module: support require()ing synchronous ESM graphs

This patch adds `require()` support for synchronous ESM graphs under
the flag --experimental-require-module.

This is based on the the following design aspect of ESM:

- The resolution can be synchronous (up to the host)
- The evaluation of a synchronous graph (without top-level await)
  is also synchronous, and, by the time the module graph is
  instantiated (before evaluation starts), this is is already known.

When the module being require()ed has .mjs extension or there are
other explicit indicators that it's an ES module, we load it as an
ES module. If the graph is synchronous, we return the module namespace
as the exports. If the graph contains top-level await, we throw an
error before evaluating the module. If an additional flag
--print-pending-tla is passed, we proceeds to evaluation but do not
run the microtasks, only to find out where the TLA is and print
their location to help users fix them.

If there are not explicit indicators whether a .js file is CJS or ESM,
we parse it as CJS first. If the parse error indicates that it contains
ESM syntax, we parse it again as ESM. If the second parsing succeeds,
we continue to treat it as ESM.

doc/api/cli.md

@@ -871,6 +871,22 @@ added: v11.8.0
 
 Use the specified file as a security policy.
 
+### `--experimental-require-module`
+
+<!-- YAML
+added: REPLACEME
+-->
+
+> Stability: 1.1 - Active Developement
+
+Supports loading a synchronous ES module graph in `require()`. If the module
+graph is not synchronous (contains top-level await), it throws an error.
+
+By default, a `.js` file will be parsed as a CommonJS module first. If it
+contains ES module syntax, Node.js will try to parse and evaluate the module
+again as an ES module. If it turns out to be synchronous and can be evaluated
+successfully, the module namespace object will be returned by `require()`.
+
 ### `--experimental-sea-config`
 
 <!-- YAML
@@ -2523,6 +2539,7 @@ Node.js options that are allowed are:
 * `--experimental-network-imports`
 * `--experimental-permission`
 * `--experimental-policy`
+* `--experimental-require-module`
 * `--experimental-shadow-realm`
 * `--experimental-specifier-resolution`
 * `--experimental-top-level-await`

lib/internal/modules/cjs/loader.js

@@ -60,6 +60,7 @@ const {
   StringPrototypeSlice,
   StringPrototypeSplit,
   StringPrototypeStartsWith,
+  Symbol,
 } = primordials;
 
 // Map used to store CJS parsing data.
@@ -107,7 +108,6 @@ const { safeGetenv } = internalBinding('credentials');
 const {
   privateSymbols: {
     require_private_symbol,
-    host_defined_option_symbol,
   },
 } = internalBinding('util');
 const {
@@ -161,6 +161,8 @@ let requireDepth = 0;
 let isPreloading = false;
 let statCache = null;
 
+const is_main_symbol = Symbol('is-main-module');
+
 /**
  * Our internal implementation of `require`.
  * @param {Module} module Parent module of what is being required
@@ -271,6 +273,7 @@ function Module(id = '', parent) {
     setOwnProperty(this.__proto__, 'require', makeRequireFunction(this, redirects));
   }
   this[require_private_symbol] = internalRequire;
+  this[is_main_symbol] = false;  // Set to true by the entry point handler.
 }
 
 /** @type {Record<string, Module>} */
@@ -396,6 +399,10 @@ function initializeCJS() {
   // TODO(joyeecheung): deprecate this in favor of a proper hook?
   Module.runMain =
     require('internal/modules/run_main').executeUserEntryPoint;
+
+  if (getOptionValue('--experimental-require-module')) {
+    Module._extensions['.mjs'] = loadESMFromCJS;
+  }
 }
 
 // Given a module name, and a list of paths to test, returns the first
@@ -1010,6 +1017,7 @@ Module._load = function(request, parent, isMain) {
     setOwnProperty(process, 'mainModule', module);
     setOwnProperty(module.require, 'main', process.mainModule);
     module.id = '.';
+    module[is_main_symbol] = true;
   }
 
   reportModuleToWatchMode(filename);
@@ -1270,46 +1278,58 @@ function wrapSafe(filename, content, cjsModuleInstance, codeCache) {
     );
 
     // Cache the source map for the module if present.
-    if (script.sourceMapURL) {
-      maybeCacheSourceMap(filename, content, this, false, undefined, script.sourceMapURL);
+    const { sourceMapURL } = script;
+    if (sourceMapURL) {
+      maybeCacheSourceMap(filename, content, this, false, undefined, sourceMapURL);
     }
 
-    return runScriptInThisContext(script, true, false);
+    return {
+      __proto__: null,
+      function: runScriptInThisContext(script, true, false),
+      sourceMapURL,
+      retryAsESM: false,
+    };
   }
 
-  try {
-    const result = compileFunctionForCJSLoader(content, filename);
-    result.function[host_defined_option_symbol] = hostDefinedOptionId;
-
-    // cachedDataRejected is only set for cache coming from SEA.
-    if (codeCache &&
-        result.cachedDataRejected !== false &&
-        internalBinding('sea').isSea()) {
-      process.emitWarning('Code cache data rejected.');
-    }
+  const result = compileFunctionForCJSLoader(content, filename);
 
