[−][src]Crate juniper_from_schema

This library contains a procedural macro that reads a GraphQL schema file, and generates the corresponding Juniper macro calls. This means you can have a real schema file and be guaranteed that it matches your Rust implementation. It also removes most of the boilerplate involved in using Juniper.

Example

Schema:

schema {
  query: Query
  mutation: Mutation
}

type Query {
  // The directive makes the return value `FieldResult<String>`
  // rather than the default `FieldResult<&String>`
  helloWorld(name: String!): String! @juniper(ownership: "owned")
}

type Mutation {
  noop: Boolean!
}

How you could implement that schema:

#[macro_use]
extern crate juniper;

use juniper_from_schema::graphql_schema_from_file;

// This is the important line
graphql_schema_from_file!("tests/schemas/doc_schema.graphql");

pub struct Context;
impl juniper::Context for Context {}

pub struct Query;

impl QueryFields for Query {
    fn field_hello_world(
        &self,
        executor: &juniper::Executor<'_, Context>,
        name: String,
    ) -> juniper::FieldResult<String> {
        Ok(format!("Hello, {}!", name))
    }
}

pub struct Mutation;

impl MutationFields for Mutation {
    fn field_noop(&self, executor: &juniper::Executor<'_, Context>) -> juniper::FieldResult<&bool> {
        Ok(&true)
    }
}

fn main() {
    let ctx = Context;

    let query = "query { helloWorld(name: \"Ferris\") }";

    let (result, errors) = juniper::execute(
        query,
        None,
        &Schema::new(Query, Mutation),
        &juniper::Variables::new(),
        &ctx,
    )
    .unwrap();

    assert_eq!(errors.len(), 0);
    assert_eq!(
        result
            .as_object_value()
            .unwrap()
            .get_field_value("helloWorld")
            .unwrap()
            .as_scalar_value::<String>()
            .unwrap(),
        "Hello, Ferris!",
    );
}

And with graphql_schema_from_file! expanded your code would look something like this:

#[macro_use]
extern crate juniper;

pub struct Context;
impl juniper::Context for Context {}

pub struct Query;

juniper::graphql_object!(Query: Context |&self| {
    field hello_world(&executor, name: String) -> juniper::FieldResult<String> {
        <Self as QueryFields>::field_hello_world(&self, &executor, name)
    }
});

trait QueryFields {
    fn field_hello_world(
        &self,
        executor: &juniper::Executor<'_, Context>,
        name: String,
    ) -> juniper::FieldResult<String>;
}

impl QueryFields for Query {
    fn field_hello_world(
        &self,
        executor: &juniper::Executor<'_, Context>,
        name: String,
    ) -> juniper::FieldResult<String> {
        Ok(format!("Hello, {}!", name))
    }
}

pub struct Mutation;

juniper::graphql_object!(Mutation: Context |&self| {
    field noop(&executor) -> juniper::FieldResult<&bool> {
        <Self as MutationFields>::field_noop(&self, &executor)
    }
});

trait MutationFields {
    fn field_noop(&self, executor: &juniper::Executor<'_, Context>) -> juniper::FieldResult<&bool>;
}

impl MutationFields for Mutation {
    fn field_noop(&self, executor: &juniper::Executor<'_, Context>) -> juniper::FieldResult<&bool> {
        Ok(&true)
    }
}

type Schema = juniper::RootNode<'static, Query, Mutation>;

fn main() {
    let ctx = Context;

    let query = "query { helloWorld(name: \"Ferris\") }";

    let (result, errors) = juniper::execute(
        query,
        None,
        &Schema::new(Query, Mutation),
        &juniper::Variables::new(),
        &ctx,
    )
    .unwrap();

    assert_eq!(errors.len(), 0);
    assert_eq!(
        result
            .as_object_value()
            .unwrap()
            .get_field_value("helloWorld")
            .unwrap()
            .as_scalar_value::<String>()
            .unwrap(),
        "Hello, Ferris!",
    );
}

Example web app

You can find an example of how to use this library together with Rocket and Diesel to make a GraphQL web app at https://github.com/davidpdrsn/graphql-app-example or an example of how to use this library with Actix and Diesel at https://github.com/husseinraoouf/graphql-actix-example.

