To me, it seems that version-change-safety and the usefulness of the generated code constitute a design tradeoff: If you mark a field as required, then the generated data structures can skip using Option/pointers, and this very common form of validation can be generated for free. If you disallow marking a field as required, then all fields must be checked for existence, even ones required for a system to function, which is quite a burden and will lead to developers having to write their own types anyway as a place to put their validated data into. If data is required to be present for an app to function, then why can't I be given the tools to express this, and benefit from the constraints applied to the data model?
Most of the time when I would like to use a schema-driven, efficient data format and code generation tool, the data contract doesn't change frequently. And when it does, assuming it's a backwards-incompatible change, I think I would be happy to generate a MyDataV2 message along with GetMyDataV2 method, allow existing clients to keep using the original version, and allow new or existing clients to use the newly supported structures at their leisure. Meanwhile, everyone that shares my schema can have much more idiomatic generated code, and in the most common cases won't have to write their own data types or be stuck with a bunch of `if data.x != null {` statements.
Protobufs are an amazing tool, but I think there is a need for a simpler tool which supports a restricted set of use cases cleanly and allows for wider expression of data models.
Most of the time when I would like to use a schema-driven, efficient data format and code generation tool, the data contract doesn't change frequently. And when it does, assuming it's a backwards-incompatible change, I think I would be happy to generate a MyDataV2 message along with GetMyDataV2 method, allow existing clients to keep using the original version, and allow new or existing clients to use the newly supported structures at their leisure. Meanwhile, everyone that shares my schema can have much more idiomatic generated code, and in the most common cases won't have to write their own data types or be stuck with a bunch of `if data.x != null {` statements.
Protobufs are an amazing tool, but I think there is a need for a simpler tool which supports a restricted set of use cases cleanly and allows for wider expression of data models.