yume/juniper
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
juniper/juniper/CHANGELOG.md
Christoph Herzog 2f5e3ab29b 0.11.1 Release 
* Changelog
* Cargo.toml version bumps
2018-12-19 03:53:33 +01:00

20 KiB
Raw Blame History

master

  • The minimum required Rust version is now 1.30.0.

[0.11.1] 2018-12-19

Changes

  • The minimum required Rust version is now 1.30.

  • All macros and the custom derives now support the macro system changes properly and also support Rust 2018 edition crates.

    #298

[0.11.0] 2018-12-17

Changes

  • The minimum required Rust version is now 1.30.0.

    #271

  • Juniper is now generic about the exact representation of scalar values. This allows downstream crates to add support for own scalar value representations.

    There are two use cases for this feature:

    • You want to support new scalar types not representable by the provided default scalar value representation like for example i64
    • You want to support a type from a third party crate that is not supported by juniper

    Note: This may need some changes in down stream code, especially if working with generic code. To retain the current behaviour use DefaultScalarValue as scalar value type

    #251

  • The GraphQLObject and GraphQLEnum derives will mark graphql fields as @deprecated when struct fields or enum variants are marked with the builtin #[deprecated] attribute.

    The deprecation reason can be set using the note = ... meta item (e.g. #[deprecated(note = "Replaced by betterField")]). The since attribute is ignored.

    #269

  • There is an alternative syntax for setting a field's description and deprecation reason in the graphql_object! and graphql_interface! macros.

    To deprecate a graphql field:

    // Original syntax for setting deprecation reason
field deprecated "Reason" my_field() -> { ... }

// New alternative syntax for deprecation reason.
#[deprecated(note = "Reason")]
field my_field() -> { ... }

// You can now also deprecate without a reason.
#[deprecated]
field my_field() -> { ... }

    To set the description of a graphql field:

    // Original syntax for field descriptions
field my_field() as "Description" -> { ... }

// Original syntax for argument descriptions
field my_field(
  floops: i32 as "The number of starfish to be returned. \
                  Can't be more than 100.",
) -> {
  ...
}

// New alternative syntax for field descriptions
/// Description
field my_field() -> { ... }

// New alternative syntax for argument descriptions
field my_field(
  /// The number of starfish to be returned.
  /// Can't be more than 100.
  arg: i32,
) -> {
  ...
}

// You can also use raw strings and const &'static str.
//
// Multiple docstrings will be collapsed into a single
// description separated by newlines.
/// This is my field.
///
/// Make sure not to filtz the bitlet.
/// Flitzing without a bitlet has undefined behaviour.
///
#[doc = my_consts::ADDED_IN_VERSION_XYZ]
field my_field() -> { ... }

    #269

[0.10.0] 2018-09-13

Changes

  • Changed serialization of NaiveDate when using the optional chronos support.

    Note: while this is not a Rust breaking change, if you relied on the serialization format (perhaps by storing serialized data in a database or making asumptions in your client code written in another language) it could be a breaking change for your application.

    #151

  • The GraphQLObject, GraphQLInputObject, and GraphQLEnum custom derives will reject invalid names at compile time.

    #170

  • Large integers (> signed 32bit) are now deserialized as floats. Previously, they produced an "integer out of range" error. For languages that do not have a distinction between integer and floating point types (such as javascript), this meant large whole floating point values could not be decoded (because they were represented without a fractional value such as .0).

    #179

  • The GraphQLObject, GraphQLInputObject, and GraphQLEnum custom derives now parse doc strings and use them as descriptions. This behavior can be overridden by using an explicit GraphQL description annotation such as #[graphql(description = "my description")]. View documentation.

    #194

  • Introduced IntoFieldError trait to allow custom error handling i.e. custom result type. The error type must implement this trait resolving the errors into FieldError. View documentation.

    #40

  • GraphQLType and ToInputValue are now implemented for Arc

    #212

  • Error responses no longer have a data field, instead, error details are stored in the extensions field

    Note: while this is a breaking change, it is a necessary one to better align with the latest GraphQL June 2018 specification, which defines the reserved extensions field for error details. View documentation.

    #219

  • The GraphQLObject and GraphQLInputObject custom derives now support lifetime annotations.

    #225

  • When using the GraphQLObject custom derive, fields can now be omitted by annotating the field with #[graphql(skip)]. View documentation.

    #220

  • Due to newer dependencies, the oldest Rust version supported is now 1.22.0

    #231

[0.9.2] 2018-01-13

Changes

__typename for unions

The __typename query meta-field now works on unions.

#112

Debug impls.

http::GraphQLRequest now implements Debug.

[0.9.0] 2017-12-03

Changes

This is the first release in a long time. Quite a few changes have accumulated since 0.8.1, including multiple breaking changes.

