Modules: Specify or infer module type for `new vm.Script`

[azz]

Is your feature request related to a problem? Please describe.

Consider the following:

const vm = require('vm');

new vm.Script(`import fs from "fs";`, { filename: 'script.mjs' });

When executed with --experimental-modules, as of 12.0.0 this throws:

script.mjs:1
import fs from "fs";
       ^^

SyntaxError: Unexpected identifier

Describe the solution you'd like

1. The "input-type" to be inferred from the filename .mjs

2. An additional option added to vm.Script, called type, which overrides file-based inference.

   new vm.Script(`import fs from "fs";`, { type: 'module' });

3. Resolve hooks should be available, e.g.:

   async function resolve(specifier, parentURL, defaultResolve) {
     return defaultResolve('./script.mjs', 'file:///');
   }
   new vm.Script(`import fs from "fs"`, { loader: { resolve } });

Describe alternatives you've considered

Alternatively, an API to determine the type/format of a given file path using the same rules that determine the type of node --experimental-modules script.js. E.g.:

const type = require.resolveType('/path/to/file.js');

This would work for some cases but the resolve hook is more powerful.

[devsnek]

use vm.SourceTextModule for source text module sources: https://nodejs.org/api/vm.html#vm_class_vm_sourcetextmodule

[azz]

Thanks. Is there a mechanism to determine if a module is CJS/ESM? Branching logic will need to introduced to either use vm.Script or vm.SourceTextModule based on this.

[devsnek]

there is no logic you can use to reliably determine if a string is a module or a script. as such, node doesn't provide anything. however, you can come up with your own logic that fits your needs. for example if you know that your modules will always contain imports or exports you can use acorn or something to parse and check for them.

[azz]

I want to follow the same log node uses:

• If .js, read package.json#type, default to CJS.
• If .mjs, ESM
• If .cjs, CJS

[azz]

Also, trying to get vm.SourceTextModule working, is there a default implementation for linker in module.link(linker)? How is one supposed to return a vm.Module for a built-in, for example?

[devsnek]

@azz you can also return a module namespace object, the result of import()

[azz]

Doesn't seem to be the case?

const vm = require('vm');
(async () => {
  const esModule = new vm.SourceTextModule(`import fs from 'fs';`, {
    url: `file:///script.mjs`,
  });
  await esModule.link(spec => import(spec));
  esModule.instantiate();
  await esModule.evaluate();
})().catch(console.error);

Results in:

(node:54937) ExperimentalWarning: The ESM module loader is experimental.
Error [ERR_VM_MODULE_NOT_MODULE]: Provided module is not an instance of Module
    at ModuleWrap.<anonymous> (internal/vm/source_text_module.js:190:15)
    at processTicksAndRejections (internal/process/task_queues.js:88:5)
    at process.runNextTicks [as _tickCallback] (internal/process/task_queues.js:58:3)
    at Function.Module.runMain (internal/modules/cjs/loader.js:823:13)
    at internal/main/run_main_module.js:17:11
    at async Promise.all (index 0)
(node:54937) ExperimentalWarning: vm.SourceTextModule is an experimental feature. This feature could change at any time

[devsnek]

yeah i'm sorry, the namespace thing is only for dynamic imports. i don't know of a good solution to this off the top of my head, but you can do something like this for now:

import * as fs from 'fs'; // or `const fs = require('fs')`, doesn't really matter
new vm.SourceTextModule(
  Object.keys(fs)
    .map((x) => `export const ${x} = import.meta.mod.${x};`)
    .join('\n'), {
  initializeImportMeta(meta) {
    meta.mod = fs;
  },
});

there are still some things you'll need to take care of, like simulating live bindings.

