[ETCM-846] make ETS stBadOpcode tests pass

[AurelienRichez]

Description

This PR is part of the ogoing effort to make ETS pass all tests. The stBadOpcode suite did not pass because the accounts were not imported with the root hash for their storage.

Proposed Solution

The getGenesisStateRoot now create an in memory trie in order to compte the resulting hash.

As this change is not specific to stBadOpcode, it also makes most the previously failing tests pass (9 failing test for GeneralStateTests while we had around 400 before).

[dzajkowski]

please also use the ticket tag in commit messages ([ECTM-846]) this enables integration with jira

[jvdp]

So it looks a lot simpler than the previous implementation and less dependent on InMemoryWorldStateProxy which sounds scary. (What's the impact of not using this anymore btw? Is it only for tests & test mode or also part of production code?)

Can you explain the main change a bit more? It's now more directly committing the genesis account to the blockchain instead of keeping it in the in-memory state?

[AurelienRichez]

@jvdp

The change is only for testmode so it should not affect normal production. We should probably at some point refactor the code so that the test specific parts are in a dedicated module though (there is a ticket for that)

The main problem here was that the rootHash for account storage was not set properly in getGenesisStateRoot.

Regarding the ditching of InMemoryWorldStateProxy :

From what I see InMemoryWorldStateProxy is more focused on building a transient state when running transactions, i.e, taking an initial state, modifying it and then commit the new resulting state.

But when we load the genesis state we do not have an initial state to start with so we have a kind of a chicken and egg problem (no initial to set the storage, and the initial state needs a storage).

My first working solution was to create the genesis state with empty contract storage, call the function with InMemoryWorldStateProxy, get the root hashes from there and then reload the genesis this time with the real contract storage.

I found it weird to have this logic (there was no reason to initialize the state in two passes), and with some guidance from Konrad I found how to directly store arbitrary MPTs. Calling directly the storage layer allows to create the desired state in the database directly without the double genesis initialization trick.