graphql/client.dart is a GraphQL client for dart modeled on the apollo client, and is currently the most popular GraphQL client for dart. It is co-developed alongside graphql_flutter on github, where you can find more in-depth examples. We also have a lively community alongside the rest of the GraphQL Dart community on discord.
As of v4, it is built on foundational libraries from the gql-dart project, including gql, gql_link, and normalize. We also depend on hive for persistence via HiveStore.
- GraphQL Client
Useful API Docs:
GraphQLCacheGraphQLDataProxyAPI docs (direct cache access)
First, depend on this package:
$ flutter pub add graphqlAnd then import it inside your dart code:
import 'package:graphql/client.dart';Find the migration from version 3 to version 4 here.
To connect to a GraphQL Server, we first need to create a GraphQLClient. A GraphQLClient requires both a cache and a link to be initialized.
In our example below, we will be using the Github Public API. we are going to use HttpLink which we will concatenate with AuthLink so as to attach our github access token.
For the cache, we are going to use GraphQLCache.
// ...
final _httpLink = HttpLink(
'https://api.github.com/graphql',
);
final _authLink = AuthLink(
getToken: () async => 'Bearer $YOUR_PERSONAL_ACCESS_TOKEN',
);
Link _link = _authLink.concat(_httpLink);
/// subscriptions must be split otherwise `HttpLink` will. swallow them
if (websocketEndpoint != null){
final _wsLink = WebSocketLink(websockeEndpoint);
_link = Link.split((request) => request.isSubscription, _wsLink, _link);
}
final GraphQLClient client = GraphQLClient(
/// **NOTE** The default store is the InMemoryStore, which does NOT persist to disk
cache: GraphQLCache(),
link: _link,
);
// ...In v4, GraphQLCache is decoupled from persistence, which is managed (or not) by its store argument.
We provide a HiveStore for easily using hive boxes as storage,
which requires a few changes to the above:
NB: This is different in
graphql_flutter, which providesawait initHiveForFlutter()for initialization inmain
GraphQLClient getClient() async {
...
/// initialize Hive and wrap the default box in a HiveStore
final store = await HiveStore.open(path: 'my/cache/path');
return GraphQLClient(
/// pass the store to the cache for persistence
cache: GraphQLCache(store: store),
link: _link,
);
}Once you have initialized a client, you can run queries and mutations.
All graphql methods accept a corresponding *Options object for configuring behavior. These options all include policies with which to override defaults, an optimisticResult for snappy client-side interactions, gql_exec Context with which to make requests, and of course a document to be requested.
Internally they are converted to gql_exec Requests with .asRequest for execution via links, and thus can also be used with the direct cache access api.
Creating a query is as simple as creating a multiline string:
const String readRepositories = r'''
query ReadRepositories($nRepositories: Int!) {
viewer {
repositories(last: $nRepositories) {
nodes {
__typename
id
name
viewerHasStarred
}
}
}
}
''';Then create a QueryOptions object:
NB: for
document- Use our built-in help function -gql(query)to convert your document string to ASTsdocument.
In our case, we need to pass nRepositories variable and the document name is readRepositories.
const int nRepositories = 50;
final QueryOptions options = QueryOptions(
document: gql(readRepositories),
variables: <String, dynamic>{
'nRepositories': nRepositories,
},
);
And finally you can send the query to the server and await the response:
// ...
final QueryResult result = await client.query(options);
if (result.hasException) {
print(result.exception.toString());
}
final List<dynamic> repositories =
result.data['viewer']['repositories']['nodes'] as List<dynamic>;
// ...Creating a mutation is similar to creating a query, with a small difference. First, start with a multiline string:
const String addStar = r'''
mutation AddStar($starrableId: ID!) {
action: addStar(input: {starrableId: $starrableId}) {
starrable {
viewerHasStarred
}
}
}
''';Then instead of the QueryOptions, for mutations we will MutationOptions, which is where we pass our mutation and id of the repository we are starring.
