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.

Table of contents

• Example
• Example web app
• Customizing ownership
• GraphQL features
  • The ID type
  • Custom scalar types
  • Interfaces
  • Union types
  • Input objects
  • Enumeration types
  • Default argument values
• GraphQL to Rust types
• Query trails
• Customizing the error type
• Inspecting the generated code

Example

Schema:

```graphql schema { query: Query mutation: Mutation }

type Query { // this makes the return value FieldResult<String> // rather than the default FieldResult<&String> "#[ownership(owned)]" helloWorld(name: String!): String! }

type Mutation { noop: Boolean! } ```

How you could implement that schema:

```rust

[macro_use]

extern crate juniper;

use juniperfromschema::graphqlschemafrom_file;

// This is the important line graphqlschemafromfile!("tests/schemas/docschema.graphql");

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

pub struct Query;

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

pub struct Mutation;

impl MutationFields for Mutation { fn fieldnoop(&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:

```rust

[macro_use]

extern crate juniper;

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

pub struct Query;

juniper::graphqlobject!(Query: Context |&self| { field helloworld(&executor, name: String) -> juniper::FieldResult { ::fieldhelloworld(&self, &executor, name) } });

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

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

pub struct Mutation;

juniper::graphqlobject!(Mutation: Context |&self| { field noop(&executor) -> juniper::FieldResult<&bool> { ::fieldnoop(&self, &executor) } });

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

impl MutationFields for Mutation { fn fieldnoop(&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

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 return owned values (such as values queried from a database) you have to annotate the field in the schema with #[ownership(owned)].

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 - 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:

graphql scalar Cursor

Would result in

rust pub struct Cursor(pub String);

Date and DateTime are the two exceptions to this. Date gets converted into chrono::naive::NaiveDate and DateTime into chrono::DateTime<chrono::offset::Utc>.

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):

```rust # graphql_schema! { schema { query: Query }

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

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 fieldsearch( &self, executor: &Executor<', Context>, trail: &QueryTrail<', SearchResult, Walked>, query: String, ) -> FieldResult> { let article: Article = Article { id: ID::new("1"), text: "Business".tostring() }; 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):

```rust # graphql_schema! { schema { query: Query }

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

union SearchResult = Article | Tweet

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

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

}

pub struct Query;

impl QueryFields for Query { fn fieldsearch( &self, executor: &Executor<', Context>, trail: &QueryTrail<', SearchResult, Walked>, query: String, ) -> FieldResult> { let article: Article = Article { id: ID::new("1"), text: "Business".tostring() }; 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):

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

type Mutation {
    "#[ownership(owned)]"
    createPost(input: CreatePost!): Post
}

input CreatePost {
    title: String!
}

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

type Query { noop: Boolean! }

}

pub struct Mutation;

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

    unimplemented!()
}

} ```

From that example CreatePost will be defined as

rust 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):

```rust # graphql_schema! { schema { query: Query }

enum Status {
    PUBLISHED
    UNPUBLISHED
}

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

type Post {
    id: ID!
}

}

pub struct Query;

impl QueryFields for Query { fn fieldallposts( &self, executor: &Executor<', Context>, trail: &QueryTrail<', Post, Walked>, status: Status, ) -> FieldResult> { 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 - Lists containing some other supported type

Input objects are currently not supported as default arguments, but might be in the future.

You also cannot have null as the default value. In that case you might as well not have a default value.

Abbreviated example (find complete example here):

```rust # graphql_schema! { schema { query: Query }

enum Status {
    PUBLISHED
    UNPUBLISHED
}

type Query {
    "#[ownership(owned)]"
    allPosts(status: Status = PUBLISHED): [Post!]!
}

type Post {
    id: ID!
}

}

pub struct Query;

impl QueryFields for Query { fn fieldallposts( &self, executor: &Executor<', Context>, trail: &QueryTrail<', Post, Walked>, status: Status, ) -> FieldResult> { // 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"),
    }
}

} ```

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 objects types (non scalar values) will also get a QueryTrail argument besides the executor.

Abbreviated example

Find complete example here

```rust # graphql_schema! { schema { query: Query }

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

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

type User {
    id: Int!
}

}

pub struct Query;

impl QueryFields for Query { fn fieldallposts( &self, executor: &Executor<', Context>, trail: &QueryTrail<', Post, Walked>, ) -> FieldResult> { // 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 fieldid(&self, executor: &Executor<', Context>) -> FieldResult<&i32> { Ok(&self.id) }

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

}

pub struct User { id: i32, }

impl UserFields for User { fn fieldid( &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:

rust 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.

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:

```rust graphqlschemafromfile!("tests/schemas/docschema.graphql", error_type: MyError);

pub struct MyError(String);

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

pub struct Query;

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

[graphql_schema!] does not support changing the error 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:

bash 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:

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.

License: MIT