-    // Cache the source map for the module if present.
-    if (result.sourceMapURL) {
-      maybeCacheSourceMap(filename, content, this, false, undefined, result.sourceMapURL);
-    }
+  // cachedDataRejected is only set for cache coming from SEA.
+  if (codeCache &&
+      result.cachedDataRejected !== false &&
+      internalBinding('sea').isSea()) {
+    process.emitWarning('Code cache data rejected.');
+  }
 
-    return result.function;
-  } catch (err) {
-    if (process.mainModule === cjsModuleInstance) {
-      const { enrichCJSError } = require('internal/modules/esm/translators');
-      enrichCJSError(err, content, filename);
-    }
-    throw err;
+  // Cache the source map for the module if present.
+  if (result.sourceMapURL) {
+    maybeCacheSourceMap(filename, content, this, false, undefined, result.sourceMapURL);
   }
+
+  return result;
+}
+
+// Resolve and evaluate as ESM, synchronously.
+function loadESMFromCJS(mod, filename) {
+  const cascadedLoader = require('internal/modules/esm/loader').getOrInitializeCascadedLoader();
+  // Note that we are still using the CJS's path resolution here.
+  const parent = moduleParentCache.get(mod)?.filename;
+  const base = parent ? pathToFileURL(parent) : parent;
+  // console.log('loadESMFromCJS', mod, filename, base);
+  const specifier = mod[is_main_symbol] ? pathToFileURL(mod.filename) : mod.id;
+  const job = cascadedLoader.getModuleJobSync(specifier, base, kEmptyObject, 'from-cjs-error');
+  const { namespace } = job.runSync();
+  // TODO(joyeecheung): maybe we can do some special handling for default here. Maybe we don't.
+  mod.exports = namespace;
 }
 
 /**
  * Run the file contents in the correct scope or sandbox. Expose the correct helper variables (`require`, `module`,
  * `exports`) to the file. Returns exception, if any.
  * @param {string} content The source code of the module
  * @param {string} filename The file path of the module
+ * @param {boolean} loadAsESM Whether it's known to be ESM - i.e. suffix is .mjs.
  */
-Module.prototype._compile = function(content, filename) {
+Module.prototype._compile = function(content, filename, loadAsESM = false) {
   let moduleURL;
   let redirects;
   const manifest = policy()?.manifest;
@@ -1319,8 +1339,21 @@ Module.prototype._compile = function(content, filename) {
     manifest.assertIntegrity(moduleURL, content);
   }
 
-  const compiledWrapper = wrapSafe(filename, content, this);
+  let compiledWrapper;
+  if (!loadAsESM) {
+    const result = wrapSafe(filename, content, this);
+    compiledWrapper = result.function;
+    loadAsESM = result.retryAsESM;
+  }
+
+  if (loadAsESM) {
+    loadESMFromCJS(this);
+    return;
+  }
 
+  // TODO(joyeecheung): the detection below is unnecessarily complex. Maybe just
+  // use the is_main_symbol, or a break_on_start_symbol that gets passed from
+  // higher level instead of doing hacky detecion here.
   let inspectorWrapper = null;
   if (getOptionValue('--inspect-brk') && process._eval == null) {
     if (!resolvedArgv) {
@@ -1344,6 +1377,7 @@ Module.prototype._compile = function(content, filename) {
       inspectorWrapper = internalBinding('inspector').callAndPauseOnStart;
     }
   }
+
   const dirname = path.dirname(filename);
   const require = makeRequireFunction(this, redirects);
   let result;
@@ -1370,6 +1404,7 @@ Module.prototype._compile = function(content, filename) {
  */
 Module._extensions['.js'] = function(module, filename) {
   // If already analyzed the source, then it will be cached.
+  // TODO(joyeecheung): pass as buffer.
   const cached = cjsParseCache.get(module);
   let content;
   if (cached?.source) {
@@ -1378,7 +1413,8 @@ Module._extensions['.js'] = function(module, filename) {
   } else {
     content = fs.readFileSync(filename, 'utf8');
   }
-  if (StringPrototypeEndsWith(filename, '.js')) {
+  if (!getOptionValue('--experimental-require-module') &&
+      StringPrototypeEndsWith(filename, '.js')) {
     const pkg = packageJsonReader.getNearestParentPackageJSON(filename);
     // Function require shouldn't be used in ES modules.
     if (pkg?.data.type === 'module') {
@@ -1414,7 +1450,8 @@ Module._extensions['.js'] = function(module, filename) {
       throw err;
     }
   }
-  module._compile(content, filename);
+
+  module._compile(content, filename, false);
 };
 
 /**

lib/internal/modules/esm/loader.js

@@ -15,6 +15,7 @@ const {
   hardenRegExp,
 } = primordials;
 
+const assert = require('internal/assert');
 const {
   ERR_REQUIRE_ESM,
   ERR_UNKNOWN_MODULE_FORMAT,
@@ -228,12 +229,12 @@ class ModuleLoader {
     return this.getJobFromResolveResult(resolveResult, parentURL, importAttributes);
   }
 
