async-graphql/src/schema.rs

567 lines
18 KiB
Rust
Raw Normal View History

2020-09-10 04:49:08 +00:00
use crate::context::{Data, ResolveId};
use crate::extensions::{BoxExtension, ErrorLogger, Extension, Extensions};
use crate::model::__DirectiveLocation;
use crate::parser::parse_query;
use crate::parser::types::{
DocumentOperations, FragmentDefinition, Name, OperationDefinition, OperationType,
};
use crate::registry::{MetaDirective, MetaInputValue, Registry};
2020-09-12 09:29:52 +00:00
use crate::resolver_utils::{resolve_object, resolve_object_serial, ObjectType};
use crate::subscription::collect_subscription_streams;
2020-03-03 11:15:18 +00:00
use crate::types::QueryRoot;
2020-05-22 03:58:49 +00:00
use crate::validation::{check_rules, CheckResult, ValidationMode};
2020-03-29 12:02:52 +00:00
use crate::{
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
BatchRequest, BatchResponse, CacheControl, ContextBase, Positioned, QueryEnv, Request,
Response, ServerError, SubscriptionType, Type, Variables, ID,
2020-03-29 12:02:52 +00:00
};
use futures::stream::{self, Stream, StreamExt};
use indexmap::map::IndexMap;
2020-05-22 03:58:49 +00:00
use itertools::Itertools;
use std::any::Any;
use std::collections::HashMap;
use std::ops::Deref;
2020-03-29 12:02:52 +00:00
use std::sync::atomic::AtomicUsize;
use std::sync::Arc;
2020-03-01 10:54:34 +00:00
2020-03-29 12:02:52 +00:00
/// Schema builder
pub struct SchemaBuilder<Query, Mutation, Subscription> {
validation_mode: ValidationMode,
query: QueryRoot<Query>,
mutation: Mutation,
subscription: Subscription,
registry: Registry,
data: Data,
complexity: Option<usize>,
depth: Option<usize>,
extensions: Vec<Box<dyn Fn() -> BoxExtension + Send + Sync>>,
enable_federation: bool,
}
2020-03-29 12:02:52 +00:00
2020-03-19 09:20:12 +00:00
impl<Query: ObjectType, Mutation: ObjectType, Subscription: SubscriptionType>
2020-03-29 12:02:52 +00:00
SchemaBuilder<Query, Mutation, Subscription>
2020-03-17 09:26:59 +00:00
{
2020-09-06 05:38:31 +00:00
/// Manually register a type in the schema.
///
/// You can use this function to register schema types that are not directly referenced.
pub fn register_type<T: Type>(mut self) -> Self {
T::create_type_info(&mut self.registry);
self
}
2020-09-06 05:38:31 +00:00
/// Disable introspection queries.
2020-03-29 12:02:52 +00:00
pub fn disable_introspection(mut self) -> Self {
self.query.disable_introspection = true;
2020-03-29 12:02:52 +00:00
self
}
2020-09-06 05:38:31 +00:00
/// Set the maximum complexity a query can have. By default there is no limit.
2020-03-29 12:02:52 +00:00
pub fn limit_complexity(mut self, complexity: usize) -> Self {
self.complexity = Some(complexity);
2020-03-29 12:02:52 +00:00
self
}
2020-09-06 05:38:31 +00:00
/// Set the maximum depth a query can have. By default there is no limit.
2020-03-29 12:02:52 +00:00
pub fn limit_depth(mut self, depth: usize) -> Self {
self.depth = Some(depth);
2020-03-29 12:02:52 +00:00
self
}
2020-09-06 05:38:31 +00:00
/// Add an extension to the schema.
2020-03-29 12:02:52 +00:00
pub fn extension<F: Fn() -> E + Send + Sync + 'static, E: Extension>(
mut self,
extension_factory: F,
) -> Self {
self.extensions
2020-03-29 12:02:52 +00:00
.push(Box::new(move || Box::new(extension_factory())));
self
}
2020-09-06 05:38:31 +00:00
/// Add a global data that can be accessed in the `Schema`. You access it with `Context::data`.
