive.

Non-empty authorization list required

Set code transactions are required to have at least one authorization to be considered valid. This is to disincentivize senders from using type 4 transactions as a generic transaction format, because this transaction has different implications on the transaction pool than, say, EIP-1559 transactions.

Backwards Compatibility

This EIP breaks a few invariants:

• An account balance can only decrease as a result of a transaction originating from that account.
  • Once an account has been delegated, any call to the account may also cause the balance to decrease.
• An EOA nonce may not increase after transaction execution has begun.
  • Once an account has been delegated, the account may call a create operation during execution, causing the nonce to increase.
• tx.origin == msg.sender can only be true in the topmost frame of execution.
  • Once an account has been delegated, it can invoke multiple calls per transaction.

Security Considerations

Implementation of secure delegate contracts

The following is a non-exhaustive list of pitfalls that delegate contracts should be wary of and require a signature over from the account’s authority:

• Replay protection (e.g., a nonce) should be implemented by the delegate and signed over. Without it, a malicious actor can reuse a signature, repeating its effects.
• value – without it, a malicious sponsor could cause unexpected effects in the callee.
• gas – without it, a malicious sponsor could cause the callee to run out of gas and fail, griefing the sponsee.
• target / calldata – without them, a malicious actor may call arbitrary functions in arbitrary contracts.

A poorly implemented delegate can allow a malicious actor to take near complete control over a signer’s EOA.

Front running initialization

Smart contract wallet developers must consider the implications of setting code in an account without execution. Contracts are normally deployed by executing initcode to determine the exact code to be placed in the account. This gives developers the opportunity to initialize storage slots at the same time. The initial values of the account cannot be replaced by an observer, because they are either signed over by an EOA in the case of a creation transaction or they are committed to by computing the contract’s address deterministically from the hash of the initcode.

This EIP does not provide developers the opportunity to run initcode and set storage slots during delegation. To secure the account from an observer front-running the initialization of the delegation with an account they control, smart contract wallet developers must verify the initial calldata to the account for setup purposes be signed by the EOA’s key using ecrecover. This ensures the account can only be initialized with desirable values.

Storage management

Changing an account’s delegation is a security-critical operation that should not be done lightly, especially if the newly delegated code is not purposely designed and tested as an upgrade to the old one.

In particular, in order to ensure a safe migration of an account from one delegate contract to another, it’s important for these contracts to use storage in a way that avoids accidental collisions among them. For example, using ERC-7201 a contract may root its storage layout at a slot dependent on a unique identifier. To simplify this, smart contract languages may provide a way of re-rooting the entire storage layout of existing contract source code.

If all contracts previously delegated to by the account used the approach described above, a migration should not cause any issues. However, if there is any doubt, it is recommended to first clear all account storage, an operation that is not natively offered by the protocol but that a special-purpose delegate contract can be designed to implement.

Setting code as tx.origin

Allowing the sender of an EIP-7702 to also set code has the possibility to:

• Break atomic sandwich protections which rely on tx.origin;
• Break reentrancy guards of the style require(tx.origin == msg.sender).

The authors of this EIP believe the risks of allowing this are acceptable for the reasons outlined in the Rationale section.

Sponsored transaction relayers

It is possible for the authorized account to cause sponsored transaction relayers to spend gas without being reimbursed by either invalidating the authorization (i.e., increasing the account’s nonce) or by sweeping the relevant assets out of the account. Relayers should be designed with these cases in mind, possibly by requiring a bond to be deposited or by implementing a reputation system.

Transaction propagation

Allowing EOAs to behave as smart contracts via the delegation indicator poses some challenges for transaction propagation. Traditionally, EOAs have only be able to send value via a transaction. This invariant allows nodes to statically determine the validity of transactions for that account. In other words, a single transaction has only been able to invalidate transactions pending from the sender’s account.

With this EIP, it becomes possible to cause transactions from other accounts to become stale. This is due to the fact that once an EOA has delegated to code, that code can be called by anyone at any point in a transaction. It becomes impossible to know if the balance of the account has been sweeped in a static manner.

While there are a few mitigations for this, the authors recommend that clients do not accept more than one pending transaction for any EOA with a non-zero delegation indicator. This minimizes the number of transactions that can be invalidated by a single transaction.

An alternative would be to expand the EIP-7702 transaction with a list of accounts the caller wishes to “hydrate” during the transaction. Those accounts behave as the delegated code only for EIP-7702 transactions which include them in such a list, thus returning to clients the ability to statically analyze and reason about pending transactions.

A related issue is that an EOA’s nonce maybe incremented more than once per transaction. Because clients already need to be robust in a worse scenario (described above), it isn’t a major concern. However, clients should be aware this behavior is possible and design their transaction propagation accordingly.

Copyright

Copyright and related rights waived via CC0.

Citation

Please cite this document as:

Vitalik Buterin (@vbuterin), Sam Wilson (@SamWilsn), Ansgar Dietrichs (@adietrichs), lightclient (@lightclient), "EIP-7702: Set Code for EOAs [DRAFT]," Ethereum Improvement Proposals, no. 7702, May 2024. [Online serial]. Available: https://eips.ethereum.org/EIPS/eip-7702.