Actually if enabled /block_results will return an error saying that the node isn't persisting abci responses

I wonder if adding the RPC part is confusing. Yes it's in the RPC section but that's because we didn't want to create a new section and wasn't sure where such a change would best fit so we landed it in RPC but really this is a storage engine config

Actually if enabled /block_results will return an error saying that the node isn't persisting abci responses

Ah, I missed this somehow. In terms of the way it's implemented now though, does this make sense given that we do actually store the last ABCI response and can actually give the latest block results? Should we really always be returning an error if we have the results? (Mainly asking for context around the decision-making there)

I wonder if adding the RPC part is confusing. Yes it's in the RPC section but that's because we didn't want to create a new section

Is it really a big deal to create a new section, especially when the flag's off by default? I'd say that, since it's technically a state store option, so it could either go into a [statestore] section, or more generically a [storage] section.

Ah, I missed this somehow. In terms of the way it's implemented now though, does this make sense given that we do actually store the last ABCI response and can actually give the latest block results? Should we really always be returning an error if we have the results? (Mainly asking for context around the decision-making there)

We definitely could. There needs to be some added logic but it's possible. The design decision was one of simplicity - easier just to close of the entire RPC endpoint than to have it error for every height except the latest

Is it really a big deal to create a new section, especially when the flag's off by default? I'd say that, since it's technically a state store option, so it could either go into a [statestore] section, or more generically a [storage] section.

It's possible to maybe have a general storage section. It's not clear to me yet what other configuration options there might be

We definitely could. There needs to be some added logic but it's possible.

Speaking to the IBC-rs team now, it may actually be better to just keep it as-is. The relayer could then just ping the /block_results endpoint on startup and, if it errors out with the current error, they can notify the user that the relevant node isn't correctly configured.

omit "be able to"