Fair enough. Yes it is probably to do with sub messages.

Letmegrpc is some of the worst code I have ever wrote, but unfortunately it works well for the use cases I and others have and I never use sub messages, oneof or windows.
It working for the use cases we had makes motivating a rewrite quite hard.

I would love to be paid to rewrite this in Elm or Purescript and using gRPC reflection, so dynamically generate the form in the frontend in a type safe language, without the possibility of null pointer exceptions.

But unfortunately at the moment I am not even a user of protobufs :(

Ok, thanks for the info. I'm dropping my test and will just write a C# client since it's fast than fixing this :) I also have a lot of the reflection stuff done, since I use it in a configuration tool. So upgrading that to also have client functionality should be the fastest option. I will still leave this issue open, just in case someone else stumples upon it.

As someone who has done this before, I really recommend doing this all in the frontend. You don't need a backend to generate a gRPC GUI.