Skip to content

Environment API for Runtimes

Experimental

Initial work for this API was introduced in Vite 5.1 with the name "Vite Runtime API". This guide describes a revised API, renamed to Environment API. This API will be released in Vite 6 as experimental. You can already test it in the latest [email protected] version.

Resources:

Please share with us your feedback as you test the proposal.

Environment factories

Environments factories are intended to be implemented by Environment providers like Cloudflare, and not by end users. Environment factories return a EnvironmentOptions for the most common case of using the target runtime for both dev and build environments. The default environment options can also be set so the user doesn't need to do it.

ts
function createWorkedEnvironment(
  userConfig: EnvironmentOptions,
): EnvironmentOptions {
  return mergeConfig(
    {
      resolve: {
        conditions: [
          /*...*/
        ],
      },
      dev: {
        createEnvironment(name, config) {
          return createWorkerdDevEnvironment(name, config, {
            hot: customHotChannel(),
          })
        },
      },
      build: {
        createEnvironment(name, config) {
          return createWorkerdBuildEnvironment(name, config)
        },
      },
    },
    userConfig,
  )
}

Then the config file can be written as:

js
import { createWorkerdEnvironment } from 'vite-environment-workerd'

export default {
  environments: {
    ssr: createWorkerdEnvironment({
      build: {
        outDir: '/dist/ssr',
      },
    }),
    rsc: createWorkerdEnvironment({
      build: {
        outDir: '/dist/rsc',
      },
    }),
  },
}

and frameworks can use an environment with the workerd runtime to do SSR using:

js
const ssrEnvironment = server.environments.ssr

Creating a new environment factory

A Vite dev server exposes two environments by default: a client environment and an ssr environment. The client environment is a browser environment by default, and the module runner is implemented by importing the virtual module /@vite/client to client apps. The SSR environment runs in the same Node runtime as the Vite server by default and allows application servers to be used to render requests during dev with full HMR support.

The transformed source code is called a module, and the relationships between the modules processed in each environment are kept in a module graph. The transformed code for these modules is sent to the runtimes associated with each environment to be executed. When a module is evaluated in the runtime, its imported modules will be requested triggering the processing of a section of the module graph.

A Vite Module Runner allows running any code by processing it with Vite plugins first. It is different from server.ssrLoadModule because the runner implementation is decoupled from the server. This allows library and framework authors to implement their layer of communication between the Vite server and the runner. The browser communicates with its corresponding environment using the server Web Socket and through HTTP requests. The Node Module runner can directly do function calls to process modules as it is running in the same process. Other environments could run modules connecting to a JS runtime like workerd, or a Worker Thread as Vitest does.

One of the goals of this feature is to provide a customizable API to process and run code. Users can create new environment factories using the exposed primitives.

ts
import { DevEnvironment, RemoteEnvironmentTransport } from 'vite'

function createWorkerdDevEnvironment(name: string, config: ResolvedConfig, context: DevEnvironmentContext) {
  const hot = /* ... */
  const connection = /* ... */
  const transport = new RemoteEnvironmentTransport({
    send: (data) => connection.send(data),
    onMessage: (listener) => connection.on('message', listener),
  })

  const workerdDevEnvironment = new DevEnvironment(name, config, {
    options: {
      resolve: { conditions: ['custom'] },
      ...context.options,
    },
    hot,
    remoteRunner: {
      transport,
    },
  })
  return workerdDevEnvironment
}

ModuleRunner

A module runner is instantiated in the target runtime. All APIs in the next section are imported from vite/module-runner unless stated otherwise. This export entry point is kept as lightweight as possible, only exporting the minimal needed to create module runners.

Type Signature:

ts
export class ModuleRunner {
  constructor(
    public options: ModuleRunnerOptions,
    public evaluator: ModuleEvaluator,
    private debug?: ModuleRunnerDebugger,
  ) {}
  /**
   * URL to execute. Accepts file path, server path, or id relative to the root.
   */
  public async import<T = any>(url: string): Promise<T>
  /**
   * Clear all caches including HMR listeners.
   */
  public clearCache(): void
  /**
   * Clears all caches, removes all HMR listeners, and resets source map support.
   * This method doesn't stop the HMR connection.
   */
  public async close(): Promise<void>
  /**
   * Returns `true` if the runner has been closed by calling `close()` method.
   */
  public isClosed(): boolean
}

The module evaluator in ModuleRunner is responsible for executing the code. Vite exports ESModulesEvaluator out of the box, it uses new AsyncFunction to evaluate the code. You can provide your own implementation if your JavaScript runtime doesn't support unsafe evaluation.

Module runner exposes import method. When Vite server triggers full-reload HMR event, all affected modules will be re-executed. Be aware that Module Runner doesn't update exports object when this happens (it overrides it), you would need to run import or get the module from evaluatedModules again if you rely on having the latest exports object.

