I see some discussion here about protobufs being widely used and preventing innovation because it's good enough and comes from Google.

I see a second effect that's probably more regressive: using protobufs (designed for RPC) as a database schema definition language or a disk data structure description language.

Much prefer something like typespec to describe types.And then derive the rpc and disk schema languages from that programatically.

My first attempt was to use the flatbuffer schema language as a starting point. But being attached to a serialization format can be a net negative. Did not find traction.

Motivation: can not stand field numbers in a database schema when other migration mechanisms exist.