Data encoding

[klntsky]

This is a summary of a discussion that happened internally, here for visibility.

We have considered three approaches to data encoding.

Scott Encoding

• Used in Plutarch.
• https://plutonomicon.github.io/plutarch-plutus/Concepts/Data%20and%20Scott%20encoding

Sum of Products

• a more recent approach to data encoding specified in CIP-0085.

PlutusData

A.k.a. no encoding: using built-in Data primitive type that can represent any algebraic data type without function-as-values. This is the approach of Aiken.

Our considerations

🟡 Both SOP and Scott encoding introduce significant constant overhead for decoding from PlutusData
🟢 We decided to go with SOP, because IOG optimization efforts will be focused on it.
🔴 PlutusData is known to be less performant benchmarks
❌ PlutusData is not usable: we need functions in ADTs, because we need to implement type class instances as records.