Prisma1: Support using Input Types for Mutation arguments

[ThisIsMissEm]

Currently when you do a mutation, you just have createPost(field1, field2, ... , fieldN), which means it's quite difficult to use variables on complex mutations.

A better route would be to allow a input type, for instance: createPost(post: Post) — in this case just using the Post type.

A further extension on this would be to allow using "Input Types", which are effectively subsets of the entity's Type, for instance: createPost(post: CreatePostInput)

Where Post is defined as:

type Post {
    id: ID!
    createdAt: DateTime
    updatedAt: DateTime
    title: String!
    body: String!
    poster: User!
}

And CreatePostInput is defined as:

type CreatePostInput {
    title: String!
    body: String!
}

By allowing these "Input Types", we can effectively achieve finer grain creation. It also prevents code like the following (taken from graphql-training/graphql-training.github.io)

// You can also use `graphql` for GraphQL mutations
export default graphql(gql`
  mutation createTrainingRequest(
      $name: String!
      $email: String!
      $companyName: String
      $attendees: Int!
      $modules: [TRAINING_REQUEST_MODULES!]!
      $additionalDetails: String
      $city: String!
      $country: String!
      $diversityRequested: Boolean!
      $onsiteRequested: Boolean!
) {
    createTrainingRequest(
        name: $name
        email: $email
        companyName: $companyName
        attendees: $attendees
        modules: $modules
        additionalDetails: $additionalDetails
        city: $city
        country: $country
        diversityRequested: $diversityRequested
        onsiteRequested: $onsiteRequested
    ) {
      id
    }
  }
`)(RequestForm);

Instead, I could've just done:

// You can also use `graphql` for GraphQL mutations
export default graphql(gql`
  mutation createTrainingRequest(
      $trainingRequest: CreateTrainingRequestInput!
) {
    createTrainingRequest(trainingRequest: $trainingRequest) {
      id
    }
  }
`)(RequestForm);

[ThisIsMissEm]

This is inspired by this example in the GraphQL Documentation:

[schickling]

Would be using the Relay API an option for you as well?

[ThisIsMissEm]

Maybe, but it greatly complicates things. I'd say using Input Types is a general great practice that should be encouraged.

[schickling]

There are a couple of alternatives from the top of my head:

1) Embed the input type into existing mutations. The upside would be that you could choose between providing a "combined object" matching input type vs spreading out all fields directly. The big downside of this approach is that all arguments would need to be optional (non-required). That means we'd lose a lot of the type system's power along the way.

2) An alternative approach would be to generate additional mutations like createPostWrapped(post: { ... }. The upside of this would be that it preserves the expressiveness of the typesystem. But obviously this approach would clutter the query namespace quite a lot. (This could be solve though by selectively configuring which queries should be exposed in the schema at all).

3) A third approach would be to make it a configuration option on a project basis which "layout" of mutations you'd prefer. This could however confuse people quite a bit.

Would be interested to hear what your thoughts are on this and especially if you'd see any other viable options?

[ThisIsMissEm]

I'd say a configuration option, perhaps per model, would be best.

[ThisIsMissEm]

But also, new projects should be created in the input type mode, and allow you to switch if you want.

[ThisIsMissEm]

👍 to being able to select which mutations and queries are exposed. This would be a great feature

[marktani]

We're looking to enhance the Simple API with an object input argument as described in this feature request. This would be a great opportunity to work on other feature requests that require an input object, like #45.

[sorenbs]

This is happening :-) Thanks for the great suggestion! https://github.com/graphcool/framework/issues/353