Customizing ownership

By default all fields return borrowed values. Specifically the type is juniper::FieldResult<&'a T> where 'a is the lifetime of self. This works well for returning data owned by self and avoids needless .clone() calls you would need if fields returned owned values.

However if you need to change the ownership you have to add the directive @juniper(ownership:) to the field in the schema.

It takes the following arguments:

@juniper(ownership: "borrowed"): The return type will be borrowed from self (FieldResult<&T>).
@juniper(ownership: "owned"): The return type will be owned (FieldResult<T>).
@juniper(ownership: "as_ref"): Only applicable for Option and Vec return types. Changes the inner type to be borrowed (FieldResult<Option<&T>> or FieldResult<Vec<&T>>).

All field arguments will be owned.

GraphQL features

The goal of this library is to support as much of GraphQL as Juniper does.

Here is the complete list of features:

Supported:

Object types including converting lists and non-nulls to Rust types
Custom scalar types including the ID type
Interfaces
Unions
Input objects
Enumeration types

Not supported yet:

Subscriptions (currently not supported by Juniper so we're unsure when or if this will happen)
Schema directives (@deprecated is supported)
Type extensions

The `ID` type

The ID GraphQL type will be generated into juniper::ID.

Custom scalar types

Custom scalar types will be generated into a newtype wrapper around a String. For example:

scalar Cursor

Would result in

pub struct Cursor(pub String);

Special case scalars

A couple of scalar names have special meaning. Those are:

Url becomes url::Url.
Uuid becomes uuid::Uuid.
Date becomes chrono::naive::NaiveDate.
DateTime becomes chrono::DateTime<chrono::offset::Utc> by default but if defined with scalar DateTime @juniper(with_time_zone: false) it will become chrono::naive::NaiveDateTime.

Juniper doesn't support chrono::Date so therefore this library cannot support that either. You can read about Juniper's supported integrations here.

Interfaces

Juniper has several ways of representing GraphQL interfaces in Rust. They are listed here along with their advantages and disadvantages.

For the generated code we use the enum pattern because we found it to be the most flexible.

Abbreviated example (find complete example here):

graphql_schema! {
    schema {
        query: Query
    }

    type Query {
        search(query: String!): [SearchResult!]! @juniper(ownership: "owned")
    }

    interface SearchResult {
        id: ID!
        text: String!
    }

    type Article implements SearchResult {
        id: ID!
        text: String!
    }

    type Tweet implements SearchResult {
        id: ID!
        text: String!
    }
}

pub struct Query;

impl QueryFields for Query {
    fn field_search(
        &self,
        executor: &Executor<'_, Context>,
        trail: &QueryTrail<'_, SearchResult, juniper_from_schema::Walked>,
        query: String,
    ) -> FieldResult<Vec<SearchResult>> {
        let article: Article = Article { id: ID::new("1"), text: "Business".to_string() };
        let tweet: Tweet = Tweet { id: ID::new("2"), text: "1 weird tip".to_string() };

        let posts = vec![
            SearchResult::from(article),
            SearchResult::from(tweet),
        ];

        Ok(posts)
    }
}

The enum that gets generated has variants for each type that implements the interface and also implements From<T> for each type.

Union types

Union types are basically just interfaces so they work in very much the same way.

Abbreviated example (find complete example here):

graphql_schema! {
    schema {
        query: Query
    }

    type Query {
        search(query: String!): [SearchResult!]! @juniper(ownership: "owned")
    }

    union SearchResult = Article | Tweet

    type Article {
        id: ID!
        text: String!
    }

    type Tweet {
        id: ID!
        text: String!
    }
}

pub struct Query;

impl QueryFields for Query {
    fn field_search(
        &self,
        executor: &Executor<'_, Context>,
        trail: &QueryTrail<'_, SearchResult, juniper_from_schema::Walked>,
        query: String,
    ) -> FieldResult<Vec<SearchResult>> {
        let article: Article = Article { id: ID::new("1"), text: "Business".to_string() };
        let tweet: Tweet = Tweet { id: ID::new("2"), text: "1 weird tip".to_string() };

        let posts = vec![
            SearchResult::from(article),
            SearchResult::from(tweet),
        ];

        Ok(posts)
    }
}

Input objects

Input objects will be converted into Rust structs with public fields.