2020-03-29 12:02:52 +00:00
pub fn data<D: Any + Send + Sync>(mut self, data: D) -> Self {
self.data.insert(data);
2020-03-29 12:02:52 +00:00
self
}
2020-04-06 11:57:21 +00:00
/// Set the validation mode, default is `ValidationMode::Strict`.
pub fn validation_mode(mut self, validation_mode: ValidationMode) -> Self {
self.validation_mode = validation_mode;
2020-04-06 11:57:21 +00:00
self
}
/// Enable federation, which is automatically enabled if the Query has least one entity definition.
pub fn enable_federation(mut self) -> Self {
self.enable_federation = true;
self
}
2020-03-29 12:02:52 +00:00
/// Build schema.
pub fn finish(mut self) -> Schema<Query, Mutation, Subscription> {
// federation
if self.enable_federation || self.registry.has_entities() {
self.registry.create_federation_types();
}
Schema(Arc::new(SchemaInner {
validation_mode: self.validation_mode,
query: self.query,
mutation: self.mutation,
subscription: self.subscription,
complexity: self.complexity,
depth: self.depth,
extensions: self.extensions,
env: SchemaEnv(Arc::new(SchemaEnvInner {
registry: self.registry,
data: self.data,
})),
}))
2020-03-29 12:02:52 +00:00
}
}
#[doc(hidden)]
pub struct SchemaEnvInner {
pub registry: Registry,
pub data: Data,
}
#[doc(hidden)]
#[derive(Clone)]
pub struct SchemaEnv(Arc<SchemaEnvInner>);
impl Deref for SchemaEnv {
type Target = SchemaEnvInner;
fn deref(&self) -> &Self::Target {
&self.0
}
}
#[doc(hidden)]
pub struct SchemaInner<Query, Mutation, Subscription> {
pub(crate) validation_mode: ValidationMode,
pub(crate) query: QueryRoot<Query>,
pub(crate) mutation: Mutation,
pub(crate) subscription: Subscription,
pub(crate) complexity: Option<usize>,
pub(crate) depth: Option<usize>,
pub(crate) extensions: Vec<Box<dyn Fn() -> BoxExtension + Send + Sync>>,
pub(crate) env: SchemaEnv,
}
2020-09-06 05:38:31 +00:00
/// GraphQL schema.
///
/// Cloning a schema is cheap, so it can be easily shared.
pub struct Schema<Query, Mutation, Subscription>(Arc<SchemaInner<Query, Mutation, Subscription>>);
2020-03-29 12:02:52 +00:00
impl<Query, Mutation, Subscription> Clone for Schema<Query, Mutation, Subscription> {
fn clone(&self) -> Self {
Schema(self.0.clone())
}
}
2020-08-28 06:19:35 +00:00
impl<Query, Mutation, Subscription> Default for Schema<Query, Mutation, Subscription>
where
Query: Default + ObjectType + Send + Sync + 'static,
Mutation: Default + ObjectType + Send + Sync + 'static,
Subscription: Default + SubscriptionType + Send + Sync + 'static,
{
fn default() -> Self {
Schema::new(
Query::default(),
Mutation::default(),
Subscription::default(),
)
}
}
2020-09-12 09:29:52 +00:00
impl<Query, Mutation, Subscription> Deref for Schema<Query, Mutation, Subscription> {
type Target = SchemaInner<Query, Mutation, Subscription>;
fn deref(&self) -> &Self::Target {
&self.0
}
}
2020-03-29 12:02:52 +00:00
impl<Query, Mutation, Subscription> Schema<Query, Mutation, Subscription>
where
Query: ObjectType + Send + Sync + 'static,
Mutation: ObjectType + Send + Sync + 'static,
Subscription: SubscriptionType + Send + Sync + 'static,
{
/// Create a schema builder
2020-03-09 10:05:52 +00:00
///
/// The root object for the query and Mutation needs to be specified.
2020-03-19 09:20:12 +00:00
/// If there is no mutation, you can use `EmptyMutation`.
/// If there is no subscription, you can use `EmptySubscription`.
