[EIP-3074] Move to two-opcode version - #2
Conversation
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
Since this is your first pull request, we kindly remind you to check out EIP-1 for guidance.
| --- | ||
| eip: 3074 | ||
| title: Native Sponsored Transactions | ||
| title: AUTHORIZE and AUTHORIZEDCALL opcodes |
There was a problem hiding this comment.
| title: AUTHORIZE and AUTHORIZEDCALL opcodes | |
| title: AUTHORIZEDCALL opcode |
I prefer this, because it is the "meat" of the EIP. AUTHORIZE is really just a helper op.
There was a problem hiding this comment.
I think I prefer "AUTH and AUTHCALL opcodes", because those might grow to a family of opcodes in the future
| - **Callee** - the target of the call from `CALLFROM`. | ||
| | Variable | Type | Initial Value | | ||
| | ------------------- | --------- |:------------------------------------------------------------------------------------ | | ||
| | `authorizedAccount` | `address` | `0x0000000000000000000000000000000000000000` | |
There was a problem hiding this comment.
Does 0x0000000000000000000000000000000000000000 == 0000000000000000000000000000000000000000000000000000000000000000? Because we wouldn't want a change in address format to break things.
There was a problem hiding this comment.
I think one alternative would be to have it be an optional address (or address pointer), so that we don't have to use the zero address to indicate no authorized account
There was a problem hiding this comment.
Once pushed on the stack, addresses are extended to 32 bytes, right?
There was a problem hiding this comment.
@SamWilsn yes, I'm just trying to think if there is any weirdness that could occur.
| | ---------- | --------- | | ||
| | `top - 0` | `valid` | | ||
| | `top - 1` | `success` | | ||
| The arguments (`yParity`, `r`, `s`) are interpreted as an ECDSA signature of the form `secp256k1(keccak256(TYPE || abi.encode(invoker, commit)))`, where: |
There was a problem hiding this comment.
| The arguments (`yParity`, `r`, `s`) are interpreted as an ECDSA signature of the form `secp256k1(keccak256(TYPE || abi.encode(invoker, commit)))`, where: | |
| The arguments (`yParity`, `r`, `s`) are interpreted as an ECDSA signature on the secp256k1 curve over the message `keccak256(TYPE || rlp([invoker, commit]))`, where: |
There was a problem hiding this comment.
At minimum, I'd like to use rlp(...) instead of abi.encode(...).
There was a problem hiding this comment.
Changed it to specifying it by hand.
| - `commit` is one of the arguments passed into `AUTHORIZE` and is a 32-byte value that can be used to commit to specific additional validity condition in the invoker's pre-processing logic. | ||
|
|
||
| If the signature is valid, the `signerAddress` is recovered. If `signerAddress != tx.origin`, the context variable `authorizedAccount` is set to `signerAddress`. | ||
| In any other case, i.e. if the signature is invalid or `signerAddress == tx.origin`, `authorizedAccount` is reset to the zero address. |
There was a problem hiding this comment.
Would rather have a bullet point list that explains what an "invalid signature" is.
Co-authored-by: lightclient <14004106+lightclient@users.noreply.github.com>
| #### Gas Cost | ||
|
|
||
| `success` must be a one in all other cases. | ||
| The gas cost for `AUTH` is `3000`. This is the same cost as for the `ecrecover` precompile. |
There was a problem hiding this comment.
Does calling into the ecrecover precompile cost 3000 gas total, or is it 3000 plus the cost of a call?
If it's 3000+call, I think this needs to be a bit more expensive.
There was a problem hiding this comment.
Looked into this. Turns out the full cost of an ecrecover (via STATICCALL) after Berlin will just be this 3000 gas plus a single WARM_STORAGE_READ_COST (100) as per EIP-2929. AUTH has none of the static call overhead. Both ORIGIN and CALLER have a gas cost of 2, so I think any extra cost here should also be on the order of ~10 gas at most - at which point a clean 3000 might just be preferable...
There was a problem hiding this comment.
Ah okay, I did actually find one extra step that is somewhat expensive, namely the hashing for the signed message. That would natively have a gas cost of 48. So I feel like the realistic values for AUTH gas cost could be 3000 / 3050 / 3100.
Co-authored-by: Sam Wilson <57262657+SamWilsn@users.noreply.github.com>
3 participants
Updated the EIP to reflect the new AUTH, AUTHCALL iteration.