perf(evm): only store changed storage entries

[mattsse]

Blocked by #3183

Motivation

only clone contract storage for step tracer
the storage entry in the geth StructLog is only the state of the contract, Ref https://docs.dcfsscan.io/docs/dapp/tracing

Therefor it's not necessary to clone the entire state. instead clone only the storage of the contract.

@shekhirin I think this is a significant factor for the recent mem issues

Solution

[shekhirin]

yes, I think you're right. This change will basically act as a diff on each step for storage which I was talking about, and that's also the way it's implemented in geth. Thank you!

I have two things to consider but I may be incorrect:

1. Since we're making a state snapshot in the step(), we've not yet executed the SLOAD/SSTORE opcodes, therefore there's no storage change made. I see geth being a bit different in that behaviour: they capture the change instantly https://github.com/ethereum/go-ethereum/blob/3a4cef5402d8b58831f95a909f8a1bdf3acffe3c/eth/tracers/logger/logger.go#L190-L205.
2. We can also do the same thing with checking opcodes for SLOAD/SSTORE as geth does, so we wouldn't loop through the storage without a need.

[mattsse]

I see, hmm how do we get the value during step?

not sure I can follow. most of the work is in #3183, I merely changed the state field's type and added the filter in this PR.

Would you like to add a fix for your 2nd point and I can just close this?

[shekhirin]

hmm how do we get the value during step?

I think we need to fill the storage in step_end(), so we could capture changed slots.

I merely changed the state field's type and added the filter in this PR.

Sorry, I'm talking about further optimization which is checking for SLOAD/SSTORE opcodes in step_end() and only in these cases iterating through the storage. I think what you've done is fine and will work!

[shekhirin]

something like that 5281088

[mattsse]

something like that 5281088

that looks great, let's merge this instead!