Hacker News Reader

Comment by 99999 | Hacker News Reader

I'm curious, in your view, why should protocols not use "oneOf"?

Not GP commenter, but a lot of the decisions to remove explicit optionality and requiredness from protocol buffer is that it made pb too complicated to change when you wanted to make a field required or optional, so they punted with 3 and said "do this in service logic"

I think the feeling is that this is the realm of business logic, and a protocol should be simple and fast and not concern itself with the rules part of the underlying service.

Oneof seems to fall under the same concept, where it says "one and only one field from this list" when they can just as easily do that in service or client logic (three fields, documentation and a server response custom written as to the WHY of oneof relationship)

I can't follow that argument, as it seems to apply just as much to wanting to change the type of a field, the name of a field or just making changes in general. Maybe I deal with relational databases too much, but I really miss being able to declare fields required and provide default values, because now instead I have to document, trust others will read the documentation, and provide explicit error handling for when constraints that used to be validated client side or server side by protobuf. I also miss not having a NULL value, but protobuf never had that and almost certainly due to dealing with too many relational databases ;)

If I lose oneof too, I might as well be using JSON.

Your points are valid, but ultimately an API is defined by its semantics in addition to data types. For anything beyond extremely primative RPC, you have to read the docs anyways.

Moving to proto3 means putting all your validation in one place, your serialization primitives in one place, and your docs in one place, instead of smattering them all over. It's a lot less error prone and way more maintenance friendly.

Why is `// required, the server will respond with a failure if this argument isn't provided` in the .proto file enough documentation? That's usually what I go to.

lowmagnet more or gave a succinct summary. I've also now added other comments here, discussing my views. I don't believe it's necessary, and I would proscribe other ways to accomplish the same design. It's a philosophy of API design, not some kind of absolute stricture.

Reply on news.ycombinator.com