Thanks guys for your ideas! I feel like it is easier to
follow pg_protobuf 's method to design and implement pg_thrift
for a postgres beginner like me. I can refer pg_protobuf's way of
using functions, writing tests, etc. I will reconsider what's the
returned format for list, sets, and struct, etc. when I touch that
part. Right now, I assume inputs are a series of thrift bytes, and
try to implement decoding logics for simple types.