2020-03-29 12:02:52 +00:00
pub fn build(
query: Query,
mutation: Mutation,
subscription: Subscription,
) -> SchemaBuilder<Query, Mutation, Subscription> {
2020-09-23 00:04:00 +00:00
SchemaBuilder {
validation_mode: ValidationMode::Strict,
query: QueryRoot {
inner: query,
disable_introspection: false,
},
mutation,
subscription,
registry: Self::create_registry(),
data: Default::default(),
complexity: None,
depth: None,
extensions: Default::default(),
enable_federation: false,
}
}
fn create_registry() -> Registry {
2020-03-08 12:35:36 +00:00
let mut registry = Registry {
types: Default::default(),
directives: Default::default(),
implements: Default::default(),
2020-09-23 00:04:00 +00:00
is_empty_query: Query::is_empty(),
2020-03-08 12:35:36 +00:00
query_type: Query::type_name().to_string(),
mutation_type: if Mutation::is_empty() {
None
} else {
Some(Mutation::type_name().to_string())
},
2020-03-17 09:26:59 +00:00
subscription_type: if Subscription::is_empty() {
None
} else {
Some(Subscription::type_name().to_string())
},
2020-03-08 12:35:36 +00:00
};
registry.add_directive(MetaDirective {
name: "include",
description: Some("Directs the executor to include this field or fragment only when the `if` argument is true."),
2020-03-05 13:34:31 +00:00
locations: vec![
__DirectiveLocation::FIELD,
__DirectiveLocation::FRAGMENT_SPREAD,
__DirectiveLocation::INLINE_FRAGMENT
],
2020-03-08 12:35:36 +00:00
args: {
let mut args = IndexMap::new();
args.insert("if", MetaInputValue {
2020-03-08 12:35:36 +00:00
name: "if",
description: Some("Included when true."),
ty: "Boolean!".to_string(),
2020-03-21 01:32:13 +00:00
default_value: None,
2020-03-22 01:34:32 +00:00
validator: None,
2020-03-08 12:35:36 +00:00
});
args
}
});
registry.add_directive(MetaDirective {
name: "skip",
description: Some("Directs the executor to skip this field or fragment when the `if` argument is true."),
2020-03-05 13:34:31 +00:00
locations: vec![
__DirectiveLocation::FIELD,
__DirectiveLocation::FRAGMENT_SPREAD,
__DirectiveLocation::INLINE_FRAGMENT
],
2020-03-08 12:35:36 +00:00
args: {
let mut args = IndexMap::new();
args.insert("if", MetaInputValue {
2020-03-08 12:35:36 +00:00
name: "if",
description: Some("Skipped when true."),
ty: "Boolean!".to_string(),
2020-03-21 01:32:13 +00:00
default_value: None,
2020-03-22 01:34:32 +00:00
validator: None,
2020-03-08 12:35:36 +00:00
});
args
}
});
2020-08-06 06:52:54 +00:00
registry.add_directive(MetaDirective {
name: "ifdef",
description: Some("Directs the executor to query only when the field exists."),
locations: vec![__DirectiveLocation::FIELD],
args: Default::default(),
});
// register scalars
bool::create_type_info(&mut registry);
i32::create_type_info(&mut registry);
f32::create_type_info(&mut registry);
String::create_type_info(&mut registry);
ID::create_type_info(&mut registry);
2020-03-08 12:35:36 +00:00
QueryRoot::<Query>::create_type_info(&mut registry);
2020-03-05 09:06:14 +00:00
if !Mutation::is_empty() {
Mutation::create_type_info(&mut registry);
}
2020-03-17 09:26:59 +00:00
if !Subscription::is_empty() {
Subscription::create_type_info(&mut registry);
}
2020-09-23 00:04:00 +00:00
registry
2020-03-25 07:07:16 +00:00
}
2020-03-29 12:02:52 +00:00
/// Create a schema
pub fn new(
query: Query,
mutation: Mutation,
subscription: Subscription,
) -> Schema<Query, Mutation, Subscription> {
Self::build(query, mutation, subscription).finish()
2020-03-05 00:39:56 +00:00
}
2020-09-23 00:04:00 +00:00
/// Returns SDL(Schema Definition Language) of this schema.
pub fn sdl() -> String {
Self::create_registry().export_sdl(false)
}
// TODO: Remove the allow
#[allow(clippy::type_complexity)]
2020-09-11 07:54:56 +00:00
fn prepare_request(
2020-05-22 03:58:49 +00:00
&self,
2020-09-11 07:54:56 +00:00
request: &Request,
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
) -> Result<
(
Positioned<OperationDefinition>,
HashMap<Name, Positioned<FragmentDefinition>>,
CacheControl,
spin::Mutex<Extensions>,
),
Vec<ServerError>,
> {
2020-05-22 03:58:49 +00:00
// create extension instances
let extensions = spin::Mutex::new(Extensions(
self.0
.extensions
.iter()
2020-09-21 07:53:07 +00:00
.chain(request.extensions.iter())
.map(|factory| factory())
.collect_vec(),
));
2020-05-22 03:58:49 +00:00
2020-09-11 07:54:56 +00:00
extensions
.lock()
.parse_start(&request.query, &request.variables);
let document = parse_query(&request.query)
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
.map_err(Into::<ServerError>::into)
.log_error(&extensions)?;
extensions.lock().parse_end(&document);
2020-05-22 03:58:49 +00:00
// check rules
extensions.lock().validation_start();
2020-05-22 03:58:49 +00:00
let CheckResult {
cache_control,
complexity,
depth,
} = check_rules(
&self.env.registry,
&document,
2020-09-11 07:54:56 +00:00
Some(&request.variables),
self.validation_mode,
)
.log_error(&extensions)?;
extensions.lock().validation_end();
2020-05-22 03:58:49 +00:00
// check limit
if let Some(limit_complexity) = self.complexity {
if complexity > limit_complexity {
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
return Err(vec![ServerError::new("Query is too complex.")]).log_error(&extensions);
2020-05-22 03:58:49 +00:00
}
}
if let Some(limit_depth) = self.depth {
if depth > limit_depth {
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
return Err(vec![ServerError::new("Query is nested too deep.")])
