[leoyvens]

Resolves #583 (the url should be tracked in a different issue).

This adds the entityCount field. Instead of storing it with the data and incrementing on every entity insert, which would be error-prone, this introduces the concept of a "calculated field", which is a field of an entity that is not in the store but rather calculated on the fly. We can probably also use this for __typename.

[Jannis]

Overall, this is great, but I would do this part differently. Adding computed fields like this will mean that we'll always count entities for all subgraphs, even if entityCount is not part of the query.

What I would do instead is resolve fields like this lazily. Here's what I'd recommend:

1. Change the signature of Resolver::resolve_scalar_value so that it takes the object type and the field name:

   fn resolve_scalar_value(
       &self,
       object_type: &s::ObjectType,
       field_name: &q::Name,
       scalar_type: &s::ScalarType,
       value: Option<&q::Value>,
   ) -> q::Value {

2. Implement resolve_scalar_value in StoreResolver. Something like this should do:

    fn resolve_scalar_value(
        &self,	
        object_type: &s::ObjectType,
        field_name: &q::Name,
        scalar_type: &s::ScalarType,
        value: Option<&q::Value>,
    ) -> q::Value {
        match field_name.as_str() {
          "entityCount" => {
            ... parse subgraph ID, if possible, fall back to value ...
            ... obtain entity count ...
          },
          _ => value
        }.and_then(|value| value.coerce(scalar_type)).unwrap_or(q::Value::Null)
    }

[Jannis]

But I'm not completely convinced of my approach because it makes adding support for GraphQL interfaces harder (which don't have ObjectType). Perhaps let's go with your approach first.

[leoyvens]

Overall, this is great, but I would do this part differently. Adding computed fields like this will mean that we'll always count entities for all subgraphs

I hadn't thought about this. I'll follow your recommendation then and keep things as they are for now, we can revisit the general issue of overfetching things later.