Multi-directional relationships will cause the runtime to fail when building the schema

[sajeetharan]

Currently the engine throws an error with the getting started sample

Here is the query to get list of books and authors:

{
  books(filter: { title: { eq: "Foundation" } })
  {
    items {
      id
      title
      authors {
          first_name
          last_name
        }
    }
  }
}

and the schema as follows,

type Author @model {
    id : ID,
    first_name : String,
    middle_name: String,
    last_name: String,
    Books: [Book]
}

type Book @model {
    id : ID,
    title : String,
    Authors: [Author]
}

it throws an error as The runtime refused to evaluate expression at this time

[aaronpowell]

I'm currently having a look at this (managed to write a unit test that can validate schemas in a much simpler way).

The problem is that when it is trying to create the InputType's for the @model types, it becomes recursive, and as a result, crashes the runtime.

Now this is only a problem on the MutationBuilder phase of the schema compiler, the QueryBuilder doesn't have this problem, despite having recursive input types (on the filter), so I'm investigating whether something changed or was just missed.

[aaronpowell]

Actually, it does raise an interesting question, how do we handle relationships within a mutation?

The problem exists because it's trying to create an input type that allows you to insert a multiple entities in a single pass, so you could do this:

mutation {
  createBook(input: { id: "...", title: "...", Authors: [{ id: "...", first_name: "...", last_name: "...", middle_name: "...", Books: [{ id: "...", title: "..." }] }]) {
    id
  }
}

Is this expected? Should we be allowing you to create recursive items such as this? I don't think you should be able to.

Note: This isn't just a problem with Cosmos, this will manifest itself in the SQL variations as well, but it's a bit harder to do as you spend a lot more time setting up the relationships. This doesn't happen in SQL, see following comment.

[aaronpowell]

After doing some more digging, I've found that this bug is caused by a change introduced with #531

It doesn't impact the SQL engine, as it's only allowing nested models on Cosmos.

Looking at the tests that were included, they only tackle one-way relationships, not two-way, which is likely why it was never picked up then.