Abbreviated example (find complete example here):

graphql_schema! {
    schema {
        query: Query
        mutation: Mutation
    }

    type Mutation {
        createPost(input: CreatePost!): Post @juniper(ownership: "owned")
    }

    input CreatePost {
        title: String!
    }

    type Post {
        id: ID!
        title: String!
    }

    type Query { noop: Boolean! }
}

pub struct Mutation;

impl MutationFields for Mutation {
    fn field_create_post(
        &self,
        executor: &Executor<'_, Context>,
        trail: &QueryTrail<'_, Post, juniper_from_schema::Walked>,
        input: CreatePost,
    ) -> FieldResult<Option<Post>> {
        let title: String = input.title;

        unimplemented!()
    }
}

From that example CreatePost will be defined as

pub struct CreatePost {
    pub title: String,
}

Enumeration types

GraphQL enumeration types will be converted into normal Rust enums. The name of each variant will be camel cased.

Abbreviated example (find complete example here):

graphql_schema! {
    schema {
        query: Query
    }

    enum Status {
        PUBLISHED
        UNPUBLISHED
    }

    type Query {
        allPosts(status: Status!): [Post!]! @juniper(ownership: "owned")
    }

    type Post {
        id: ID!
    }
}

pub struct Query;

impl QueryFields for Query {
    fn field_all_posts(
        &self,
        executor: &Executor<'_, Context>,
        trail: &QueryTrail<'_, Post, juniper_from_schema::Walked>,
        status: Status,
    ) -> FieldResult<Vec<Post>> {
        match status {
            Status::Published => unimplemented!("find published posts"),
            Status::Unpublished => unimplemented!("find unpublished posts"),
        }
    }
}

Default argument values

In GraphQL you are able to provide default values for field arguments, provided the argument is nullable.

Arguments of the following types support default values:

Float
Int
String
Boolean
Enumerations
Input objects (as field arguments, see below)
Lists containing some other supported type

Abbreviated example (find complete example here):

graphql_schema! {
    schema {
        query: Query
    }

    enum Status {
        PUBLISHED
        UNPUBLISHED
    }

    input Pagination {
        pageSize: Int!
        cursor: ID
    }

    type Query {
        allPosts(
            status: Status = PUBLISHED,
            pagination: Pagination = { pageSize: 20 }
        ): [Post!]! @juniper(ownership: "owned")
    }

    type Post {
        id: ID!
    }
}

pub struct Query;

impl QueryFields for Query {
    fn field_all_posts(
        &self,
        executor: &Executor<'_, Context>,
        trail: &QueryTrail<'_, Post, juniper_from_schema::Walked>,
        status: Status,
        pagination: Pagination,
    ) -> FieldResult<Vec<Post>> {
        // `status` will be `Status::Published` if not given in the query

        match status {
            Status::Published => unimplemented!("find published posts"),
            Status::Unpublished => unimplemented!("find unpublished posts"),
        }
    }
}

Input object gotchas

Defaults for input objects are only supported as field arguments. The following is not supported

input SomeType {
  field: Int = 1
}

This isn't supported because the spec is unclear about how to handle multiple nested defaults.

Also, defaults are only used if no arguments are passed. So given the schema

input Input {
  a: String
  b: String
}

type Query {
  field(arg: Input = { a: "a" }): Int!
}

and the query

query MyQuery {
  field(arg: { b: "my b" })
}

The value of arg inside the resolver would be Input { a: None, b: Some("my b") }. Note that even though a has a default value in the field doesn't get used here because we set arg in the query.

GraphQL to Rust types

This is how the standard GraphQL types will be mapped to Rust:

Int -> i32
Float -> f64
String -> String
Boolean -> bool
ID -> juniper::ID

Query trails

If you're not careful about preloading associations for deeply nested queries you risk getting lots of N+1 query bugs. Juniper provides a look ahead api which lets you inspect things coming up further down a query. However the API is string based, so you risk making typos and checking for fields that don't exist.

QueryTrail is a thin wrapper around Juniper look aheads with generated methods for each field on all your types. This means the compiler will reject your code if you're checking for invalid fields.

Fields that return object types (non scalar values) will also get a QueryTrail argument besides the executor.