// ...
final MutationOptions options = MutationOptions(
document: gql(addStar),
variables: <String, dynamic>{
'starrableId': repositoryID,
},
);
// ...And finally you can send the mutation to the server and await the response:
// ...
final QueryResult result = await client.mutate(options);
if (result.hasException) {
print(result.exception.toString());
return;
}
final bool isStarred =
result.data['action']['starrable']['viewerHasStarred'] as bool;
if (isStarred) {
print('Thanks for your star!');
return;
}
// ...gql_http_link provides support for the GraphQL Upload spec as proposed at https://github.com/jaydenseric/graphql-multipart-request-spec
mutation($files: [Upload!]!) {
multipleUpload(files: $files) {
id
filename
mimetype
path
}
}import "package:http/http.dart" show Multipartfile;
import "package:http_parser/http_parser.dart" show MediaType;
// ...
final myFile = MultipartFile.fromString(
"",
"just plain text",
filename: "sample_upload.txt",
contentType: MediaType("text", "plain"),
);
final result = await graphQLClient.mutate(
MutationOptions(
document: gql(uploadMutation),
variables: {
'files': [myFile],
},
)
);To use subscriptions, a subscription-consuming link must be split from your HttpLink or other terminating link route:
link = Link.split((request) => request.isSubscription, websocketLink, link);Then you can subscribe to any subscriptions provided by your server schema:
final subscriptionDocument = gql(
r'''
subscription reviewAdded {
reviewAdded {
stars, commentary, episode
}
}
''',
);
// graphql/client.dart usage
subscription = client.subscribe(
SubscriptionOptions(
document: subscriptionDocument
),
);
subscription.listen(reactToAddedReview)In order to add auth header or any other header to websocket connection use initialPayload property
initialPayload: () {
var headers = <String, String>{};
headers.putIfAbsent(HttpHeaders.authorizationHeader, () => token);
return headers;
},In order to refresh auth header you need to setup onConnectionLost function
onConnectionLost: (int? code, String? reason) async {
if (code == 4001) {
await authTokenService.issueToken(refresh: true);
return Duration.zero;
}
return null;
}Where code and reason are values returned from the server on connection close. There is no such code like 401 in WebSockets so you can use your custom and server code could look similar:
subscriptions: {
'graphql-ws': {
onConnect: async (context: any) => {
const { connectionParams } = context;
if (!connectionParams) {
throw new Error('Connection params are missing');
}
const authToken = connectionParams.authorization;
if (authToken) {
const isValid await authService.isTokenValid(authToken);
if (!isValid) {
context.extra.socket.close(4001, 'Unauthorized');
}
return;
}
},
},
},onConnectionLost function returns Duration which is basically delayBetweenReconnectionAttempts for current reconnect attempt. If duration is null then default delayBetweenReconnectionAttempts will be used. Otherwise returned value. For example upon expired auth token there is not much sense to wait after token is refreshed.
toggleConnection stream was introduced to allow connect or disconnect manually.
var toggleConnection = PublishSubject<ToggleConnectionState>;
SocketClientConfig(
toggleConnection: toggleConnection,
),later from your code call
toggleConnection.add(ToggleConnectionState.disconnect);
//OR
toggleConnection.add(ToggleConnectionState.connect);When disconnect event is called autoReconnect stops. When connect is called autoReconnect resumes.
this is useful when for some reason you want to stop reconnection. For example when user logouts from the system and reconnection would cause auth error from server causing infinite loop.
WebSocketLink now has an experimental connect parameter that can be
used to supply custom headers to an IO client, register custom listeners,
and extract the socket for other non-graphql features.
Warning: if you want to listen to the stream,
wrap your channel with our GraphQLWebSocketChannel using the .forGraphQL() helper:
connect: (url, protocols) {
var channel = WebSocketChannel.connect(url, protocols: protocols)
// without this line, our client won't be able to listen to stream events,
// because you are already listening.
