Upgrade Mesh to Go 1.14

[jalextowle]

This PR upgrades Mesh to Go version 1.14. All of the Go forks were updated, and their branches have been documented within the go.mod file.

Fixes #726.

[jalextowle]

Was there a particular reason that we used goreleaser at @albrow. I decided to use the script that was provided by golangci, but I'm happy to change this.

[albrow]

The golangci team probably just changed their recommendation. This looks fine to me.

[jalextowle]

@albrow Is there a reason not to check IsNull here?

[albrow]

We never set it to null at any point. But it doesn't really hurt to check anyways so I'm okay with this change.

[jalextowle]

@albrow This setting allows us to include Javascript files along with our Typescript. tsc generates the appropriate declaration and map files.

This allowed me to avoid converting wasm_exec.js to a Typescript file. Any reason why you think this is a bad idea?

[jalextowle]

It has been moved to @0x/mesh-browser-lite's tsconfig.json now, but this comment is still relevant.

[jalextowle]

Geth removed the LengthNonIndexed function entirely.

[jalextowle]

abigen now generates good names for all of the generated types, so it wasn't necessary to alias these names to TrimmedOrder anymore.

[albrow]

What's the reason for this change? It functionally changes how we handle UnsupportedEventError and looks like it introduces a bug.

[jalextowle]

This change was made due to a change in the way that Go's abi package unmarshals events. There is no longer a way to distinguish invalid events by looking at the difference in length between indexed and len(topics) - 1 from what I could tell by trying to fix the bug that was introduced by upgrading.

The error that is now emitted when an invalid ERC721 asset data is passed to unpackLog is something that begins with the error seen above.

FWIW, this passes all of the tests with the same behavior as the old if statement, but I'll grant you that the tests passing is not the same as a bug not existing.

[albrow]

Can you move the strings.Contains(erc721Err.Error(), "Unpack(no-values unmarshalled") to unpackLog instead of doing it here?

Basically change this:

if len(indexed) != len(log.Topics[1:]) {
	return UnsupportedEventError{Topics: log.Topics, ContractAddress: log.Address}
}

To:

if err := _abi.Unpack(decodedEvent, event, log.Data); err != nil {
	if strings.Contains(err.Error(), "Unpack(no-values unmarshalled") {
		return UnsupportedEventError{Topics: log.Topics, ContractAddress: log.Address}
	}
	return err
}

We should avoid error string comparisons altogether if possible, however, we can't avoid it here because of how the error is created in go-ethereum. In this case, we should try and keep the string comparison as close as possible to the original call site (i.e. in this case where we call ABI.Unpack).

This will also make the code more consistent with the previous behavior. As it is currently written there are two differences that could cause problems:

1. There are situations where we used to return UnsupportedEventError from unpackLog but we currently do not. Namely, if ABI.Unpack fails.
2. There are situations where we used to return UnsupportedEventError from decodeERC721 but we currently do not. Namely, if the fallback encoding fails we return erc721Err.

[albrow]

@jalextowle Before we merge I still want to double check the changes that you made to go-ethereum, go-ws-transport, and goleveldb. We also have an open discussion on Slack about using --composite together with --allowJs.

[jalextowle]

@albrow I needed to place this in a slightly different place than what you suggested. This is actually where the event seems to be emitted.

What's weird is that it seems like the event only has two topics, which suggests that there could be a bug in the way that Geth handles non-indexed topics.

[albrow]

@jalextowle can you update CONTRIBUTING.md with new instructions for Go 1.14? After that, LGTM.