High level summary
The ledger team finished up the remaining work for tracking individual depots, built out the new Conway era transaction body (in line with CIP-1694), greatly reduce some problematically large calculations on the epoch boundary, and addressed technical debt.
Lower level summary
Finishing the deposit tracking
The initial work on the individual deposit tracking project focused only on correctness. As this is a large data structure (since its size is linear with respect to the number of registered stake credentials), it is very important that we also reduce the memory overhead as much as possible. Fortunately, we were able to add very little overhead for the deposits by using existing efficient data structures. The extra tracking now only incurs one word (8 bytes) per registered stake credential.
New Conway era transaction
We implemented the Conway era transaction body, which is in line with CIP-1694. Note that the Conway era implements, losing speaking, the parts of CIP-1694 that are not related to the liquid democracy (the "DReps"). The new transaction body adds the new governance actions and votes, while also deprecating the old governance structures (i.e. the old protocol parameter updates and MIR certificates).
We also now have the wire specification (CDDL file) and serialization code in place. The wire specification is still subject to change while we work on the Conway era, but it is now usable and has proper testing support (so that, for example, the serialization round-trips, etc).
Every since the release of the Shelley era, we have been working to reduce the computational load placed on the node by the ledger at the epoch boundary. While still not perfect, we believe that we have removed one of the final problematically long epoch boundary computations that exacerbate situations like this. In particular, the problem involved the way in which the consensus layer obtains a view of the ledger for the purposes of checking the leadership schedule in a new epoch. We implemented a stopgap measure which now only incurs a single multi-second cost once per epoch instead of potentially several multi-second costs while the networks waits for the first block of a new epoch to be minted.
We closed the year out with a lot of reduction to the technical debt!
Improved ledger event
- pull-3212 - The ledger events are not guaranteed to appear in any given order within a block.
For this reason, motivated by the use case in db-sync, the
TotalDepositsevent now includes a transaction ID and emits the change in deposits instead of the value.
Improved type saftey
- pull-3208 - We replaced
NominalDiffTimewith a newtype wrapper. The problem was that our CBOR encoders and decoders were using the wrong level of precision, having to due with with the Shelley genesis file. We removed the potential problem with a newtype wrapper.
- pull-3167 - We now use a GADT to ensure consistency of the Plutus language in the types
- pull-3175 - The Allegra and Mary eras had an unusual relationship in our codebase,
due to the uncertainly of release dates while we were implementing them.
In particular, they were coupled in way that is different from the rest of the code base.
With hindsight on our side, we split the combined
shelley-maHaskell package into two separate ledger era packages, which is now consistent with the rest of the repository and module structure.
- pull-3184 - We created a core test sub-library, cleaning up a lot of our property test generator code.
- pull-3210 - We moved the
KeyPairtype to the test library. Outside of testing, the ledger does not need to deal with signing keys, and since this is a topic that deserves the utmost care, it is best to make it clear that our use of signing keys is only for testing.
- pull-3229 - We split the
Cardano.Ledger.Alonzo.Datamodule, which is more consistent with the rest of the codebase.
Revert pointer address deprecation
Thanks to one of our excellent internal auditors, @jmhrpr, we now have a better plan for deprecating pointer addresses. This meant that we had to revert the previous work to deprecate them.
- pull-3205 - We removed deprecated type synonyms.
- pull-3218 - We cleaned up the address deserialization.
- pull-3223 - We fixed faulty address deserialization tests.
- pull-3222 - We switched to a general type family
TxOutfrom concrete ones, reducing many constraints.
- pull-3224 -
ShelleyGenesisis now parameterized by crypto instead of by era.
- pull-3170 - We set the cabal-version to 3.0 in our projects.
- pull-3172 - We removed the now useless
- pull-3225 - We switch from ormolu to fourmolu. The reason was to be able to finally have more diff friendly code!