Since the QueryTrail type itself is defined in this crate (rather than being inserted into your code) we cannot directly add methods for your GraphQL fields. Those methods have to be added through "extension traits". So if you see an error like

   |  trail.foo();
   |        ^^^ method not found in `&juniper_from_schema::QueryTrail<'a, User, juniper_from_schema::Walked>`
   |
   = help: items from traits can only be used if the trait is in scope
help: the following trait is implemented but not in scope, perhaps add a `use` for it:
   |
2  | use crate::graphql_schema::query_trails::QueryTrailUserExtensions;
   |

Then adding use crate::graphql_schema::query_trails::* to you module should fix it. This is necessary because all the extention traits are generated inside a module called query_trails. This is done so you can glob import the QueryTrail extension traits without glob importing everything from your GraphQL schema.

If you just want everything from the schema use crate::graphql_schema::* will also bring in the extension traits.

Abbreviated example

Find complete example here

graphql_schema! {
    schema {
        query: Query
    }

    type Query {
        allPosts: [Post!]! @juniper(ownership: "owned")
    }

    type Post {
        id: Int!
        author: User!
    }

    type User {
        id: Int!
    }
}

pub struct Query;

impl QueryFields for Query {
    fn field_all_posts(
        &self,
        executor: &Executor<'_, Context>,
        trail: &QueryTrail<'_, Post, juniper_from_schema::Walked>,
    ) -> FieldResult<Vec<Post>> {
        // Check if the query includes the author
        if let Some(_) = trail.author().walk() {
            // Somehow preload the users to avoid N+1 query bugs
            // Exactly how to do this depends on your setup
        }

        // Normally this would come from the database
        let post = Post {
            id: 1,
            author: User { id: 1 },
        };

        Ok(vec![post])
    }
}

pub struct Post {
    id: i32,
    author: User,
}

impl PostFields for Post {
    fn field_id(&self, executor: &Executor<'_, Context>) -> FieldResult<&i32> {
        Ok(&self.id)
    }

    fn field_author(
        &self,
        executor: &Executor<'_, Context>,
        trail: &QueryTrail<'_, User, juniper_from_schema::Walked>,
    ) -> FieldResult<&User> {
        Ok(&self.author)
    }
}

pub struct User {
    id: i32,
}

impl UserFields for User {
    fn field_id(
        &self,
        executor: &Executor<'_, Context>,
    ) -> FieldResult<&i32> {
        Ok(&self.id)
    }
}

Types

A query trial has two generic parameters: QueryTrail<'a, T, K>. T is the type the current field returns and K is either Walked or NotWalked.

The lifetime 'a comes from Juniper and is the lifetime of the incoming query.

`T`

The T allows us to implement different methods for different types. For example in the example above we implement id and author for QueryTrail<'_, Post, K> but only id for QueryTrail<'_, User, K>.

If your field returns a Vec<T> or Option<T> the given query trail will be QueryTrail<'_, T, _>. So Vec or Option will be removed and you'll only be given the inner most type. That is because in the GraphQL query syntax it doesn't matter if you're querying a User or [User]. The fields you have access to are the same.

`K`

The Walked and NotWalked types are used to check if a given trail has been checked to actually be part of a query. Calling any method on a QueryTrail<'_, T, K> will return QueryTrail<'_, T, NotWalked>, and to check if the trail is actually part of the query you have to call .walk() which returns Option<QueryTrail<'_, T, Walked>>. If that is a Some(_) you'll know the trail is part of the query and you can do whatever preloading is necessary.

Example:

if let Some(walked_trail) = trail
    .some_field()
    .some_other_field()
    .third_field()
    .walk()
{
    // preload stuff
}

You can always run cargo doc and inspect all the methods on QueryTrail and in which contexts you can call them.

Downcasting for interface and union `QueryTrail`s

This section is mostly relevant if you're using juniper-eager-loading however it isn't specific to that library.

If you have a QueryTrail<'a, T, Walked> where T is an interface or union type you can use .downcast() to convert that QueryTrail into one of the implementors of the interface or union.

Example:

graphql_schema! {
    schema {
        query: Query
    }

    type Query {
        search(query: String!): [SearchResult!]!
    }

    interface SearchResult {
        id: ID!
    }

    type Article implements SearchResult {
        id: ID!
    }

    type Tweet implements SearchResult {
        id: ID!
    }
}

pub struct Query;

impl QueryFields for Query {
    fn field_search(
        &self,
        executor: &Executor<'_, Context>,
        trail: &QueryTrail<'_, SearchResult, juniper_from_schema::Walked>,
        query: String,
    ) -> FieldResult<&Vec<SearchResult>> {
        let article_trail: QueryTrail<'_, Article, Walked> = trail.downcast();
        let tweet_trail: QueryTrail<'_, Tweet, Walked> = trail.downcast();

        // ...
    }
}

Why is this useful?