channel = channel.forGraphQL();
channel.stream.listen(myListener)
return channel;
}To supply custom headers (not supported on Flutter Web):
SocketClient(
wsUrl,
config: SocketClientConfig(
autoReconnect: autoReconnect,
headers: customHeaders,
delayBetweenReconnectionAttempts: delayBetweenReconnectionAttempts,
),
);In some cases, you want to update your web socket connection (as described in #727), in order to implement this functionality you need to do the following steps:
1- create your custom websocket link
class SocketCustomLink extends Link {
SocketCustomLink(this.url);
final String url;
_Connection? _connection;
/// this will be called every time you make a subscription
@override
Stream<Response> request(Request request, [forward]) async* {
/// first get the token by your own way
String? token = "...";
/// check is connection is null or the token changed
if (_connection == null || _connection!.token != token) {
connectOrReconnect(token);
}
yield* _connection!.client.subscribe(request, true);
}
/// Connects or reconnects to the server with the specified headers.
void connectOrReconnect(String? token) {
_connection?.client.dispose();
var _url = Uri.parse(url);
if (kIsWeb) {
/// web don't support headers in sockets so you may need to pass it as query string
/// server must support that
_url = _url.replace(queryParameters: {"access_token": token});
}
_connection = _Connection(
client: SocketClient(
_url.toString(),
config: SocketClientConfig(
autoReconnect: true,
headers: kIsWeb || token == null
? null
: {"Authorization": " Bearer $token"},
),
),
token: token,
);
}
@override
Future<void> dispose() async {
await _connection?.client.dispose();
_connection = null;
}
}
/// this a wrapper for web socket to hold the used token
class _Connection {
SocketClient client;
String? token;
_Connection({
required this.client,
required this.token,
});
}
2- if you need to update your socket just cancel your subscription and resubscribe again using usual way and if the token changed it will be reconnect with the new token otherwise it will use the same client
client.watchQuery
can be used to execute both queries and mutations, then reactively listen to changes to the underlying data in the cache.
final observableQuery = client.watchQuery(
WatchQueryOptions(
fetchResults: true,
document: gql(
r'''
query HeroForEpisode($ep: Episode!) {
hero(episode: $ep) {
name
}
}
''',
),
variables: {'ep': 'NEWHOPE'},
),
);
/// Listen to the stream of results. This will include:
/// * `options.optimisitcResult` if passed
/// * The result from the server (if `options.fetchPolicy` includes networking)
/// * rebroadcast results from edits to the cache
observableQuery.stream.listen((QueryResult result) {
if (!result.isLoading && result.data != null) {
if (result.hasException) {
print(result.exception);
return;
}
if (result.isLoading) {
print('loading');
return;
}
doSomethingWithMyQueryResult(myCustomParser(result.data));
}
});
// ... cleanup:
observableQuery.close();ObservableQuery is a bit of a kitchen sink for reactive operation logic – consider looking at the API docs if you'd like to develop a deeper understanding.
The default CacheRereadPolicy of client.watchQuery merges optimistic data from the cache into the result on every related cache change. This is great for queries, but an undesirable default for mutations, as their results should not change due to subsequent mutations.
While eventually we would like to decouple mutation and query logic, for now we have client.watchMutation (used in the Mutation widget of graphql_flutter) which has the default policy CacheRereadPolicy.ignoreAll. Otherwise, its behavior is exactly the same. It still takes WatchQueryOptions and returns ObservableQuery, and both methods can take either mutation or query documents. The watchMutation method should be thought of as a stop-gap.
See Rebroadcasting for more details.
NB:
watchQuery,watchMutation, andObservableQuerycurrently don't have a nice APIs forupdateonCompletedandonErrorcallbacks, but you can have a look at howgraphql_flutterregisters them throughonDatainMutation.runMutation.
The GraphQLCache automatically normalizes data from the server, and heavily leverages the normalize library. Data IDs are pulled from each selection set and used as keys in the cache.
The default approach is roughly:
String dataIdFromObject(Map<String, Object> data) {
final typename = data['__typename'];
if (typename == null) return null;
final id = data['id'] ?? data['_id'];
return id == null ? null : '$typename:$id';
}To disable cache normalization entirely, you could pass (data) => null.
If you only cared about nodeId, you could pass (data) => data['nodeId'].
