Batching experiments with different server-side GraphQL implementations

scenario.graphql

fragment profileFragment on ProfileRef {
    __typename
 
    ... on ActiveProfileRef {
      id
      firstName
      lastName
    }
 
    ... on InactiveProfileRef {
      id
    }
}
 
query profileRefQuery {
  me {
    profileRef {
      ... profileFragment
    } 
     
    profileVisits {
      collection {
        visitorConnection {
          ... profileFragment
        }
      }
      total
    }
  }
   
  profileRefs(ids: [2,4,5]) {
    ... profileFragment
  }
}

Behind me/profileRef, profileVisits/visitorConnection and profileRefs are batched resolvers that point to the same REST API. profileVisits points to another one.

• Sangria and graphql-ruby / graphql-batch manage to do this in 2 requests
• graphql-js / dataloader does 3 requests (1 per level in hierarchie as far as I can tell)
• absinthe does 4 requests (basically only batches the visitorConnection together)

The main reason that dataloader and absinthe differ here is presumably because me is synchronous, so profileRef and profileRefs get resolved within the same tick. Can you clarify the details of which fields are actually making data requests vs resolving instantly? It's hard to conclude anything based on these numbers alone.