If you were do perform some kind of preloading of data you might have a function that inspects a QueryTrail and loads the necessary data from a database. Such a function could look like this:

fn preload_users(
    mut users: Vec<User>,
    query_trail: &QueryTrail<'_, User, Walked>,
    db: &Database,
) -> Vec<User> {
    // ...
}

This function works well when we have field that returns [User!]!. That field is going to get a QueryTrail<'a, User, Walked> which is exactly what preload_users needs.

However, now imagine you have a schema like this:

type Query {
    search(query: String!): [SearchResult!]!
}

union SearchResult = User | City | Country

type User {
    id: ID!
    city: City!
}

type City {
    id: ID!
    country: Country!
}

type Country {
    id: ID!
}

The method QueryFields::field_search will receive a QueryTrail<'a, SearchResult, Walked>. That type doesn't work with preload_users. So we have to convert our QueryTrail<'a, SearchResult, Walked> into QueryTrail<'a, User, Walked>.

This can be done by calling .downcast() which automatically gets implemented for interface and union query trails. See above for an example.

Customizing the error type

By default the return type of the generated field methods will be juniper::FieldResult<T>. That is just a type alias for std::result::Result<T, juniper::FieldError>. Should you want to use a different error type than juniper::FieldError that can be done by passing , error_type: YourType to graphql_schema_from_file!.

Just keep in that your custom error type must implement juniper::IntoFieldError to type check.

Example:

graphql_schema_from_file!("tests/schemas/doc_schema.graphql", error_type: MyError);

pub struct MyError(String);

impl juniper::IntoFieldError for MyError {
    fn into_field_error(self) -> juniper::FieldError {
        // Perform custom error handling
        juniper::FieldError::from(self.0)
    }
}

pub struct Query;

impl QueryFields for Query {
    fn field_hello_world(
        &self,
        executor: &Executor<'_, Context>,
        name: String,
    ) -> Result<String, MyError> {
        Ok(format!("Hello, {}!", name))
    }
}

graphql_schema! does not support changing the error type.

Customizing the context type

By default the generate code will assume your context type is called Context. If that is not the case you can customize it by calling graphql_schema_from_file! with context_type: NewName.

Example:

graphql_schema_from_file!("tests/schemas/doc_schema.graphql", context_type: MyContext);

pub struct MyContext;
impl juniper::Context for MyContext {}

pub struct Query;

impl QueryFields for Query {
    fn field_hello_world(
        &self,
        executor: &Executor<'_, MyContext>,
        name: String,
    ) -> juniper::FieldResult<String> {
        Ok(format!("Hello, {}!", name))
    }
}

graphql_schema! does not support changing the context type.

Inspecting the generated code

If you wish to see exactly what code gets generated you can set the env var JUNIPER_FROM_SCHEMA_DEBUG to 1 when compiling. For example:

JUNIPER_FROM_SCHEMA_DEBUG=1 cargo build

The code will not be formatted so it might be tricky to read. The easiest way to fix this is to copy the printed code to a file and run it through rustfmt.

Alternatively you can include the feature called "format-debug-output". This will run the output through rustfmt before printing it. That way you don't have to do that manually. Example Cargo.toml:

[dependencies]
juniper-from-schema = { version = "x.y.z", features = ["format-debug-output"] }

Unfortunately this requires that you're using nightly, because rustfmt requires nightly. It might also break your build because rustfmt doesn't always compile for some reason ¯\_(ツ)_/¯. If you experience this just remove the "format-debug-output" feature and format the output manually.

Macros

graphql_schema	Write your GraphQL schema directly in your Rust code.
graphql_schema_from_file	Read a GraphQL schema file and generate corresponding Juniper macro calls.

Structs

NotWalked	A type used to parameterize `QueryTrail` to know that `walk` has not been called.
QueryTrail	A wrapper around a `juniper::LookAheadSelection` with methods for each possible child.
Walked	A type used to parameterize `QueryTrail` to know that `walk` has been called.

Traits

MakeQueryTrail

Make a QueryTrail from something, normally a juniper::LookAheadSelection.