Custom derive & macros

Juniper has gained custom derive implementations for input objects, objects and enums.

  • #[derive(GraphQLInputObject)]
  • #[derive(GraphQLEnum)]
  • #[derive(GraphQLObject)]

The graphql_enum! and graphql_input_object! macros did not provide any more benefits, so they have been removed! All functionality is now covered by custom derive. Check the docs to find out more.

Web framework integrations - Iron & Rocket

The iron and rocket integrations were removed from the main crate, and are now available via the juniper_iron and juniper_rocket crates.

FieldError rewrite (custom data)

The FieldError type now supports custom data with the Value type from serde_json. Use this to populate the data field in returned errors.

This also means that try! and ? now work in resolvers, which is quite nice.

Also, the ResultExt extension and the jtry! macro were removed, since they are redundant now!

Dynamic Schemas

Juniper has gained support for dynamic schemas, thanks to @srijs.

That also means the type of RootNode has changed to include a lifetime.

The repository was restructured to a multi crate workspace to enable several new features like custom_derive and an extracted parser.

##66

Data Type Integrations

Integrations with multiple popular crates was added to make working with them easier.

  • uuid
  • url
  • chrono

Field Order

To better comply with the specification, order of requested fields is now preserved.

[#82](https://github.com/graphql-rust/juniper/issues/82

From/ToInputValue

The ::from and ::to methods in From/ToInputValue were renamed to from/to_input_value() to not conflict with other methods.

#90

Other changes

  • Several small performance improvements
  • Use fnv hash map for better performance

Contributors

A big shoutout to the many contributors for this version, sorted alphabetically.

[0.8.1] – 2017-06-15

Tiny release to fix broken crate metadata on crates.io.

0.8.0 – 2017-06-15

Breaking changes

  • To better comply with the specification, and to avoid weird bugs with very large positive or negative integers, support for i64 has been completely dropped and replaced with i32. i64 is no longer a valid GraphQL type in Juniper, and InputValue/Value can only represent 32 bit integers.

    If an incoming integer is out of range for a 32 bit signed integer type, an error will be returned to the client. (#52, #49)

  • Serde has been updated to 1.0. If your application depends on an older version, you will need to first update your application before you can upgrade to a more recent Juniper. (#43)

  • rustc_serialize support has been dropped since this library is now deprecated. (#51)

New features

  • A new rocket-handlers feature now includes some tools to use the Rocket framework. A simple example has been added to the examples folder.

Bugfixes

  • A panic in the parser has been replaced with a proper error (#44)

0.7.0 – 2017-02-26

Breaking changes

  • The iron-handlers feature now depends on Iron 0.5 (#30). Because of this, support for Rust 1.12 has been dropped. It might still work if you're not using the Iron integrations feature, however.

New features

  • Input objects defined by the graphql_input_object! can now be used as default values to field arguments and other input object fields.

0.6.3 – 2017-02-19

New features

  • Add support for default values on input object fields (#28)

0.6.2 – 2017-02-05

New features

  • The null literal is now supported in the GraphQL language. (#26)
  • Rustc-serialize is now optional, but enabled by default. If you only want Serde support, include Juniper without default features and enable Serde. (#12)
  • The built-in ID type now has a public constructor and derives a few traits (Clone, Debug, Eq, PartialEq, From<String>, Deref<Target=str>). (#19)
  • Juniper is now built and tested against all Rust compilers since version 1.12.1.

Minor breaking change

  • Serde has been updated to 0.9. (#25)

Bugfixes

  • The built-in GraphiQL handler had a bug in variable serialization. (#16)
  • The example should now build and run without problems on Windows. (#15)
  • Object types now properly implement __typename. (#22)
  • String variables are now properly parsed into GraphQL enums. (#17)

0.6.1 – 2017-01-06

New features

  • Optional Serde support (#8)

Improvements

  • The graphql_input_object! macro can now be used to define input objects as public Rust structs.
  • GraphiQL in the Iron GraphiQL handler has been updated to 0.8.1 (##11)

Minor breaking changes

Some undocumented but public APIs were changed.

  • to_snake_case correctly renamed to to_camel_case (#9)
  • JSON serialization of GraphQLError changed to be more consistent with how other values were serialized (#10).

0.6.0 – 2017-01-02

TL;DR: Many big changes in how context types work and how they interact with the executor. Not too much to worry about if you're only using the macros and not deriving GraphQLType directly.

Breaking changes

  • The executor argument in all resolver methods is now immutable. The executor instead uses interior mutability to store errors in a thread-safe manner.

    This change could open up for asynchronous or multi-threaded execution: you can today use something like rayon in your resolve methods to let child nodes be concurrently resolved.

    How to fix: All field resolvers that looked like field name(&mut executor now should say field name(&executor.

  • The context type of GraphQLType is moved to an associated type; meaning it's no longer generic. This only affects people who implement the trait manually, not macro users.