node/lib/internal/bootstrap/loaders.js

Lines 233 to 288 in b9f1e57

	// Provide named exports for all builtin libraries so that the libraries
	// may be imported in a nicer way for ESM users. The default export is left
	// as the entire namespace (module.exports) and wrapped in a proxy such
	// that APMs and other behavior are still left intact.
	NativeModule.prototype.proxifyExports = function() {
	this.exportKeys = Object.keys(this.exports);
	const update = (property, value) => {
	if (this.reflect !== undefined &&
	ObjectPrototype.hasOwnProperty(this.reflect.exports, property))
	this.reflect.exports[property].set(value);
	};
	const handler = {
	__proto__: null,
	defineProperty: (target, prop, descriptor) => {
	// Use `Object.defineProperty` instead of `Reflect.defineProperty`
	// to throw the appropriate error if something goes wrong.
	Object.defineProperty(target, prop, descriptor);
	if (typeof descriptor.get === 'function' &&
	!Reflect.has(handler, 'get')) {
	handler.get = (target, prop, receiver) => {
	const value = Reflect.get(target, prop, receiver);
	if (ObjectPrototype.hasOwnProperty(target, prop))
	update(prop, value);
	return value;
	};
	}
	update(prop, getOwn(target, prop));
	return true;
	},
	deleteProperty: (target, prop) => {
	if (Reflect.deleteProperty(target, prop)) {
	update(prop, undefined);
	return true;
	}
	return false;
	},
	set: (target, prop, value, receiver) => {
	const descriptor = Reflect.getOwnPropertyDescriptor(target, prop);
	if (Reflect.set(target, prop, value, receiver)) {
	if (descriptor && typeof descriptor.set === 'function') {
	for (const key of this.exportKeys) {
	update(key, getOwn(target, key, receiver));
	}
	} else {
	update(prop, getOwn(target, prop, receiver));
	}
	return true;
	}
	return false;
	}
	};
	this.exports = new Proxy(this.exports, handler);
	};

node/lib/internal/modules/esm/translators.js

Lines 90 to 107 in b9f1e57

	translators.set('builtin', async function builtinStrategy(url) {
	debug(`Translating BuiltinModule ${url}`);
	// Slice 'node:' scheme
	const id = url.slice(5);
	const module = NativeModule.map.get(id);
	if (!module) {
	throw new ERR_UNKNOWN_BUILTIN_MODULE(id);
	}
	module.compileForPublicLoader(true);
	return createDynamicModule(
	[...module.exportKeys, 'default'], url, (reflect) => {
	debug(`Loading BuiltinModule ${url}`);
	module.reflect = reflect;
	for (const key of module.exportKeys)
	reflect.exports[key].set(module.exports[key]);
	reflect.exports.default.set(module.exports);
	});
	});

[azz]

Looks like vm.ScriptTextModule is a very low-level API, which is great, but not really what we're after. We want a batteries-included API (like vm.Script) that is effectively running node file.mjs except provides:

• Ability to intercept and potentially rewire imports, falling back to a default.
• Ability to run the module in a new context. (Doesn't have to be vm.createContext, just need to define the global this.)

Would there be any interest in providing this? Maybe called new vm.Module?

Basically we want to write something like this:

async function importModule(specifier, parentURL, defaultImport) {
  return defaultImport(specifier, parentURL);
}

const context = vm.createContext();

if (inputIsModule(filePath)) {
  let mod = new vm.Module(code, { filename: filePath, importModule });
  await mod.runInContext(context);
} else {
  let script = new vm.Script(code, { filename: filePath })
  script.runInContext(context);
}

[devsnek]

@azz vm.Script and vm.SourceTextModule are the same "level", its just that you need to do more things to run a module than to run a script.

It seems like you want us to expose node's internal module loader, more than a specific vm api, is that right? There are definitely changes we can still make to the SourceTextModule api, that's why it's experimental, but exposing node's internal module loader is kind of no-no territory until we finish designing it.

we want to write something like this

who is "we" and what is the specific use case here? knowing that will help me provide more useful information.

[azz]

Ah sorry, I didn't explicitly mention (there's an issue reference above). The context is that we're trying to determine if Jest can run tests on files that are ES modules without first transpiling them to CJS via Babel. With CJS, Jest intercepts require() to implement module mocking, and provides a context so it can install fake timers, globals, etc.

Some relevant files:
https://github.com/facebook/jest/blob/master/packages/jest-environment-node/src/index.ts
https://github.com/facebook/jest/blob/master/packages/jest-transform/src/ScriptTransformer.ts