ICS23: Generic Merkle types

[mossid]

The commitment proofs are defined as opaque objects dependent on the concrete types. For Merkle commitment proofs, the implementation is currently using tendermint/crypto/merkle, but it should use ICS23 implementation instead.

[mossid]

https://github.com/confio/ics23-tendermint provides an easy conversion method, can/should be used

To reduce onchain overhead all of the conversion should happen offchain side; the relayer retrieves crypto/merkle format from the chain(simply using CLIContext), convert it into ICS23 format, submit it into the other chain.

[fedekunze]

@mossid are you saying that we should remove the 23-commitment package in favor of https://github.com/confio/ics23/tree/master/go instead?

[AdityaSripal]

Should the confio package just be copied over wholesale? Or should I wrap the confio package with the interfaces we currently have in 23-commitment such that 23 no longer relies on crypto/merkle.Proof.

I also notice there are changes in the confio implementation that I'm not sure if it can be accepted as is, or have to be wrapped to conform to the previous interfaces.

1. confio uses a ProofSpec in the VerifyMembership functions, does this break expected interface in a way that needs to be hidden. For example, the proof Spec can be part of the struct implementing proof interface to maintain old interface.

2. There is no ApplyPrefix function with the ExistenceProof as far as i can tell, is this no longer necessary with the proofs in confio? Should i implement an ApplyPrefix function as a no-op or remove it entirely?

@fedekunze

[cwgoes]

ProofSpec will become part of either the client type or the connection version in IBC.

(both are possible, the latter is necessary only if the ProofSpec would vary across connections to the same chain using the same client - I believe ApplyPrefix is subsumed by ProofSpec, which can contain prefix information, so if we want that flexibility ProofSpec should go in the connection version)

confio/ics23 implements the ICS 23 interface pretty directly, I think I would recommend trying to fuse it and the current 23-commitment interface - we don't want to rely on crypto/merkle.Proof at all (in the 1.0.0 production IBC version).

cc @mossid - Do you agree with the above?

[fedekunze]

confio/ics23 implements the ICS 23 interface pretty directly, I think I would recommend trying to fuse it and the current 23-commitment interface - we don't want to rely on crypto/merkle.Proof at all (in the 1.0.0 production IBC version).

++

@ethanfrey thoughts on 1. and 2. above?

[cwgoes]

Blocked on ICS 8.

[cwgoes]

This is still blocked on spec work & cosmos/ics23#28.

[cwgoes]

Done.