Changelog | API Docs | Discord Channel
Server process backing the GraphQL Language Service.
GraphQL Language Service Server provides an interface for building GraphQL language services for IDEs.
Partial support for Microsoft's Language Server Protocol is in place, with more to come in the future.
Supported features include:
- Diagnostics (GraphQL syntax linting/validations) (spec-compliant)
- Autocomplete suggestions (spec-compliant)
- Hyperlink to fragment definitions and named types (type, input, enum) definitions (spec-compliant)
- Outline view support for queries
- Support for
gql
graphql
and other template tags inside javascript, typescript, jsx, ts, vue and svelte files, and an interface to allow custom parsing of all files.
- An LSP compatible client with its own file watcher, that sends watch notifications to the server, such as vscode, nvim, or sublime-lsp.
- Node.js
^18.18.0 || >=20.9.0
or later is required. - (for now) a graphql config file is required
npm install graphql-language-service-server
# or
yarn add graphql-language-service-server
We also provide a CLI interface to this server, see
graphql-language-service-cli
Initialize the GraphQL Language Server with the startServer
function:
import { startServer } from 'graphql-language-service-server';
await startServer({
method: 'node',
});
If you are developing a service or extension, this is the LSP language server you want to run.
When developing vscode extensions, just the above is enough to get started for
your extension's ServerOptions.run.module
, for example.
startServer
function takes the following parameters:
Parameter | Required | Description |
---|---|---|
port | true when method is socket , false otherwise |
port for the LSP server to run on |
method | false |
socket , streams , or node (ipc) |
config | false |
custom graphql-config instance from loadConfig (see example above) |
configDir | false |
the directory where graphql-config is found |
extensions | false |
array of functions to transform the graphql-config and add extensions dynamically |
parser | false |
Customize all file parsing by overriding the default parseDocument function |
fileExtensions | false . defaults to ['.js', '.ts', '.tsx, '.jsx'] |
Customize file extensions used by the default LSP parser |
You must provide a graphql config file
Check out graphql-config to learn the many ways you can define your graphql config
schema: 'packages/api/src/schema.graphql'
documents: 'packages/app/src/components/**/*.{tsx,ts}'
extensions:
endpoints:
example:
url: 'http://localhost:8000'
customExtension:
foo: true
{
"schema": "https://localhost:8000"
}
module.exports = { schema: 'https://localhost:8000' };
use graphql config loadConfig
for
further customization:
import { loadConfig } from 'graphql-config'; // 3.0.0 or later!
// with required params
const config = await loadConfig();
await startServer({
method: 'node',
// or instead of configName, an exact path (relative from rootDir or absolute)
// deprecated for: loadConfigOptions.rootDir. root directory for graphql config file(s), or for relative resolution for exact `filePath`. default process.cwd()
// configDir: '',
loadConfigOptions: {
// any of the options for graphql-config@3 `loadConfig()`
schema: await config.getSchema(),
// rootDir is same as `configDir` before, the path where the graphql config file would be found by cosmic-config
rootDir: 'config/',
// or - the relative or absolute path to your file
filePath: 'exact/path/to/config.js', // (also supports yml, json, ts, toml)
// myPlatform.config.js/json/yaml works now!
configName: 'myPlatform',
},
});
The graphql-config features we support are:
module.exports = {
extensions: {
// add customDirectives (legacy). you can now provide multiple schema pointers to config.schema/project.schema, including inline strings. same with scalars or any SDL type that you'd like to append to the schema
customDirectives: ['@myExampleDirective'],
// a function that returns an array of validation rules, ala https://github.com/graphql/graphql-js/tree/main/src/validation/rules
// note that this file will be loaded by the vscode runtime, so the node version and other factors will come into play
customValidationRules: require('./config/customValidationRules'),
schemaCacheTTL: 1000, // reduce or increase minimum schema cache lifetime from 30000ms (30 seconds). you may want to reduce this if you are developing fullstack with network schema
languageService: {
// this is enabled by default if non-local files are specified in the project `schema`
// NOTE: this will disable all definition lookup for local SDL files
cacheSchemaFileForLookup: true,
// undefined by default which has the same effect as `true`, set to `false` if you are already using // `graphql-eslint` or some other tool for validating graphql in your IDE. Must be explicitly `false` to disable this feature, not just "falsy"
enableValidation: true,
// (experimental) enhanced auto expansion of graphql leaf fields and arguments
fillLeafsOnComplete: true,
// instead of jumping directly to the SDL file, you can override definition peek/jump results to point to different files or locations
// (for example, source files for your schema in any language!)
