esm: move hook execution to separate thread

PR-URL: #44710
Backport-PR-URL: #50669
Reviewed-By: Yagiz Nizipli <yagiz@nizipli.com>
Reviewed-By: Geoffrey Booth <webadmin@geoffreybooth.com>
Co-authored-by: Antoine du Hamel <duhamelantoine1995@gmail.com>
Co-authored-by: Geoffrey Booth <webadmin@geoffreybooth.com>
Co-authored-by: Michaël Zasso <targos@protonmail.com>

Author: 4 people

doc/api/esm.md

@@ -7,6 +7,9 @@
 <!-- YAML
 added: v8.5.0
 changes:
+  - version: REPLACEME
+    pr-url: https://github.com/nodejs/node/pull/44710
+    description: Loader hooks are executed off the main thread.
   - version:
     - v18.6.0
     pr-url: https://github.com/nodejs/node/pull/42623
@@ -325,6 +328,9 @@ added:
   - v13.9.0
   - v12.16.2
 changes:
+  - version: REPLACEME
+    pr-url: https://github.com/nodejs/node/pull/44710
+    description: This API now returns a string synchronously instead of a Promise.
   - version:
       - v16.2.0
       - v14.18.0
@@ -340,29 +346,26 @@ command flag enabled.
 * `specifier` {string} The module specifier to resolve relative to `parent`.
 * `parent` {string|URL} The absolute parent module URL to resolve from. If none
   is specified, the value of `import.meta.url` is used as the default.
-* Returns: {Promise}
+* Returns: {string}
 
 Provides a module-relative resolution function scoped to each module, returning
-the URL string.
+the URL string. In alignment with browser behavior, this now returns
+synchronously.
 
-<!-- eslint-skip -->
+> **Caveat** This can result in synchronous file-system operations, which
+> can impact performance similarly to `require.resolve`.
 
 ```js
-const dependencyAsset = await import.meta.resolve('component-lib/asset.css');
+const dependencyAsset = import.meta.resolve('component-lib/asset.css');
 ```
 
 `import.meta.resolve` also accepts a second argument which is the parent module
-from which to resolve from:
-
-<!-- eslint-skip -->
+from which to resolve:
 
 ```js
-await import.meta.resolve('./dep', import.meta.url);
+import.meta.resolve('./dep', import.meta.url);
 ```
 
-This function is asynchronous because the ES module resolver in Node.js is
-allowed to be asynchronous.
-
 ## Interoperability with CommonJS
 
 ### `import` statements
@@ -729,6 +732,11 @@ A hook that returns without calling `next<hookName>()` _and_ without returning
 `shortCircuit: true` also triggers an exception. These errors are to help
 prevent unintentional breaks in the chain.
 
+Hooks are run in a separate thread, isolated from the main. That means it is a
+different [realm](https://tc39.es/ecma262/#realm). The hooks thread may be
+terminated by the main thread at any time, so do not depend on asynchronous
+operations to (like `console.log`) complete.
+
 #### `resolve(specifier, context, nextResolve)`
 
 <!-- YAML
@@ -759,7 +767,7 @@ changes:
   Node.js default `resolve` hook after the last user-supplied `resolve` hook
   * `specifier` {string}
   * `context` {Object}
-* Returns: {Object}
+* Returns: {Object|Promise}
   * `format` {string|null|undefined} A hint to the load hook (it might be
     ignored)
     `'builtin' | 'commonjs' | 'json' | 'module' | 'wasm'`
@@ -769,6 +777,9 @@ changes:
     terminate the chain of `resolve` hooks. **Default:** `false`
   * `url` {string} The absolute URL to which this input resolves
 
