Drop support for JSON encoded transactions

[ghost]

Why

• As new Eras come into play and Tx representation changes, storing data in non-canonical JSON runs the risk of inconsistencies as we observed when trying to migrate to Conway and found out the txid changed because canonical CBOR representation changed
• We maintain our own JSON representation which is cumbersome and error-prone as we don't test for all the possible transaction contents and are not safe against upstream changes

What

• Migrate how we store and share Transactions as JSON everywhere

  • All Tx JSON representations have the transaction data as cborHex (see ADR)
  • NewTx websocket command and /cardano-transaction HTTP endpoint accept transactions as objects with a cborHex field (nothing else required)
    • If an txId is present, it must match the transaction body hash
  • The /commit response and TxValid, TxInvalid etc. websocket messages are compatible with cardano-cli transaction commands and also contain a txId
  • Anything we store on disk must us CBOR encoded transactions (can use the same schema as above)
  • Documentation is updated to instruct users to use CBOR encoded transactions and how to deal with them (e.g. debugging)

• Ensure our maintained hydra clients and integrations still work

  • hydra-tui and hydraw (in the same repo)
  • hydra-poll and hydra-chess (maintained by us, different repos)
  • kupo indexing of a hydra head (we originally contributed and is affected)

• Out of scope: Any changes to UTxO (as initially drafted in the ADR)

How

• ADR to use cbor for tx #1215
• Replace ToJSON/FromJSON Tx with cbor encoding #1240
• Ensure consistency of published schemas through tests
• Maybe: Version our persisted data
• Fix hydra-poll
• Fix hydra-chess
• Fix kupo
• Fix hydraw

[ghost]

After having spent some time working on a Chess game DApp for Hydra, I am now convinced of the benefits of a proper JSON API over a CBOR one, at least for UTxO-related interactions. Being able to commit a JSON-formatted UTxO through the HTTP interface, and being able to extract information from JSON-formatted UTxO from the Head protocol, is very useful as it alleviates the need for the client to be aware of all the details of the CBOR formatting of those UTxOs.

In general, a JSON API is much more versatile from a client perspective, notably because it allows those clients to pick and choose what they support, for example ignoring some parts of a data structure they don't care about and using symbolic references (eg. field names) instead of having to be aware of all the details of the structure's layout as in CBOR. There's support for JSON-schema based documentation and even code generation, which might not be stellar but is still better than what's available for CBOR. And it's always easier for a client to drop to raw JSON handling.

I think we should continue to push for first-class support of such an interface in the cardano ecosystem, and for example improve on Cardanonical, and/or propose/use canonical JSON schema for cardano-ledger data structures.

What I think is the most glaring issue right now is the lack of proper versioning of the serialised data, whether it's persisted on disk or exchanged across the network of nodes or through the interfaces we provide:

• A starting hydra-node should be able to check what's the version of the data it's loading
• A client should know what's the version of the data is supported by the node it connects to

This can be easily done by extending these data structures with a generic version field, and adding check/conversion code to the hydra-node. Note this realisation also dawned on the cardano-ledger team leading them to change all their serialisation formats to include a version.

[ch1bo]

Discussion from tactical:

• Any "self-hashing" data should not be accepted as JSON, although for informational purposes it might be just fine and helpful to output this data as JSON on the API
• UTxO for example are helpful and fine to accept as JSON on /commit endpoint

[ch1bo]

I realized that we also should contribute a PR to kupo to decode the cbor (as a follow-up)

[ch1bo]

This should actually become a 💬 feature item and put on our roadmap. I'll bring it up on our next grooming session.

[ch1bo]

We decided not to include hydra-poll and hydra-chess fixes as they are targeting older versions of the released hydra-node right now and we will assist anyone who wants to put it o use when needed.