Here's a more detailed example where the system involved contains versioned entities you don't want to clobber:
String customDataIdFromObject(Map<String, Object> data) {
final typeName = data['__typename'];
final entityId = data['entityId'];
final version = data['version'];
if (typeName == null || entityId == null || version == null){
return null;
}
return '${typeName}/${entityId}/${version}';
}Normalize requires you to specify the possible types map for fragments to work correctly. This is a mapping from abstract union and interface types to their concrete object types. E.g. take the schema
interface PersonI {
name: String
age: Int
}
type Employee implements PersonI {
name: String
age: Int
daysOfEmployement: Int
}
type InStoreCustomer implements PersonI {
name: String
age: Int
numberOfPurchases: Int
}
type OnlineCustomer implements PersonI {
name: String
age: Int
numberOfPurchases: Int
}
union CustomerU = OnlineCustomer | InStoreCustomer
the possible types map would be:
const POSSIBLE_TYPES = const {
'CustomerU': {'InStoreCustomer', 'OnlineCustomer'},
'PersonI': {'Employee', 'InStoreCustomer', 'OnlineCustomer'},
}
// Here's how it's parsed to the cache
final client = GraphQLClient(
cache: GraphQLCache(
possibleTypes: POSSIBLE_TYPES,
),
)You can generate the POSSIBLE_TYPES map, e.g., using graphql_codegen.
Furthermore, for normalize to correctly resolve the type you should always make sure you're querying the __typename. Given the example above a query could look something like
query {
people {
__typename # Needed to decide where which entry to update in the cache
... on Employee {
name
age
}
... on Customer {
name
age
}
}
}
if you're not providing the possible type map and introspecting the typename, the cache can't be updated.
The GraphQLCache
leverages normalize to give us a fairly apollo-ish direct cache access API, which is also available on GraphQLClient.
This means we can do local state management in a similar fashion as well.
The cache access methods are available on any cache proxy, which includes the GraphQLCache the OptimisticProxy passed to update in the graphql_flutter Mutation widget, and the client itself.
NB counter-intuitively, you likely never want to use use direct cache access methods directly on the
cache, as they will not be rebroadcast automatically. Preferclient.writeQueryandclient.writeFragmentto those on theclient.cachefor automatic rebroadcasting
In addition to this overview, a complete and well-commented rundown of can be found in the
GraphQLDataProxy API docs.
The query-based direct cache access methods readQuery and writeQuery leverage gql_exec Requests used internally in the link system. These can be retrieved from options.asRequest available on all *Options objects, or constructed manually:
const int nRepositories = 50;
final QueryOptions options = QueryOptions(
document: gql(readRepositories),
variables: {
'nRepositories': nRepositories,
},
);
var queryRequest = Request(
operation: Operation(
document: gql(readRepositories),
),
variables: {
'nRepositories': nRepositories,
},
);
/// experimental convenience api
queryRequest = Operation(document: gql(readRepositories)).asRequest(
variables: {
'nRepositories': nRepositories,
},
);
print(queryRequest == options.asRequest);
final data = client.readQuery(queryRequest);
client.writeQuery(queryRequest, data);The cache access methods are available on any cache proxy, which includes the GraphQLCache the OptimisticProxy passed to update in the graphql_flutter Mutation widget, and the client itself.
NB counter-intuitively, you likely never want to use use direct cache access methods on the cache cache.readQuery(queryRequest); client.readQuery(queryRequest); //
FragmentRequest has almost the same api as Request, but is provided directly from graphql for consistency.
It is used to access readFragment and writeFragment. The main differences are that they cannot be retreived from options, and that FragmentRequests require idFields to find their cooresponding entities:
final fragmentDoc = gql(
r'''
fragment mySmallSubset on MyType {
myField,
someNewField
}
''',
);
var fragmentRequest = FragmentRequest(
fragment: Fragment(
document: fragmentDoc,
),
idFields: {'__typename': 'MyType', 'id': 1},
);
/// same as experimental convenience api
fragmentRequest = Fragment(document: fragmentDoc).asRequest(
idFields: {'__typename': 'MyType', 'id': 1},
);
final data = client.readFragment(fragmentRequest);
client.writeFragment(fragmentRequest, data);NB You likely want to call the cache access API from your
clientfor automatic broadcasting support.
As of #754 we can now enforce strict structural constraints on data written to the cache. This means that if the client receives structurally invalid data from the network or on client.writeQuery, it will throw an exception.
