We have recently discussed the possibility of migrating the accessWitness from the StateManager to the EVM. It makes conceptual sense for the EVM to be in charge of this low-level tracking rather than the StateManager.
The accessWitness file itself is fairly minimal, only importing from the common and util packages, so the migration of the file itself is straightforward.
However there are some additional considerations:
accessWitness is deeply tied to verkle logic, notably for pricing execution witnesses (ref: https://eips.ethereum.org/EIPS/eip-4762), and we should only enable when verkle (EIP-6800) is also enabled. I think we want this to be hardfork/EIP-inferred rather than manually specified so that we are able to dynamically transition in the context of the merkle->verkle transition.
verkle state managers (both stateless and stateful) will have to be adjusted so that they obtain the acccess witness from the EVM somehow for things like postStateVerify. This shouldn't be too tricky, as I think postStateVerify is called by the VM, which could also pass in the accessWitness as an arg.
We'll need to create an AccessWitness interface or type for use for e.g. in the StateManager (we wouldn't want to add an EVM dependency just for importing the AccessWitness types, enums, etc)
acolytec3
commented
6 days ago
We have recently discussed the possibility of migrating the accessWitness from the StateManager to the EVM. It makes conceptual sense for the EVM to be in charge of this low-level tracking rather than the StateManager.
The accessWitness file itself is fairly minimal, only importing from the
commonandutilpackages, so the migration of the file itself is straightforward.However there are some additional considerations: