Update geth-utils and bus-mapping to build more witness

[han0110]

This PR is once part of #278, and aims to fulfill all required witness when parsing.

It first update geth-utils to better match how geth trace API works, and parse all fields of call for further usage.

The most tricky field of call is rw_counter_end_of_reversion, which takes 2 rounds of iteration on the traces to infer. Briefly speaking, the first round is for inspecting if a call ends up success (e.g. STOP or RETURN) or failure (REVERT or all kind of Exceptions), the second round is for recording the correct amount of read/write, with proper reversions if needed, then knowing the correct rw_counter_end_of_reversion. For more details on state write reversion, please refer to this note with illustration.

Also due to the interpreter changes of geth, error parsing of InvalidOpcode and WriteProtection are handled by get_step_err instead of directly from error string (see ethereum/go-ethereum#24031 and ethereum/go-ethereum#23970)

Resolves #174
Resolves #188
Resolves #225
Resolves #250
Resolves #289

[ed255]

🎉
This resolves some of the tasks defined here #165

[ed255]

This PR also resolves #174

[ed255]

I've left some comments and notes. I plan to review the circuit input builder changes in more detail after this first review.

[ed255]

How/Where will this method be used?

[han0110]

Ah, this is used here in #278 and I forgot to remove it. Should I remove it?

[ed255]

Thanks! No need to remove it, I just wanted to understand where it was needed because I was thinking of a possible edge case but it may not apply here. The edge case I was thinking about is that an Account has a Storage which we model as a HashMap. But keys can be deleted in EVM by setting the value to 0, but our HashMap will not be empty in that case. Maybe this edge case never occurs, because to write to the Storage you would need a code hash != empty.

Nevertheless let me confirm this: is_empty should return true when either the account never existed, or it has been destroyed, and no other case right?

[han0110]

I see, to put something in storage of an account, its nonce must be non-zero (being created or already created), so I guess this edge case would not happen (actually contract could be have some storage without code, we can do this by store something and return nothing when creation).

is_empty should return true when either the account never existed, or it has been destroyed, and no other case right?

Not really. You just reminds me, for destroyed one, it depends on if we have finalized or not (tx ends or not), because before finalizing, destroyed account is not empty. We might need to add a tag to record if this account is destroyed or not, and finalize to actual remove it after each tx ends.

[ed255]

Overall the PR looks good to me! Quite a lot of changes!
I really like the introduction of ReversionGroup. I think it makes the code more readable and easier to follow what's going on. I want to add that I think diagrams are extremely useful to understand the concepts and details of how reversion works, and the possible cases that can happen. @han0110 you've already made some diagrams and documentation about this in hackmd; I think we should move them to the zkevm-specs repository so that they are easily accessible.

[han0110]

@ed255 Thanks a lot for reviewing and suggestion of docs! All suggested docs are applied in 20ed665.