F# implementation of Facebook GraphQL query language specification.

type Person = 
    { FirstName: string
      LastName: string }

// Define GraphQL type 
let PersonType = Define.Object(
    name = "Person",
    fields = [
        // Property resolver will be auto-generated
        Define.AutoField("firstName", String)   
        // Asynchronous explicit member resolver
        Define.AsyncField("lastName", String, resolve = fun context person -> async { return person.LastName })
    ])

// Include person as a root query of a schema
let schema = Schema(query = PersonType)
// Create an Exector for the schema
let executor = Executor(schema)

// Retrieve person data
let johnSnow = { FirstName = "John"; LastName = "Snow" }
let reply = executor.AsyncExecute(parse "{ firstName, lastName }", johnSnow) |> Async.RunSynchronously
// #> { data: { "firstName", "John", "lastName", "Snow" } } 

It's type safe. Things like invalid fields or invalid return types will be checked at compile time.

Go to GraphiQL sample directory. In order to run it, build and run FSharp.Data.GraphQL.Samples.GiraffeServer project on Debug settings - this will create a Giraffe server compatible with GraphQL spec, running on port 8084. Then what you need is to run node.js graphiql frontend. To do so, run npm i to get all dependencies, and then run npm run serve | npm run dev - this will start a webpack server running on http://localhost:8090/ . Visit this link, and GraphiQL editor should appear. You may try it by applying following query:

{
  hero(id:"1000") {
    id,
    name,
    appearsIn,
    homePlanet,
    friends {
      ... on Human {
        name
      }
      ... on Droid {
        name
      }
    }
  }
}

A second sample is a F#-backed version of of popular Relay Starter Kit - an example application using React.js + Relay with Relay-compatible server API.

To run it, build FSharp.Data.GraphQL and FSharp.Data.GraphQL.Relay projects using Debug settings. Then start server by running server.fsx script in your FSI - this will start relay-compatible F# server on port 8083. Then build node.js frontend by getting all dependencies (npm i) and running it (npm run serve | npm run dev) - this will start webpack server running React application using Relay for managing application state. You can visit it on http://localhost:8083/ .

In order to update client schema, visit http://localhost:8083/ and copy-paste the response (which is introspection query result from current F# server) into data/schema.json.

You can create and use middlewares on top of the Executor<'Root> object.

The query execution process through the use of the Executor involves three phases:

• Schema compile phase: this phase happens when the Executor<'Root> class is instantiated. In this phase, the Schema map of types is used to build a field execute map, which contains all field definitions alongside their field resolution functions. This map is used later on the planning and execution phases to retrieve the values of the queried fields of the schema.

• Operation planning phase: this phase happens before running a query that has no execution plan. This phase is responsible to analyze the AST document generated by the query, and build an ExecutionPlan to execute it.

• Operation execution phase: this phase is actually the phase that executes the query itself. It needs an execution plan, so, it commonly happens after the operation planning phase.

All those phases wraps needed data to do the phase job inside an Context object. They are expressed internally by functions:

let internal compileSchema (ctx : SchemaCompileContext) : unit =
  // ...

let internal planOperation (ctx: PlanningContext) : ExecutionPlan =
  // ...

let internal executeOperation (ctx : ExecutionContext) : AsyncVal<GQLResponse> =
  // ...

That way, in the compile schema phase, the Schema is modified and execution maps are generated inside the SchemaCompileContext object. On the operation planning phase, values of the PlanningContext object are used to generate an execution plan, and finally, this plan is passed alongside other values in the ExecutionContext object to the operation execution phase, wich finally uses them to execute the query and generate a GQLResponse.

With that being said, a middleware can be used to intercept each of those phases, and make customizations to them, modifying operations as needed. Each middleware must be implemented as a function with specific signature, and wrapped inside an IExecutorMiddleware interface:

type SchemaCompileMiddleware = 
    SchemaCompileContext -> (SchemaCompileContext -> unit) -> unit

type OperationPlanningMiddleware = 
    PlanningContext -> (PlanningContext -> ExecutionPlan) -> ExecutionPlan

type OperationExecutionMiddleware =
    ExecutionContext -> (ExecutionContext -> AsyncVal<GQLResponse>) -> AsyncVal<GQLResponse>

type IExecutorMiddleware =
    abstract CompileSchema : SchemaCompileMiddleware option
    abstract PlanOperation : OperationPlanningMiddleware option
    abstract ExecuteOperationAsync : OperationExecutionMiddleware option

Optionally, for ease of implementation, concrete class to derive from can be used, receiving only the optional sub-middleware functions in the constructor:

type ExecutorMiddleware(?compile, ?plan, ?execute) =
    interface IExecutorMiddleware with
        member __.CompileSchema = compile
        member __.PlanOperation = plan
        member __.ExecuteOperationAsync = execute

