Test staking: Part 2

[aakoshh]

Continuation of #299

• apply update immediately until activated
• activate on collateral reached
• join and leave command
• implement join and leave on the model state
• implement unstake in the contract
• call the SUT for join
• call the SUT for stake
• call the SUT for leave
• call the SUT for unstake
• call the SUT for checkpoint
• validate balances
• validate stakes
• validate top N validators
• generate a non-empty cross message vector to check hashing

Fixes:

• On the contract side Integration testing fixes ipc-solidity-actors#255
• Use non-masked ID for gateway address
• Pass the checkpoint as a tuple to ABI hashing
• Add 27 to the signature recovery byte
• Various fixes on the staking test model

Running

The tests can be executed as follows:

cargo test --release -p contract-test --test smt_staking 

To run them on modified Solidity contracts, change the Cargo.toml file like so:

[patch."https://github.com/consensus-shipyard/ipc-solidity-actors.git"]
ipc_actors_abis = { path = "../ipc-solidity-actors/binding" }

And then run make ipc-actors-abi to recompile the ABI artifacts.

Notes

Ranking

I cannot replicate the exact ordering of validators when multiple validators have the same collateral, which can lead to disagreements between Rust and Solidity about who exactly are the top N validators, when validator N and N+1 both have the same collateral.

I used a stable sorting mechanism hoping that it will work fairly: given a sorted list of validators, we update the collateral, and the move them up or down to their new place, leaving others in place.

However, in Solidity the priority queues are unstable. For example:

• Say we we have the following values in a min-heap: [1a, 2b, 2c, 3d, 3e, 3f], where the letter is the validator identity and the number is their collateral
• Say we want to pop/delete the minimum value 1a
• The algorithm first swaps 1a with the last element 3f, resulting in [3f, 2b, 2c, 3d, 3e, 1a]. Then it removes the last element and makes the queue shorter. Finally it sinks the first element back into its place, resulting in [2b, 2c, 3f, 3d, 3e].
• So 3f moved from being the last to the top of the "threes".
• By contrast, in Rust we just remove 1a from the queue if it leaves, leaving the others in their relative order.

If this situation arises, I stop testing the system further and start with a fresh state. I don't think it's worth trying to replicate this exact ordering mechanism.

[adlrocha]

These tests are actual gold, @aakoshh. Thanks! This makes testing significantly easier than what we were doing pre-Fendermint ❤️

[adlrocha]

I like this from_fvm and from_eth conversion utils. I feel we should expose them as a crate and make them available to the ipc repo. I feel we have a lot of redundant conversion functions here and there :)

//cc @cryptoAtwill, thoughts?

[aakoshh]

They are exposed in the messages crate, but some low level ones might be useful on their own for someone who doesn't want to deal with Fendermint message types.