New NodeToClientVersionOf typeclass

[newhoggy]

This allows code to check what version of the node to client protocol is needed to use the query. For example:

ntcVersionOf $ QueryInEra qeInMode $ QueryInShelleyBasedEra qSbe QueryProtocolParameters

Whilst this function can be called directly, it is intended to be called by queryExpr eventually so that only queries that are valid for the current protocol are ever sent.

[Jimbo4350]

Nice, a few comments

[Jimbo4350]

Why are we defaulting to NodeToClientV_9 for any era? Shouldn't we be explicit with each era and assign the version as follows:

--
data NodeToClientVersion
    = NodeToClientV_9
    -- ^ enabled @CardanoNodeToClientVersion7@, i.e., Alonzo
    | NodeToClientV_10
    -- ^ added 'GetChainBlockNo' and 'GetChainPoint' queries
    | NodeToClientV_11
    -- ^ added 'GetRewardInfoPools` Block query
    | NodeToClientV_12
    -- ^ added 'LocalTxMonitor' mini-protocol
    | NodeToClientV_13
    -- ^ enabled @CardanoNodeToClientVersion9@, i.e., Babbage
    | NodeToClientV_14
    -- ^ added @GetPoolDistr, @GetPoolState, @GetSnapshots

[Jimbo4350]

I.e NodeToClientV_12 for everything up to Alonzo and NodeToClientV_14 for Babbage? The downside is we have to manually maintain this pairing.

[newhoggy]

Why are we defaulting to NodeToClientV_9 for any era?

NodeToClientV_9 is the oldest era that is still being tracked. Any queries that exist but are not in the above list above are presume to be from an era older than NodeToClientV_9, but we set it to NodeToClientV_9 because it is long enough ago that we don't care about the precise version anymore.

Note that our API doesn't export everything. For example GetRewardInfoPools is never used in any of our code.

The downside is we have to manually maintain this pairing.

Yeah, I believe some form of this might belong in consensus. It's implemented here for the moment because getting changes in consensus can take a long time.

Even if it is implemented in consensus, some form of this will exist in our code because the type of the argument to ntcVersionOf is a type that exists in our code base.

[Jimbo4350]

NodeToClientV_9 is the oldest era that is still being tracked

Sure but the NodeToClientV_* depends on the era, so I don't think we should always default to NodeToClientV_9. E.g as things are implemented now if we call ntcVersionOf on AnyCardanoEra BabbageEra there are more queries available in Babbage than NodeToClientV_9 will allow.

[newhoggy]

That's an interesting point. Would it be sufficient for nodeToClientVersionOf to take an era argument?

[newhoggy]

Note that we won't know the node's era unless we query for it, so it's not free.

[Jimbo4350]

I think manually doing this for now is ok.

[Jimbo4350]

#4787 (comment)

[Jimbo4350]

Still a couple of import changes needed

[Jimbo4350]

Sorry, instance NodeToClientVersionOf AnyCardanoEra where needs to be updated to reflect the latest versions for the relevant era

Removed unused typeclass instance

[newhoggy]

I think there is not a strong relationship between the node to client version and the era. For example a query could be added during an era, but depending on the circumstances, the added query might work for an earlier era.

[Jimbo4350]

LGTM

[Jimbo4350]

Random diff