Each of the middleware functions acts like an intercept function, with two parameters: the context of the phase, the function of the next middleware (or the actual phase itself, wich is the last to run), and the return value. Those functions can be passed as an argument to the constructor of the Executor<'Root> object:

let middlewares = [ ExecutorMiddleware(compileFn, planningFn, executionFn) ]
let executor = Executor(schema, middlewares)

A simple example of a practical middleware can be one that measures the time needed to plan a query, and returns it on the Metadata of the planning context. The metadata object is a Map<string, obj> implementation that acts like a bag of information to be passed through each phase, until it is returned inside the GQLResponse object. You can use it to thread custom information through middlewares:

let planningMiddleware (ctx : PlanningContext) (next : PlanningContext -> ExecutionPlan) =
    let watch = Stopwatch()
    watch.Start()
    let result = next ctx
    watch.Stop()
    let metadata = result.Metadata.Add("planningTime", watch.ElapsedMilliseconds)
    { result with Metadata = metadata }

There are some built-in middlewares inside FSharp.Data.GraphQL.Server.Middlewares package:

This middleware can be used to place weights on fields of the schema. Those weightened fields can now be used to protect the server from complex queries that otherwise could be used to create things like a DDoS attack.

When defining a field, we use the extension method WithQueryWeight to place a weight on it:

let resolveFn (h : Human) =
  h.Friends |> List.map getCharacter |> List.toSeq

let field =
  Define.Field("friends", ListOf (Nullable CharacterType), 
    resolve = resolveFn).WithQueryWeight(0.5)

Then we define the threshold middleware for the Executor. If we execute a query that ask for "friends of friends" in a recursive way, the executor will only accept nesting them 4 times before the query exceeds the weight threshold of 2.0:

let middlewares = [ Define.QueryWeightMiddleware(2.0) ]

This middleware can be used to automatically generate a filter for list fields inside an object of the schema. This filter can be passed as an argument for the field on the query, and recovered in the ResolveFieldContext argument of the resolve function of the field.

For example, we can create a middleware for filtering list fields of an Human object, that are of the type Character option:

let middlewares = [ Define.ObjectListFilterMiddleware<Human, Character option>() ]

The filter argument is an object that is mapped through a JSON definition inside an filter argument on the field. A simple example would be filtering friends of a hero that have their names starting with the letter A:

query TestQuery {
    hero(id:"1000") {
        id
        name
        appearsIn
        homePlanet
        friends (filter : { name_starts_with: "A" }) {
            id
            name
        }
    }
}

This filter is mapped by the middleware inside an ObjectListFilter definition:

type FieldFilter<'Val> =
    { FieldName : string
      Value : 'Val }

type ObjectListFilter =
    | And of ObjectListFilter * ObjectListFilter
    | Or of ObjectListFilter * ObjectListFilter
    | Not of ObjectListFilter
    | Equals of FieldFilter<System.IComparable>
    | GreaterThan of FieldFilter<System.IComparable>
    | LessThan of FieldFilter<System.IComparable>
    | StartsWith of FieldFilter<string>
    | EndsWith of FieldFilter<string>
    | Contains of FieldFilter<string>
    | FilterField of FieldFilter<ObjectListFilter>

And the value recovered by the filter in the query is usable in the ResolveFieldContext of the resolve function of the field. To easy access it, you can use the extension method Filter, wich returns an ObjectListFilter option (it does not have a value if the object doesn't implement a list with the middleware generic definition, or if the user didn't provide a filter on the query).

Define.Field("friends", ListOf (Nullable CharacterType),
    resolve = fun ctx (d : Droid) -> 
        ctx.Filter |> printfn "Droid friends filter: %A"
        d.Friends |> List.map getCharacter |> List.toSeq)

By retrieving this filter on the field resolution context, it is possible to use client code to customize the query against a database, for example, and extend your GraphQL API features.

You can use extension methods provided by the FSharp.Data.GraphQL.Shared package to help building your own middlewares. When making a middleware, often you will need to modify schema definitions to add features to the schema defined by the user code. The ObjectListFilter middleware is an example, where all fields that implements lists of a certain type needs to be modified, by accepting an argument called filter.

As field definitions are immutable by default, generating copies of them with improved features can be a hard work sometimes. This is where the extension methods can help: for example, if you need to add an argument to an already defined field inside the schema compile phase, you can use the method WithArgs of the FieldDef<'Val> interface:

let field : FieldDef<'Val> = // Search for field inside ISchema...
let arg : Define.Input("id", String)
let fieldWithArg = field.WithArgs([ arg ])

To see the complete list of extensions used for improve definitions, you can take a look at the TypeSystemExtensions module of the FSharp.Data.GraphQL.Shared package.