    This greatly simplifies a lot of code by ensuring that there only can be one GraphQLType implementation for any given Rust type. However, it has the downside that support for generic contexts previously used in scalars, has been removed. Instead, use the new context conversion features to accomplish the same task.

    How to fix: Instead of impl GraphQLType<MyContext> for ..., you use impl GraphQLType for ... { type Context = MyContext;.

  • All context types must derive the Context marker trait. This is part of an overarching change to allow different types to use different contexts.

    How to fix: If you have written e.g. graphql_object!(MyType: MyContext ...) you will need to add impl Context for MyContext {}. Simple as that.

  • Registry and all meta type structs now takes one lifetime parameter, which affects GraphQLType's meta method. This only affects people who implement the trait manually.

    How to fix: Change the type signature of meta() to read fn meta<'r>(registry: &mut Registry<'r>) -> MetaType<'r>.

  • The type builder methods on Registry no longer return functions taking types or fields. Due to how the borrow checker works with expressions, you will have to split up the instantiation into two statements. This only affects people who implement the GraphQLType trait manually.

    How to fix: Change the contents of your meta() methods to something like this:

    fn meta<'r>(registry: &mut Registry<r>) -> MetaType<'r> {
    let fields = &[ /* your fields ... */ ];

    registry.build_object_type::<Self>(fields).into_meta()
}

Added

  • Support for different contexts for different types. As GraphQL schemas tend to get large, narrowing down the context type to exactly what a given type needs is great for encapsulation. Similarly, letting different subsystems use different resources thorugh the context is also useful for the same reasons.

    Juniper supports two different methods of doing this, depending on your needs: if you have two contexts where one can be converted into the other without any extra knowledge, you can implement the new FromContext trait. This is useful if you have multiple crates or modules that all belong to the same GraphQL schema:

    struct TopContext {
  db: DatabaseConnection,
  session: WebSession,
  current_user: User,
}

struct ModuleOneContext {
  db: DatabaseConnection, // This module only requires a database connection
}

impl Context for TopContext {}
impl Context for ModuleOneContext {}

impl FromContext<TopContext> for ModuleOneContext {
  fn from(ctx: &TopContext) -> ModuleOneContext {
    ModuleOneContext {
      db: ctx.db.clone()
    }
  }
}

graphql_object!(Query: TopContext |&self| {
  field item(&executor) -> Item {
    executor.context().db.get_item()
  }
});

// The `Item` type uses another context type - conversion is automatic
graphql_object!(Item: ModuleOneContext |&self| {
  // ...
});

    The other way is to manually perform the conversion in a field resolver. This method is preferred when the child context needs extra knowledge than what exists in the parent context:

    // Each entity has its own context
struct TopContext {
  entities: HashMap<i32, EntityContext>,
  db: DatabaseConnection,
}

struct EntityContext {
  // fields
}

impl Context for TopContext {}
impl Context for EntityContext {}

graphql_object!(Query: TopContext |&self| {
  // By returning a tuple (&Context, GraphQLType), you can tell the executor
  // to switch out the context for the returned value. You can wrap the
  // tuple in Option<>, FieldResult<>, FieldResult<Option<>>, or just return
  // the tuple without wrapping it.
  field entity(&executor, key: i32) -> Option<(&EntityContext, Entity)> {
    executor.context().entities.get(&key)
      .map(|ctx| (ctx, executor.context().db.get_entity(key)))
  }
});

graphql_object!(Entity: EntityContext |&self| {
  // ...
});

Improvements

  • Parser and query execution has now reduced the allocation overhead by reusing as much as possible from the query source and meta type information.

0.5.3 – 2016-12-05

Added

  • jtry!: Helper macro to produce FieldResults from regular Results. Wherever you would be using try! in a regular function or method, you can use jtry! in a field resolver:

    graphql_object(MyType: Database |&self| {
  field count(&executor) -> FieldResult<i32> {
    let txn = jtry!(executor.context().transaction());

    let count = jtry!(txn.execute("SELECT COUNT(*) FROM user"));

    Ok(count[0][0])
  }
});

Changes

  • Relax context type trait requirements for the iron handler: your contexts no longer have to be Send + Sync.

  • RootNode is now Send and Sync if both the mutation and query types implement Send and Sync.

Bugfixes

  • return statements inside field resolvers no longer cause syntax errors.

0.5.2 – 2016-11-13

Added

  • Support for marking fields and enum values deprecated.
  • input_object! helper macro

Changes

  • The included example server now uses the simple Star Wars schema used in query/introspection tests.

Bugfixes

  • The query validators - particularly ones concerned with validation of input data and variables - have been improved significantly. A large number of test cases have been added.

  • Macro syntax stability has also been improved. All syntactical edge cases of the macros have gotten tests to verify their correctness.