Tests for EIP-3860: Limit and meter initcode

[gumb0]

https://eips.ethereum.org/EIPS/eip-3860
Geth implementation in ethereum/go-ethereum#23847

[jochem-brouwer]

Are these tests up to date? The EIP seems to state that once we violate the max initcode on a CREATE or CREATE2, we put 0 on the stack but do not consume all gas (not sure if this is correct?). However, in these tests, that seems to be the case..?

[axic]

I don't think these have been updated yet.

[jochem-brouwer]

@axic sorry to bother you on this one again, but I think that the EIP is now unclear the gas costs regarding this one. If I try to CREATE or CREATE2 a new contract and the length parameter is larger than the max init code size, then what gas cost should be charged? I think this logically leads to multiple options:

1. The base fee of the CREATE opcode. We can pop the values from stack, if the length parameter > max init code size then charge the gas and push 0 on the stack
2. Calculate the entire gas cost to start executing the message. So this would involve the initcode jump analysis cost (introduced by the EIP), memory expansion, access costs and hash costs (CREATE2). So this is the upfront gas cost before starting to execute the initcode of the CREATE message.
3. Consume the provided gasLimit of the message

I am not sure if this is practically justified, but for gas costs, (1) seems to be the most logical one to me. This charges the CREATE base cost, but we just pop the codesize length item from stack, check if it is too large, and then just continue. (2) seems overkill since we do not execute a message at all (so no memory expansion, access, hashing or jumpdest analysis is needed) and (3) seems extremely overkill with the same arguments as (2).

Regarding the EIP: for me, it is hinting at (2) or possibly (1), but this is not very clear. So the question is: what gas costs should be charged when the initcode is larger than the max init code?

[gumb0]

@jochem-brouwer In the tests currently if initcode is larger then limit, 0 is pushed on stack and all gas given to CREATE/CREATE2 (63/64th of caller's gas) is consumed.

We're reconsidering the gas cost of failure and will clarify in EIP.

[gumb0]

I've updated now the tests with the rule that all gas is not consumed when initcode is larger than limit, this is the probable direction EIP spec will go.

The cost of CREATE in this case is base static cost + memory expansion cost + initcode charge.
(+ hashing charge for CREATE2)
It is similar to the failures due to call stack limit and not enough balance to transfer.

[jochem-brouwer]

Great, this makes sense, thanks @gumb0! :)

[winsvega]

are this two samples identical?
you can have 2 different expect sections of the same data but for different forks, saving the space

[gumb0]

As comments say, they are different in size. (invalid is one byte longer than valid)

[winsvega]

its better not to check the balance as the gasPrice change a lot by other eips
you can measure gas before the code you want to check and after the execution and sstore the difference.
msotre(1, gas())
... code ...
msotre(10, gas())
sstore(1, sub(mload(1), mload(10)))

[gumb0]

I reworked it using yul variables and gas()

[winsvega]

empty code both for valid and invalid creations ?

[gumb0]

(it's not empty, but one byte 00 of code deployed) and yes before EIP result is the same for valid and invalid.

With EIP account is not created for invalid initcode.

[winsvega]

I mean isn't it better to deploy some code so not to mistake the created contract with smth else.

[gumb0]

I made it deploy the codecopied init code itself (0x600a80600080396000f3)

[jochem-brouwer]

It does not seem that it is being tested if CREATE/CREATE2 charges the initcode per words gas (of 2 gas per 32-byte word).

What I would also suggest is to add a test which CREATE/CREATE2 a contract, and inside this contract the GASLIMIT is stored (this verifies that the initcode cost is deducted before forwarding the 63/64 of the remaining gas)

[winsvega]

better to have it in state tests too
especially we can check the tx exception

[winsvega]

need to create stEIP3860 folder and generate from those.
Merge+3860 forkname

[winsvega]

make a folder stEIP3860
change the forkname Merge+3860
check invalid transaction exception in state tests.

[gumb0]

Requires addition to retesteth config:

      "TR_InitCodeLimitExceeded" : "max initcode size exceeded:",

[winsvega]

ready?

[gumb0]

Ready, I've noted what would be nice to add later in #1101