.log_error(&extensions);
2020-05-22 03:58:49 +00:00
}
}
let operation = if let Some(operation_name) = &request.operation_name {
match document.operations {
DocumentOperations::Single(_) => None,
DocumentOperations::Multiple(mut operations) => {
operations.remove(operation_name.as_str())
}
}
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
.ok_or_else(|| {
ServerError::new(format!(r#"Unknown operation named "{}""#, operation_name))
})
} else {
match document.operations {
DocumentOperations::Single(operation) => Ok(operation),
DocumentOperations::Multiple(map) if map.len() == 1 => {
Ok(map.into_iter().next().unwrap().1)
}
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
DocumentOperations::Multiple(_) => {
Err(ServerError::new("Operation name required in request."))
}
}
};
let operation = match operation {
Ok(operation) => operation,
Err(e) => {
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
extensions.lock().error(&e);
return Err(vec![e]);
2020-09-10 04:49:08 +00:00
}
};
Ok((operation, document.fragments, cache_control, extensions))
2020-09-11 07:54:56 +00:00
}
async fn execute_once(
&self,
operation: Positioned<OperationDefinition>,
fragments: HashMap<Name, Positioned<FragmentDefinition>>,
2020-09-11 07:54:56 +00:00
extensions: spin::Mutex<Extensions>,
variables: Variables,
ctx_data: Data,
) -> Response {
// execute
let inc_resolve_id = AtomicUsize::default();
let env = QueryEnv::new(
extensions,
variables,
operation,
fragments,
Arc::new(ctx_data),
);
2020-09-10 04:49:08 +00:00
let ctx = ContextBase {
path_node: None,
resolve_id: ResolveId::root(),
inc_resolve_id: &inc_resolve_id,
item: &env.operation.node.selection_set,
2020-09-10 08:39:43 +00:00
schema_env: &self.env,
2020-09-10 04:49:08 +00:00
query_env: &env,
};
env.extensions.lock().execution_start();
let data = match &env.operation.node.ty {
OperationType::Query => resolve_object(&ctx, &self.query).await,
2020-09-14 01:46:22 +00:00
OperationType::Mutation => resolve_object_serial(&ctx, &self.mutation).await,
2020-09-10 04:49:08 +00:00
OperationType::Subscription => {
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
return Response::from_errors(vec![ServerError::new(
"Subscriptions are not supported on this transport.",
)])
2020-09-10 04:49:08 +00:00
}
};
env.extensions.lock().execution_end();
2020-09-10 08:39:43 +00:00
let extensions = env.extensions.lock().result();
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
match data {
Ok(data) => Response::new(data),
Err(e) => Response::from_errors(vec![e]),
}
.extensions(extensions)
2020-09-10 04:49:08 +00:00
}
2020-09-11 07:54:56 +00:00
/// Execute an GraphQL query.
pub async fn execute(&self, request: impl Into<Request>) -> Response {
let request = request.into();
match self.prepare_request(&request) {
Ok((operation, fragments, cache_control, extensions)) => self
.execute_once(
operation,
fragments,
extensions,
request.variables,
request.data,
)
.await
.cache_control(cache_control),
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
Err(errors) => Response::from_errors(errors),
}
2020-09-11 07:54:56 +00:00
}
2020-09-17 08:39:55 +00:00
/// Execute an GraphQL batch query.