+> **Caveat** Despite support for returning promises and async functions, calls
+> to `resolve` may block the main thread which can impact performance.
+
 The `resolve` hook chain is responsible for telling Node.js where to find and
 how to cache a given `import` statement or expression. It can optionally return
 its format (such as `'module'`) as a hint to the `load` hook. If a format is
@@ -794,7 +805,7 @@ Node.js module specifier resolution behavior_ when calling `defaultResolve`, the
 `context.conditions` array originally passed into the `resolve` hook.
 
 ```js
-export async function resolve(specifier, context, nextResolve) {
+export function resolve(specifier, context, nextResolve) {
   const { parentURL = null } = context;
 
   if (Math.random() > 0.5) { // Some condition.
@@ -1067,6 +1078,25 @@ import CoffeeScript from 'coffeescript';
 
 const baseURL = pathToFileURL(`${cwd()}/`).href;
 
+// CoffeeScript files end in .coffee, .litcoffee, or .coffee.md.
+const extensionsRegex = /\.coffee$|\.litcoffee$|\.coffee\.md$/;
+
+export function resolve(specifier, context, nextResolve) {
+  if (extensionsRegex.test(specifier)) {
+    const { parentURL = baseURL } = context;
+
+    // Node.js normally errors on unknown file extensions, so return a URL for
+    // specifiers ending in the CoffeeScript file extensions.
+    return {
+      shortCircuit: true,
+      url: new URL(specifier, parentURL).href,
+    };
+  }
+
+  // Let Node.js handle all other specifiers.
+  return nextResolve(specifier);
+}
+
 export async function load(url, context, nextLoad) {
   if (extensionsRegex.test(url)) {
     // Now that we patched resolve to let CoffeeScript URLs through, we need to

lib/internal/main/worker_thread.js

@@ -10,6 +10,7 @@ const {
   ObjectDefineProperty,
   PromisePrototypeThen,
   RegExpPrototypeExec,
+  SafeWeakMap,
   globalThis: {
     Atomics,
     SharedArrayBuffer,
@@ -88,23 +89,25 @@ port.on('message', (message) => {
     const {
       argv,
       cwdCounter,
-      filename,
       doEval,
-      workerData,
       environmentData,
-      publicPort,
+      filename,
+      hasStdin,
       manifestSrc,
       manifestURL,
-      hasStdin,
+      publicPort,
+      workerData,
     } = message;
 
-    if (argv !== undefined) {
-      ArrayPrototypePushApply(process.argv, argv);
-    }
+    if (doEval !== 'internal') {
+      if (argv !== undefined) {
+        ArrayPrototypePushApply(process.argv, argv);
+      }
 
-    const publicWorker = require('worker_threads');
-    publicWorker.parentPort = publicPort;
-    publicWorker.workerData = workerData;
+      const publicWorker = require('worker_threads');
+      publicWorker.parentPort = publicPort;
+      publicWorker.workerData = workerData;
+    }
 
     require('internal/worker').assignEnvironmentData(environmentData);
 
@@ -137,31 +140,47 @@ port.on('message', (message) => {
     debug(`[${threadId}] starts worker script ${filename} ` +
           `(eval = ${doEval}) at cwd = ${process.cwd()}`);
     port.postMessage({ type: UP_AND_RUNNING });
-    if (doEval === 'classic') {
-      const { evalScript } = require('internal/process/execution');
-      const name = '[worker eval]';
-      // This is necessary for CJS module compilation.
-      // TODO: pass this with something really internal.
-      ObjectDefineProperty(process, '_eval', {
-        __proto__: null,
-        configurable: true,
-        enumerable: true,
-        value: filename,
-      });
-      ArrayPrototypeSplice(process.argv, 1, 0, name);
-      evalScript(name, filename);
-    } else if (doEval === 'module') {
-      const { evalModule } = require('internal/process/execution');
-      PromisePrototypeThen(evalModule(filename), undefined, (e) => {
-        workerOnGlobalUncaughtException(e, true);
-      });
-    } else {
-      // script filename
-      // runMain here might be monkey-patched by users in --require.
-      // XXX: the monkey-patchability here should probably be deprecated.
-      ArrayPrototypeSplice(process.argv, 1, 0, filename);
-      const CJSLoader = require('internal/modules/cjs/loader');
-      CJSLoader.Module.runMain(filename);
+    switch (doEval) {
+      case 'internal': {
+        // Create this WeakMap in js-land because V8 has no C++ API for WeakMap.
+        internalBinding('module_wrap').callbackMap = new SafeWeakMap();
+        require(filename)(workerData, publicPort);
+        break;
+      }
+
+      case 'classic': {
+        const { evalScript } = require('internal/process/execution');
+        const name = '[worker eval]';
+        // This is necessary for CJS module compilation.
+        // TODO: pass this with something really internal.
+        ObjectDefineProperty(process, '_eval', {
+          __proto__: null,
+          configurable: true,
+          enumerable: true,
+          value: filename,
+        });
+        ArrayPrototypeSplice(process.argv, 1, 0, name);
+        evalScript(name, filename);
+        break;
+      }
+
+      case 'module': {
+        const { evalModule } = require('internal/process/execution');
+        PromisePrototypeThen(evalModule(filename), undefined, (e) => {
+          workerOnGlobalUncaughtException(e, true);
+        });
+        break;
+      }
+
+      default: {
+        // script filename
+        // runMain here might be monkey-patched by users in --require.
+        // XXX: the monkey-patchability here should probably be deprecated.
+        ArrayPrototypeSplice(process.argv, 1, 0, filename);
+        const CJSLoader = require('internal/modules/cjs/loader');
+        CJSLoader.Module.runMain(filename);
+        break;
+      }
     }
   } else if (message.type === STDIO_PAYLOAD) {
     const { stream, chunks } = message;