Add fake response testing

[sztomi]

This is a continuation of #13 using fake. This still doesn't test all possible response structures, but it tests a lot more. It might be useful to implement a form of fuzzing where the seeds can be recorded and similar tests can be ran. That shouldn't run in Github Actions though, because it will be potentially unstable.

[sztomi]

This is added to be able to implement the Dummy trait for Value.

[pileghoff]

Overall it looks good, but i must admit, i haven't looked too deep into how all this works. Two things i noticed while playing around:

1. I wont detect if i remove Option<_> from a member (but it will detect if you add one, so thats good!). Not sure what can be done about that.
2. Because it does not test the entire message, but only the body, it did not catch that the raw string variant of ResponseMessage is wrong, as we need flatten it. Currently its wrapped inError(_). Im not sure how important it is to test the entire message in general, especially with all the issues you are encountering with the schema.

[sztomi]

Because it does not test the entire message, but only the body, it did not catch that the raw string variant of ResponseMessage is wrong, as we need flatten it.

Can you give me an example?

[pileghoff]

Because it does not test the entire message, but only the body, it did not catch that the raw string variant of ResponseMessage is wrong, as we need flatten it.

Can you give me an example?

fn main() {
  let a = Response {
    request_seq: 1,
    success: false,
    message: Some(ResponseMessage::Error("test".to_string())),
    body: None,
  };
  println!("{}", serde_json::to_string(&a).unwrap());
}

prints

{"request_seq":1,"success":false,"message":{"Error":"test"}}

but should print

{"request_seq":1,"success":false,"message":"test"}

[pileghoff]

I also just now noticed that because we moved the dap source folder, cargo run --examples no longer works, as the examples folder also needs to be moved into the dap folder.

[sztomi]

Thanks, I filed an issue about the error flattening. It's a good point that we probably should test entire messages. That might be useful future work or we might ultimately end up just iterating enough (and adding individual regression tests) that we weed out all of these mistakes.

I'll look into making the examples work again. I remember this being painful in workspaces.

[sztomi]

oh that somehow didn't get autosquashed when I rebased ... I'll amend on main.