-  getModuleJobSync(specifier, parentURL, importAttributes) {
-    const resolveResult = this.resolveSync(specifier, parentURL, importAttributes);
-    return this.getJobFromResolveResult(resolveResult, parentURL, importAttributes, true);
+  getModuleJobSync(specifier, parentURL, importAttributes, requireESMHint) {
+    const resolveResult = this.resolveSync(specifier, parentURL, importAttributes, requireESMHint);
+    return this.getJobFromResolveResult(resolveResult, parentURL, importAttributes, true, requireESMHint);
   }
 
-  getJobFromResolveResult(resolveResult, parentURL, importAttributes, sync) {
+  getJobFromResolveResult(resolveResult, parentURL, importAttributes, sync, requireESMHint) {
     const { url, format } = resolveResult;
     const resolvedImportAttributes = resolveResult.importAttributes ?? importAttributes;
     let job = this.loadCache.get(url, resolvedImportAttributes.type);
@@ -244,7 +245,7 @@ class ModuleLoader {
     }
 
     if (job === undefined) {
-      job = this.#createModuleJob(url, resolvedImportAttributes, parentURL, format, sync);
+      job = this.#createModuleJob(url, resolvedImportAttributes, parentURL, format, sync, requireESMHint);
     }
 
     return job;
@@ -261,7 +262,7 @@ class ModuleLoader {
    *                          `resolve` hook
    * @returns {Promise<ModuleJob>} The (possibly pending) module job
    */
-  #createModuleJob(url, importAttributes, parentURL, format, sync) {
+  #createModuleJob(url, importAttributes, parentURL, format, sync, requireESMHint) {
     const callTranslator = ({ format: finalFormat, responseURL, source }, isMain) => {
       const translator = getTranslators().get(finalFormat);
 
@@ -274,7 +275,7 @@ class ModuleLoader {
     const context = { format, importAttributes };
 
     const moduleProvider = sync ?
-      (url, isMain) => callTranslator(this.loadSync(url, context), isMain) :
+      (url, isMain) => callTranslator(this.loadSync(url, context, requireESMHint), isMain) :
       async (url, isMain) => callTranslator(await this.load(url, context), isMain);
 
     const inspectBrk = (
@@ -358,26 +359,30 @@ class ModuleLoader {
    * Just like `resolve` except synchronous. This is here specifically to support
    * `import.meta.resolve` which must happen synchronously.
    */
-  resolveSync(originalSpecifier, parentURL, importAttributes) {
-    if (this.#customizations) {
+  resolveSync(originalSpecifier, parentURL, importAttributes, requireESMHint) {
+    // If this comes from the require(esm) fallback, don't apply loader hooks which are on
+    // a separate thread. This is ignored by require(cjs) already anyway.
+    // TODO(joyeecheung): add support in hooks for this?
+    if (this.#customizations && !requireESMHint) {
       return this.#customizations.resolveSync(originalSpecifier, parentURL, importAttributes);
     }
-    return this.defaultResolve(originalSpecifier, parentURL, importAttributes);
+    return this.defaultResolve(originalSpecifier, parentURL, importAttributes, requireESMHint);
   }
 
   /**
    * Our `defaultResolve` is synchronous and can be used in both
    * `resolve` and `resolveSync`. This function is here just to avoid
    * repeating the same code block twice in those functions.
    */
-  defaultResolve(originalSpecifier, parentURL, importAttributes) {
+  defaultResolve(originalSpecifier, parentURL, importAttributes, requireESMHint) {
     defaultResolve ??= require('internal/modules/esm/resolve').defaultResolve;
 
     const context = {
       __proto__: null,
       conditions: this.#defaultConditions,
       importAttributes,
       parentURL,
+      requireESMHint,
     };
 
     return defaultResolve(originalSpecifier, context);
@@ -398,14 +403,23 @@ class ModuleLoader {
     return result;
   }
 
-  loadSync(url, context) {
+  loadSync(url, context, requireESMHint) {
     defaultLoadSync ??= require('internal/modules/esm/load').defaultLoadSync;
-
-    let result = this.#customizations ?
+    const isRequireModuleAllowed = getOptionValue('--experimental-require-module');
+    if (requireESMHint === 'from-cjs-error') {
+      assert(isRequireModuleAllowed);
+      context.format = 'module';
+    }
+    // If this comes from the require(esm) fallback, don't apply loader hooks which are on
+    // a separate thread. This is ignored by require(cjs) already anyway.
+    // TODO(joyeecheung): add support in hooks for this?
+    let result = this.#customizations && !requireESMHint ?
       this.#customizations.loadSync(url, context) :
       defaultLoadSync(url, context);
+
+    // TODO(joyeecheung): we need a better way to detect the format and cache the result.
     let format = result?.format;
-    if (format === 'module') {
+    if (format === 'module' && !isRequireModuleAllowed) {
       throw new ERR_REQUIRE_ESM(url, true);
     }
     if (format === 'commonjs') {