pub async fn execute_batch(&self, batch_request: BatchRequest) -> BatchResponse {
match batch_request {
BatchRequest::Single(request) => BatchResponse::Single(self.execute(request).await),
BatchRequest::Batch(requests) => BatchResponse::Batch(
futures::stream::iter(requests.into_iter())
.then(|request| self.execute(request))
.collect()
.await,
),
}
}
2020-09-11 08:41:56 +00:00
pub(crate) fn execute_stream_with_ctx_data(
&self,
2020-09-17 18:22:54 +00:00
request: impl Into<Request> + Send,
2020-09-11 08:41:56 +00:00
ctx_data: Arc<Data>,
2020-09-17 18:22:54 +00:00
) -> impl Stream<Item = Response> + Send {
let schema = self.clone();
2020-09-11 07:54:56 +00:00
async_stream::stream! {
let request = request.into();
let (operation, fragments, cache_control, extensions) = match schema.prepare_request(&request) {
2020-09-11 07:54:56 +00:00
Ok(res) => res,
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
Err(errors) => {
yield Response::from_errors(errors);
return;
}
};
if operation.node.ty != OperationType::Subscription {
yield schema
.execute_once(operation, fragments, extensions, request.variables, request.data)
.await
.cache_control(cache_control);
return;
}
2020-03-29 12:02:52 +00:00
let resolve_id = AtomicUsize::default();
let env = QueryEnv::new(
extensions,
request.variables,
operation,
fragments,
2020-09-11 08:41:56 +00:00
ctx_data,
);
let ctx = env.create_context(
&schema.env,
None,
&env.operation.node.selection_set,
&resolve_id,
);
// TODO: Invoke extensions
let mut streams = Vec::new();
if let Err(e) = collect_subscription_streams(&ctx, &schema.subscription, &mut streams) {
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
yield Response::from_errors(vec![e]);
return;
}
2020-05-18 16:03:15 +00:00
let mut stream = stream::select_all(streams);
while let Some(data) = stream.next().await {
let is_err = data.is_err();
let extensions = env.extensions.lock().result();
Rework errors This completely overhauls the error system used in async-graphql. - `Error` has been renamed to `ServerError` and `FieldError` has been renamed to just `Error`. This is because `FieldError` is by far the most common error that users will have to use so it makes sense to use the most obvious error name. Also, the current name didn't make sense as it was used for things other than field errors, such as the data callback for websockets. - `ServerError` has been made completely opaque. Before it was an enum of all the possible errors, but now it just contains an error message, the locations, the path and extensions. It is a shame that we lose information, it makes more sense as _conceptually_ GraphQL does not provide that information. It also frees us to change the internals of async-graphql a lot more. - The path of errors is no longer an opaque JSON value but a regular type, `Vec<PathSegment>`. The type duplication of `PathSegment` and `QueryPathSegment` is unfortunate, I plan to work on this in the future. - Now that `ServerError` is opaque, `RuleError` has been removed from the public API, making it simpler. - Additionally `QueryError` has been completely removed. Instead the error messages are constructed ad-hoc; I took care to never repeat an error message. - Instead of constructing field-not-found errors inside the implementations of field resolvers they now return `Option`s, where a `None` value is representative of the field not being found. - As an unfortunate consequence of the last change, self-referential types based on the output of a subscription resolver can no longer be created. This does not mean anything for users, but causes lifetime issues in the implementation of merged objects. I fixed it with a bit of a hack, but this'll have to be looked into further. - `InputValueError` now has a generic parameter - it's kind of weird but it's necessary for ergonomics. It also improves error messages. - The `ErrorExtensions` trait has been removed. I didn't think the `extend` method was necessary since `From` impls exist. But the ergonomics are still there with a new trait `ExtendError`, which is implemented for both errors and results. - `Response` now supports serializing multiple errors. This allows for nice things like having multiple validation errors not be awkwardly shoved into a single error. - When an error occurs in execution, data is sent as `null`. This is slightly more compliant with the spec but the algorithm described in <https://spec.graphql.org/June2018/#sec-Errors-and-Non-Nullability> has yet to be implemented.
2020-09-29 19:06:44 +00:00
yield match data {
Ok((name, value)) => Response::new(
serde_json::json!({
name: value,
})
),
Err(e) => Response::from_errors(vec![e]),
}.extensions(extensions);
if is_err {
break;
}
2020-09-10 08:39:43 +00:00
}
2020-09-11 07:54:56 +00:00
}
2020-03-29 12:02:52 +00:00
}
2020-09-11 08:41:56 +00:00
/// Execute an GraphQL subscription.
2020-09-17 18:22:54 +00:00
pub fn execute_stream(
&self,
request: impl Into<Request>,
) -> impl Stream<Item = Response> + Send {
2020-09-11 08:41:56 +00:00
let mut request = request.into();
2020-09-12 16:07:46 +00:00
let ctx_data = std::mem::take(&mut request.data);
2020-09-11 08:41:56 +00:00
self.execute_stream_with_ctx_data(request, Arc::new(ctx_data))
}
2020-03-01 10:54:34 +00:00
}