Allow empty resolvers about graphql HOT 7 CLOSED

Hey @F21, I never tested it, but it could be done with the current graphql-js implementation? That would be such an interesting feature. A friend of mine ran into the same issue some weeks ago, we had to define the resolve in the parent even without needing all fields, for example:

{
  memory {
    virtual {
      total
      used
      available
    }
  }
}

In memory we have both virtual and swap, but we needed to resolve both virtual and swap data which leads to performance-related issues.

/cc @txgruppi

from graphql.

@CentaurWarchief I haven't really had a chance to play with the graphql-js version (only went through the relay tutorial). However, I think the javascript implementation also has this limitation: graphql/graphql-js#154

The current work around is to resolve in the virtual and swap object and set the resolver of memory to be empty:

Resolve: func(p graphql.ResolveParams) (interface{}, error) {
                    return struct{}{}, nil 
                }

from graphql.

Our solution was to create a type for memory and use it in the resolve function.

func NewMemoryData() *MemoryData {
    return &MemoryData{}
}

type MemoryData struct {
    Swap    *SwapData    `json:"swap"`
    Virtual *VirtualData `json:"virtual"`
}

Resolve: func(p graphql.ResolveParams) (interface{}, error) {
    return NewMemoryData(), nil
},

from graphql.

This is kinda tricky, because the way the executor works is by evaluating the query top-down.
As part of its pruning process, it would only look at the child if the parent node has something to evaluate.

I think it would be useful if a field that does not have a resolver is allowed to go "deeper", so that the child resolvers can resolve.

Every field actually has a default resolver, which you override by specifying it in the schema you defined.

So if the resolver (regardless if it's default or user-specified) returns a nil value, the executor won't try to see if its child fields has a value.

It is possible to design the executor to try evaluate all child fields even when the parent field evaluates to nil, but that could possibly lead to evaluating every field in the schema for every query.

In addition to that, having the value of resources field to be nil but its own fields to have values are kinda counter-intuitive.
Consider:

type Car struct {
    Name string
    Year int
}
var car *Car
log.Println(car.Year) // <-- panic

@F21 solution of returning an empty anonymous struct at first glance seems hacky but is perfectly valid.
(It returns an empty but non-nil value), equivalent to:

var car Car
log.Println(car.Year) // 0

Can I suggest an alternate solution (which is practically the same as the one posted above), but probably more natural way of writing the schema.

type Memory struct {
    Total float64 `json:"total"`
    Used float64 `json:"used"`
}

type Resources struct {
    Memory *Memory `json:"memory"`
}

func (resources *Resources) GetMemory() *Memory {
    // resource-heavy operation to fetch stats
    return &Memory {
        Total: 1024,
        Used: 100,
    }
}

func main() {
    resourceType := graphql.NewObject(graphql.ObjectConfig{
        Name:        "Resource",
        Description: "Resource utilization information for a type.",
        Fields: graphql.Fields{
            "total": &graphql.Field{
                Type:        graphql.Float,
                Description: "The total available.",
            },
            "used": &graphql.Field{
                Type:        graphql.Float,
                Description: "The amount used.",
            },
        },
    })

    resourcesType := graphql.NewObject(graphql.ObjectConfig{
        Name:        "Resources",
        Description: "Resource utilization information.",
        Fields: graphql.Fields{
            "memory": &graphql.Field{
                Type:        resourceType,
                Description: "Memory utilization information.",
                Resolve: func(p graphql.ResolveParams) (interface{}, error) {
                    // `p.Source` is given by it's parent field `resources`
                    if resources, ok := p.Source.(*Resources); ok {
                        return resources.GetMemory(), nil
                    }
                    return nil, nil
                },
            },
        },
    })
    queryType := graphql.NewObject(graphql.ObjectConfig{
        Name: "Query",
        Fields: graphql.Fields{
            "resources": &graphql.Field{
                Type: resourcesType,
                Resolve: func(p graphql.ResolveParams) (interface{}, error) {
                    // pass `&Resource{}` down to it's child fields, available as `p.Source`
                    return &Resources{}, nil
                },
            },
        },
    })
    ...
}

from graphql.

@sogko Thanks for posting the sample code. That really helped!

I think your solution is exactly what I am looking for! 😄

from graphql.

@sogko Not sure if you meant to do something else but, this bit:

log.Printf(car.Year) 

Panics because car.Year is an int while log.Printfexpects a string. If you change to Println or give it a format string it prints 0 as should be expected. Again, you may have meant something else but I just wanted to point that out.

from graphql.

@bbuck hahaha yeah thanks for pointing that out, it should be log.Println(), I just meant to give an example of getting a value of a field from a nil struct, thanks! 👍🏻

from graphql.