Motivation

Currently our decoders serve as adapter only to dill pickle. Because they type information is stored in the pickle, you can de-serialize a pickle and get the correct type without having to specify it at deserialization. Unfortunately this process breaks down when you've moved or renamed a class that you want to deserialize into.

So to be able to de-couple code and data serialization, and separate reader/writer schema, we need the decoder to be type-aware. The Encoder has to be type aware for certain operations as well, for example if we pass a TypedDict to encode, we won't know which type to encode since python doesn't retain type information for typed dicts. So this is just to make the encoding API a more flexible.

Changes

• state Encoder and Decoder APIs now optionally take a expected_type argument that allows an underlying implementation to be specific about the python type to serialize or deserialize into