Update EIP-7702: refine based on discussions

[lightclient]

Based on some discussions I've made the following updates:

• deploy code via template (copy from existing address) instead of specifying code directly
• reduced the per auth base cost to 2500, mostly to account for reading the template account
• sign over the template address instead of code

Additionally, I have added a lot of rationale for why decisions were made.

[eth-bot]

✅ All reviewers have approved.

[derekchiang]

I just want to clarify that, given the discussions on eth magicians, it's clear that many (most) of the updates in this PR are still in contention (even among the EIP authors), so I personally don't think that this is a good time to merge this PR, considering that we will have another ACDE meeting coming up to discuss 7702 and hopefully align on some of these potential updates.

Merging this PR now might mislead the core devs and other community members into believing that decisions have already been made one way or the other, when in reality the discussion is still very much ongoing.

[drortirosh]

Missing from the list of broken assumptions is signature malleability:
Specifically, a user might sign an unverified experimental 7702 contract on a testnet under the assumption (which was correct until now) that it is a sandbox, and nothing affects other network.
This becomes disastrous, as this contract can be used anywhere - and currently is un-revocable.
The user never opt-in to this new feature: he merely continued his work with an existing (though updated) wallet.

[smartprogrammer93]

@drortirosh chainId variable will be added to the signature. Also, this type of signed data has a specific magic. Wallets are expected to warn the user heavily about what they are about to sign.

[lightclient]

Signing an unverified / experimental 7702 contract would never happen with a wallet that correctly implements 7702.

[drortirosh]

I don't have any problem with "wallet that correctly implements the standard".

Leaving the chainid out of signature EIP breaks the invariant that "testnet are sandbox and can't affect any other network".
Please add that text into your security consideration, if you think that's acceptable.

[drortirosh]

Signing an unverified / experimental 7702 contract would never happen with a wallet that correctly implements 7702.

The only way to implement this is by completely blocking (by the wallet) from ever signing anything un-approved by the wallet, not even with a warning message.
That's the only way you can block this concern.

Even with a warning message such as:

Beware of this signature: it opens your account to malicious operations not only on this network, but on any other network, and this operation is un-revocable. are you sure you want to continue

And I'm afraid that on testnets, some people will STILL sign this message, as the current narrative is "testnets are sandbox"

[lightclient]

Which networks are banning chain id 0?

[drortirosh]

Which networks are banning chain id 0?

The question is, on which network you can submit a non-EIP-155 transaction? All block builders on all networks I know of block them.
That's a de-facto standard (that all TXs are EIP-155).

[lightclient]

You're conflating protocol functionality with protections offered at the RPC level.

[drortirosh]

I'm saying that there is an implicit assumptions today by users, that this EIP breaks:

1. That signing data on testnet stays in testnet and doesn't affect other networks.
2. That signing anything has a one-shot effect, not long-lasting (or permanent)

And I think that breaking these assumptions AT LEAST deserves being mentioned in the "security consideration" (or backward compatibility) section.
After specifying them, you can add your argument why you think it is OK to break them - but at least acknowledge that we're introducing a network change that alters existing users assumptions.
This is not a change that affect new users that use those new features, or new wallets: it affect all existing users that happen to upgrade their existing wallet, and are unaware of such changes.

[lightclient]

That signing data on testnet stays in testnet and doesn't affect other networks.

Again, this is not correct. The protocol allows for non EIP-155 txs to be replayed on any chain. If users think they're safe because the wallet says they're on a testnet today, it's because apps, wallets, and RPC providers have done well to create this feeling of security. They could have just as easily continued signing and allowing txs w/o chain id baked in. I have no reason to believe post-7702 things will be any different.

That signing anything has a one-shot effect, not long-lasting (or permanent)

This would be true for 7702 auths. Happy to put reword the security considerations if it is not clear there.

[github-actions]

The commit 23f2ec0 (as a parent of 133edf9) contains errors.
Please inspect the Run Summary for details.

[Jrachman]

Some thoughts I'd like to bring up as I review this revision:

In the case of multiple tuples for the same authority, set the code specified by the address in the first occurrence. (ref)

So this means that if we have tupleA and tupleB where tupleA comes first in the list and both ecrecover calls come out to be authority, then the code of authority would be set to tupleA and tupleB would be ignored?

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. (ref)

Isn't this a good thing? I believe the context this is in makes it sound like a bad thing. I bias towards maximal flexibility. This "creation by template" doesn't necessarily mean that what the EOA can sign/execute is constrained right? It is just choosing the way that code should be signed over?

The main families of instructions where a ban was considered were storage related and contract creation related. (ref)

Potentially dumb question but what happens if an SSTORE opcode is called within a 7702 transaction? Would it just not do anything or act kind of like memory storage because EOA's don't have built-in storage? Also, what are the unnecessary inefficiencies with an external storage contract? - would love to identify them to keep note of them.

A hotly debated element of this EIP is the need for in-protocol revocation. (ref)

I wrote some thoughts here around time-based revocation. Not sure if it is at all useful here but thought I'd throw it into the ring.

The following is a non-exhaustive list of checks/pitfalls/conditions that delegate contracts should be wary of and require a signature over from the account's authority: (ref)

Would this list of params be signed over in the data field of the 7702 transaction? Thinking about this like it's what you sign over in execData for 3074...

[frangio]

If I produce and broadcast one 7702 signature without nonce, I can never again use 7702 without the risk of being frontrun with my previous signature to prevent the new code from being deployed on my account. This does seem like something the protocol should give tools to prevent.

[smartprogrammer93]

If I produce and broadcast one 7702 signature without nonce, I can never again use 7702 without the risk of being frontrun with my previous signature to prevent the new code from being deployed on my account. This does seem like something the protocol should give tools to prevent.

@frangio I am not sure how would be front run. If you send a 7702 tx that you control what authorizations go inside it. Even if there was a 7702 tx before, it wont affect how your tx is executed.

[frangio]

Right, I was thinking of a bundle submitted by a sponsor, and "frontrunning" in the sense of including an authorization before another in a 7702 tx authorization list.

I guess that clarifies why it shouldn't be considered a problem? A sponsor can always just choose not to include your signed authorization, so including two of your authorizations as a way to censor one of them is not really a new capability.

[smartprogrammer93]

@lightclient can we specify exactly the behavior in case the validations of an authorization fails? my assumption is that it gets ignored but the tx goes through regardless. if that is not the case and tx should fail, do we spend all gas? seems excessive.