// based on Relay vscode's pathToLocateCommand
// see LocateCommand type!
locateCommand(projectName, typePath, info) {
// pass more info, such as GraphQLType with the ast node. info.project is also available if you need it
const { path, range } = ourLookupUtility(
projectName,
typePath,
info.type.node,
);
return { uri: path, range }; // range.start.line/range.end.character/etc, base 1
// you can also return relay LSP style
// return '/path/to/file.py:20:23'; // (range: 20:1 )
// return '/path/to/file.py'; // (range: 1:1 1:1)
},
},
},
};
or for multi-project workspaces:
// graphql.config.ts
export default {
projects: {
myProject: {
schema: [
// internally in `graphql-config`, an attempt will be made to combine these schemas into one in-memory schema to use for validation, lookup, etc
'http://localhost:8080',
'./my-project/schema.graphql',
'./my-project/schema.ts',
'@customDirective(arg: String!)',
'scalar CustomScalar',
],
// project specific defaults
extensions: {
languageService: {
cacheSchemaFileForLookup: true,
enableValidation: false,
},
},
},
anotherProject: {
schema: {
'http://localhost:8081': {
customHeaders: { Authorization: 'Bearer example' },
},
},
},
},
// global defaults for all projects
extensions: {
languageService: {
cacheSchemaFileForLookup: false,
enableValidation: true,
},
},
};
You can specify any of these settings globally as above, or per project. Read the graphql-config docs to learn more about this!
For secrets (headers, urls, etc), you can import dotenv()
and set a base path
as you wish in your graphql-config
file to pre-load process.env
variables.
- you may need to manually restart the language server for some of these configurations to take effect
- graphql-config's multi-project support is not related to multi-root workspaces in vscode - in fact, each workspace can have multiple graphql config projects, which is what makes multi-root workspaces tricky to support. coming soon!
The LSP Server reads config by sending workspace/configuration
method when it
initializes.
Note: We still do not support LSP multi-root workspaces but will tackle this very soon!
Many LSP clients beyond vscode offer ways to set these configurations, such as
via initializationOptions
in nvim.coc. The options are mostly designed to
configure graphql-config's load parameters, the only thing we can't configure
with graphql config. The final option can be set in graphql-config
as well
Parameter | Default | Description |
---|---|---|
graphql-config.load.baseDir |
workspace root or process.cwd() | the path where graphql config looks for config files |
graphql-config.load.filepath |
null |
exact filepath of the config file. |
graphql-config.load.configName |
graphql |
config name prefix instead of graphql |
graphql-config.load.legacy |
true |
backwards compatibility with graphql-config@2 |
graphql-config.dotEnvPath |
null |
backwards compatibility with graphql-config@2 |
vscode-graphql.cacheSchemaFileForLookup |
true if schema contains non-SDL files or URLs |
generate an SDL file based on your graphql-config schema configuration for definition lookup and other features. enabled by default when your schema config are URLs or introspection JSON, or if you have any non-local SDL files in schema |
vscode-graphql.schemaCacheTTL |
30000 |
an integer value in milliseconds for the desired minimum cache lifetime for all schemas, which also causes the generated file to be re-written. set to 30s by default. effectively a "lazy" form of polling. if you are developing a schema alongside client queries, you may want to decrease this |
vscode-graphql.debug |
false |
show more verbose log output in the output channel |
all the graphql-config.load.*
configuration values come from static
loadConfig()
options in graphql config.
(more coming soon!)
GraphQL Language Service currently communicates via Stream transport with the IDE server. GraphQL server will receive/send RPC messages to perform language service features, while caching the necessary GraphQL artifacts such as fragment definitions, GraphQL schemas etc. More about the server interface and RPC message format below.
The IDE server should launch a separate GraphQL server with its own child
process for each .graphqlrc.yml
file the IDE finds (using the nearest ancestor
directory relative to the file currently being edited):
./application
./productA
.graphqlrc.yml
ProductAQuery.graphql
ProductASchema.graphql
./productB
.graphqlrc.yml
ProductBQuery.graphql
ProductBSchema.graphql
A separate GraphQL server should be instantiated for ProductA
and ProductB
,
each with its own .graphqlrc.yml
file, as illustrated in the directory
structure above.
The IDE server should manage the lifecycle of the GraphQL server. Ideally, the IDE server should spawn a child process for each of the GraphQL Language Service processes necessary, and gracefully exit the processes as the IDE closes. In case of errors or a sudden halt the GraphQL Language Service will close as the stream from the IDE closes.
GraphQL Language Server uses JSON-RPC to communicate with the IDE servers. Microsoft's language server currently supports two communication transports: Stream (stdio) and IPC. For IPC transport, the reference guide to be used for development is the language server protocol documentation.
For each transport, there is a slight difference in JSON message format, especially in how the methods to be invoked are defined - below are the currently supported methods for each transport (will be updated as progress is made):
Stream | IPC | |
---|---|---|
Diagnostics | getDiagnostics |
textDocument/publishDiagnostics |
Autocompletion | getAutocompleteSuggestions |
textDocument/completion |
Outline | getOutline |
textDocument/outline |
Document Symbols | getDocumentSymbols |
textDocument/symbols |
Workspace Symbols | getWorkspaceSymbols |
workspace/symbols |
Go-to definition | getDefinition |
textDocument/definition |
Workspace Definition | getWorkspaceDefinition |
workspace/definition |
File Events | Not supported yet | didOpen/didClose/didSave/didChange events |