Support submessages in multitest

[ethanfrey]

Closes #259

• Add optional reply to contract wrapper contructor
• Add sample contract that handles submessages
• Add (failing) submessage tests
• Support submessages

[ethanfrey]

I think I finally finished this.

Sometimes it felt like brute forcing to get the levels of caching working and I am sure it could use cleanup.

Happy for a review, even if only stylistic changes.

[ethanfrey]

Okay, with #328 I fixed how we handle attributes to make it compatible with the events as returned with the SDK. And simplified/corrected how we handle SubMsg/reply.

I feel better about the functionality, but still think I need to organize the code better... especially all those traits I added to unify *Router and *Cache

[yun-yeo]

For the better reply test, maybe it is possible to wrap the data with Protobuf wrappers depends on the message type (bank & wasm::execute, wasm::instantiate)

message MsgInstantiateContractResponse {
    // Address is the bech32 address of the new contract instance.
    string address = 1;
    // Data contains base64-encoded bytes to returned from the contract
    bytes data = 2;
}

https://github.com/terraswap/terraswap/blob/55c9ac2e54e95462c5c2ed3507cbdc27993b1b64/contracts/terraswap_factory/src/response.proto#L4-L9

https://github.com/CosmWasm/wasmd/blob/8bec411e4bd947cc7d872bd14bd1da865bf703bb/proto/cosmwasm/wasm/v1beta1/tx.proto#L69-L74

Then, we can properly test Protobuf decoding part at reply like

let res: MsgInstantiateContractResponse =
     Message::parse_from_bytes(msg.result.unwrap().data.unwrap().as_slice()).map_err(|_| {
         StdError::parse_err("MsgInstantiateContractResponse", "failed to parse data")
     })?;

https://github.com/terraswap/terraswap/blob/55c9ac2e54e95462c5c2ed3507cbdc27993b1b64/contracts/terraswap_factory/src/contract.rs#L152-L155

[ethanfrey]

Nice catch. I did something in init_response, but that was the old (pre-protobuf) version. Thanks for pointing this out.

[ethanfrey]

Added #330 for this

[ethanfrey]

And closing with #331

[yun-yeo]

LGTM 👏 👏 👏

I think Protobuf style data decoding testing at reply can be done later.

Thanks for making good test-kit for all CosmWasm contracts.

[maurolacy]

Partial review of wasm.rs. lgtm.

Will continue reviewing later, as I have some errands to do now.

[maurolacy]

I started reviewing this file, and this is a little above my head at this point, but:

I guess we need a multi-level (contract state) cache, to separate "server" contracts state from "client" contracts states. This, to mimic the transactional nature of the different contracts in the SDK.

Impressive piece of work, by the way.

[ethanfrey]

Before this PR we had a one-level cache in App.execute_multi. That is, all the messages are run together in a cache, and any error will cause the state changes to rollback. They will only be committed to the "real" storage if all return a success.

With submessages, each submessage will need it's own cache. They can write state and then fail, and then reply could catch the error and turn it to a success. That would mean the "parent contract" writes should be committed but the submessage writes not. These semanatics are described here: https://github.com/CosmWasm/cosmwasm/blob/main/SEMANTICS.md#submessages

This leads to the need of an arbitrarily multi-level caching technique. I feel my approach in this PR is a bit brute force and could probably use a bigger refactor to only use one Storage (rather than multiple HashMaps), and unify a lot on the caching in the end, but that would be much more work than needed to get a 0.7.0 release out. This package really should get some more refactors when we have capacity.