Example Usage:

js
import { ModuleRunner, ESModulesEvaluator } from 'vite/module-runner'
import { root, fetchModule } from './rpc-implementation.js'

const moduleRunner = new ModuleRunner(
  {
    root,
    fetchModule,
    // you can also provide hmr.connection to support HMR
  },
  new ESModulesEvaluator(),
)

await moduleRunner.import('/src/entry-point.js')

ModuleRunnerOptions

ts
export interface ModuleRunnerOptions {
  /**
   * Root of the project
   */
  root: string
  /**
   * A set of methods to communicate with the server.
   */
  transport: RunnerTransport
  /**
   * Configure how source maps are resolved. Prefers `node` if `process.setSourceMapsEnabled` is available.
   * Otherwise it will use `prepareStackTrace` by default which overrides `Error.prepareStackTrace` method.
   * You can provide an object to configure how file contents and source maps are resolved for files that were not processed by Vite.
   */
  sourcemapInterceptor?:
    | false
    | 'node'
    | 'prepareStackTrace'
    | InterceptorOptions
  /**
   * Disable HMR or configure HMR options.
   */
  hmr?:
    | false
    | {
        /**
         * Configure how HMR communicates between the client and the server.
         */
        connection: ModuleRunnerHMRConnection
        /**
         * Configure HMR logger.
         */
        logger?: false | HMRLogger
      }
  /**
   * Custom module cache. If not provided, it creates a separate module cache for each module runner instance.
   */
  evaluatedModules?: EvaluatedModules
}

ModuleEvaluator

Type Signature:

ts
export interface ModuleEvaluator {
  /**
   * Number of prefixed lines in the transformed code.
   */
  startOffset?: number
  /**
   * Evaluate code that was transformed by Vite.
   * @param context Function context
   * @param code Transformed code
   * @param id ID that was used to fetch the module
   */
  runInlinedModule(
    context: ModuleRunnerContext,
    code: string,
    id: string,
  ): Promise<any>
  /**
   * evaluate externalized module.
   * @param file File URL to the external module
   */
  runExternalModule(file: string): Promise<any>
}

Vite exports ESModulesEvaluator that implements this interface by default. It uses new AsyncFunction to evaluate code, so if the code has inlined source map it should contain an offset of 2 lines to accommodate for new lines added. This is done automatically by the ESModulesEvaluator. Custom evaluators will not add additional lines.

RunnerTransport

Type Signature:

ts
interface RunnerTransport {
  /**
   * A method to get the information about the module.
   */
  fetchModule: FetchFunction
}

Transport object that communicates with the environment via an RPC or by directly calling the function. By default, you need to pass an object with fetchModule method - it can use any type of RPC inside of it, but Vite also exposes bidirectional transport interface via a RemoteRunnerTransport class to make the configuration easier. You need to couple it with the RemoteEnvironmentTransport instance on the server like in this example where module runner is created in the worker thread:

ts
import { parentPort } from 'node:worker_threads'
import { fileURLToPath } from 'node:url'
import {
  ESModulesEvaluator,
  ModuleRunner,
  RemoteRunnerTransport,
} from 'vite/module-runner'

const runner = new ModuleRunner(
  {
    root: fileURLToPath(new URL('./', import.meta.url)),
    transport: new RemoteRunnerTransport({
      send: (data) => parentPort.postMessage(data),
      onMessage: (listener) => parentPort.on('message', listener),
      timeout: 5000,
    }),
  },
  new ESModulesEvaluator(),
)
ts
import { BroadcastChannel } from 'node:worker_threads'
import { createServer, RemoteEnvironmentTransport, DevEnvironment } from 'vite'

function createWorkerEnvironment(name, config, context) {
  const worker = new Worker('./worker.js')
  return new DevEnvironment(name, config, {
    hot: /* custom hot channel */,
    remoteRunner: {
      transport: new RemoteEnvironmentTransport({
        send: (data) => worker.postMessage(data),
        onMessage: (listener) => worker.on('message', listener),
      }),
    },
  })
}

await createServer({
  environments: {
    worker: {
      dev: {
        createEnvironment: createWorkerEnvironment,
      },
    },
  },
})

RemoteRunnerTransport and RemoteEnvironmentTransport are meant to be used together, but you don't have to use them at all. You can define your own function to communicate between the runner and the server. For example, if you connect to the environment via an HTTP request, you can call fetch().json() in fetchModule function:

ts
import { ESModulesEvaluator, ModuleRunner } from 'vite/module-runner'

export const runner = new ModuleRunner(
  {
    root: fileURLToPath(new URL('./', import.meta.url)),
    transport: {
      async fetchModule(id, importer) {
        const response = await fetch(
          `http://my-vite-server/fetch?id=${id}&importer=${importer}`,
        )
        return response.json()
      },
    },
  },
  new ESModulesEvaluator(),
)

await runner.import('/entry.js')

ModuleRunnerHMRConnection

Type Signature:

ts
export interface ModuleRunnerHMRConnection {
  /**
   * Checked before sending messages to the server.
   */
  isReady(): boolean
  /**
   * Send a message to the server.
   */
  send(payload: HotPayload): void
  /**
   * Configure how HMR is handled when this connection triggers an update.
   * This method expects that the connection will start listening for HMR updates and call this callback when it's received.
   */
  onUpdate(callback: (payload: HotPayload) => void): void
}

This interface defines how HMR communication is established. Vite exports ServerHMRConnector from the main entry point to support HMR during Vite SSR. The isReady and send methods are usually called when the custom event is triggered (like, import.meta.hot.send("my-event")).

onUpdate is called only once when the new module runner is initiated. It passed down a method that should be called when connection triggers the HMR event. The implementation depends on the type of connection (as an example, it can be WebSocket/EventEmitter/MessageChannel), but it usually looks something like this:

js
function onUpdate(callback) {
  this.connection.on('hmr', (event) => callback(event.data))
}

The callback is queued and it will wait for the current update to be resolved before processing the next update. Unlike the browser implementation, HMR updates in a module runner will wait until all listeners (like, vite:beforeUpdate/vite:beforeFullReload) are finished before updating the modules.

ត្រូវបានចេញផ្សាយដោយស្ថិតក្រោមអញ្ញាប័ណ្ណ MIT ។ (dev)