By default, optimistic data is excluded from these constraints for ease of use via PartialDataCachePolicy.acceptForOptimisticData, as it is easy to miss __typename, etc.
This behavior is configurable via GraphQLCache.partialDataPolicy, which can be set to accept for no constraints or reject for full constraints.
At link execution time, one of the following exceptions can be thrown:
CacheMisconfigurationExceptionif the structure seems like it should write properly, and is perhaps failing due to atypePolicyUnexpectedResponseStructureExceptionif the server response looks malformed.MismatchedDataStructureExceptionin the event of a malformed optimistic result (andPartialDataCachePolicy.reject).CacheMissExceptionif write succeeds butreadQuerythen returnsnull(though data will not be overwritten)
Policies are used to configure various aspects of a request process, and can be set on any *Options object:
// override policies for a single query
client.query(QueryOptions(
// return result from network and save to cache.
fetchPolicy: FetchPolicy.networkOnly,
// ignore all GraphQL errors.
errorPolicy: ErrorPolicy.ignore,
// ignore cache data.
cacheRereadPolicy: CacheRereadPolicy.ignore,
// ...
));Defaults can also be overridden via defaultPolices on the client itself:
GraphQLClient(
defaultPolicies: DefaultPolicies(
// make watched mutations behave like watched queries.
watchMutation: Policies(
FetchPolicy.cacheAndNetwork,
ErrorPolicy.none,
CacheRereadPolicy.mergeOptimistic,
),
),
// ...
)FetchPolicy: determines where the client may return a result from, and whether that result will be saved to the cache.
Possible options:
- cacheFirst: return result from cache. Only fetch from network if cached result is not available.
- cacheAndNetwork: return result from cache first (if it exists), then return network result once it's available.
- cacheOnly: return result from cache if available, fail otherwise.
- noCache: return result from network, fail if network call doesn't succeed, don't save to cache.
- networkOnly: return result from network, fail if network call doesn't succeed, save to cache.
ErrorPolicy: determines the level of events for errors in the execution result.
Possible options:
- none (default): Any GraphQL Errors are treated the same as network errors and any data is ignored from the response.
- ignore: Ignore allows you to read any data that is returned alongside GraphQL Errors, but doesn't save the errors or report them to your UI.
- all: Using the all policy is the best way to notify your users of potential issues while still showing as much data as possible from your server. It saves both data and errors into the Apollo Cache so your UI can use them.
CacheRereadPolicy determines whether and how cache data will be merged into the final QueryResult.data before it is returned.
Possible options:
- mergeOptimistic: Merge relevant optimistic data from the cache before returning.
- ignoreOptimistic: Ignore optimistic data, but still allow for non-optimistic cache rebroadcasts if applicable.
- ignoreAll: Ignore all cache data besides the result, and never rebroadcast the result, even if the underlying cache data changes.
Rebroadcasting behavior only applies to watchMutation and watchQuery, which both return an ObservableQuery.
There is no rebroadcasting option for subscriptions, because it would be indistiguishable from the previous event in the stream.
Rebroadcasting is enabled unless either FetchPolicy.noCache or CacheRereadPolicy.ignoreAll are set,
and whether it considers optimistic results is controlled by the specific CacheRereadPolicy.
If there were problems encountered during a query or mutation, the QueryResult will have an OperationException in the exception field:
/// Container for both [graphqlErrors] returned from the server
/// and any [linkException] that caused a failure.
class OperationException implements Exception {
/// Any graphql errors returned from the operation
List<GraphQLError> graphqlErrors = [];
/// Errors encountered during execution such as network or cache errors
LinkException linkException;
}Example usage:
if (result.hasException) {
if (result.exception.linkException is NetworkException) {
// handle network issues, maybe
}
return Text(result.exception.toString())
}graphql and graphql_flutter now use the gql_link system, re-exporting
gql_http_link,
gql_error_link,
gql_dedupe_link,
and the api from gql_link,
as well as our own custom WebSocketLink and AuthLink.
This makes all link development coordinated across the ecosystem, so that we can leverage existing links like gql_dio_link, and all link-based clients benefit from new link development (such as ferry).
