@smallcircles @mauve @silverpill Good points.
The rational for choosing Borsh is that we want a simple, binary serialization that has only one canonical byte representation for any given data.
This "one canonical representation" feature ensures that two components/schemas/etc. with the same data will get exactly the same hash, maximizing deduplication for data across the network.
I really like how simple Borsh is, but I should verify that CBOR doesn't already do what we need, too. 👍️
@zicklag @smallcircles @silverpill If you want stability in serialization you may be interested at looking at IPLD and soecifically the dag-cbor codec.
@mauve @smallcircles @silverpill After thinking more about it I realized that CBOR isn't really an option anyway because we specifically want a non-self-describing format.
This allows the description of the data to be stored separately in immutable Schemas, that Components reference by hash.
@mauve @smallcircles @silverpill So it looks like CBOR by itself doesn't have the canonical propery we need. DAG-CBOR adds that canonical property, along with the concept of CID links.
In our case we don't use CIDs, so we can't directly use either CBOR or DAG-CBOR without either adding our own specification/modifications on top or having extra stuff that we don't need.
So it seems like Borsh is still the simplest option.