NB:
WebSocketLinkand other "terminating links" must be used withsplitwhen there are multiple terminating links.
The gql_link systm has a well-specified routing system:
a rundown of the composition api:
// kitchen sink:
Link.from([
// common links run before every request
DedupeLink(), // dedupe requests
ErrorLink(onException: reportClientException),
]).split( // split terminating links, or they will break
(request) => request.isSubscription,
MyCustomSubscriptionAuthLink().concat(
WebSocketLink(mySubscriptionEndpoint),
), // MyCustomSubscriptionAuthLink is only applied to subscriptions
AuthLink(getToken: httpAuthenticator).concat(
HttpLink(myAppEndpoint),
)
);
// adding links after here would be pointless, as they would never be accessed
/// both `Link.from` and `link.concat` can be used to chain links:
final Link _link = _authLink.concat(_httpLink);
final Link _link = Link.from([_authLink, _httpLink]);
/// `Link.split` and `link.split` route requests to the left or right based on some condition
/// for instance, if you do `authLink.concat(httpLink).concat(websocketLink)`,
/// `websocketLink` won't see any `subscriptions`
link = Link.split((request) => request.isSubscription, websocketLink, link);When combining links, it is important to note that:
- Terminating links like
HttpLinkandWebsocketLinkmust come at the end of a route, and will not call links following them. - Link order is very important. In
HttpLink(myEndpoint).concat(AuthLink(getToken: authenticate)), theAuthLinkwill never be called.
Cognito Pools
To use with an AppSync GraphQL API that is authorized with AWS Cognito User Pools, simply pass the JWT token for your Cognito user session in to the AuthLink:
// Where `session` is a CognitorUserSession
// from amazon_cognito_identity_dart_2
final token = session.getAccessToken().getJwtToken();
final AuthLink authLink = AuthLink(
getToken: () => token,
);See more: Issue #209
Other Authorization Types
API key, IAM, and Federated provider authorization could be accomplished through custom links, but it is not known to be supported. Anyone wanting to implement this can reference AWS' JS SDK AuthLink implementation.
- Making a custom link: Comment on Issue 173
- AWS JS SDK
auth-link.ts: aws-mobile-appsync-sdk-js:auth-link.ts
This package does not support code-generation out of the box, but graphql_codegen does!
This package extensions on the client which takes away the struggle of serialization and gives you confidence through type-safety. It is also more performant than parsing GraphQL queries at runtime.
For example, by creating the .graphql file
# add_star.graphql
mutation AddStar($starrableId: ID!) {
action: addStar(input: { starrableId: $starrableId }) {
starrable {
viewerHasStarred
}
}
}after building, you'll be able to execute your mutation on the client as:
// add_star.dart
import 'add_star.graphql.dart';
// ..
await client.mutateAddStar(
OptionsMutationAddStar(
variables: VariablesMutationAddStar(starableId: repositoryID)
)
);NOTE: There is a PR for migrating the v3 PersistedQueriesLink, and it works, but requires more consideration. It will be fixed before v4 stable is published
To improve performance you can make use of a concept introduced by apollo called Automatic persisted queries (or short "APQ") to send smaller requests and even enabled CDN caching for your GraphQL API.
ATTENTION: This also requires you to have a GraphQL server that supports APQ, like Apollo's GraphQL Server and will only work for queries (but not for mutations or subscriptions).
You can than use it simply by prepending a PersistedQueriesLink to your normal HttpLink:
final PersistedQueriesLink _apqLink = PersistedQueriesLink(
// To enable GET queries for the first load to allow for CDN caching
useGETForHashedQueries: true,
);
final HttpLink _httpLink = HttpLink(
'https://api.url/graphql',
);
final Link _link = _apqLink.concat(_httpLink);If you are receiving csrf error from your apollo graphql server while uploading file,
your need to add some additional headers to the HttpLink:
Also ensure that you're add contentType to MultipartFile as MediaType
HttpLink httpLink = HttpLink('https://api.url/graphql', defaultHeaders: {
'Content-Type': 'application/json; charset=utf-8',
'X-Apollo-Operation-Name': 'post'
})