Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

loader, docs, test: named exports from commonjs modules #16675

Closed
wants to merge 13 commits into from
Closed
29 changes: 25 additions & 4 deletions doc/api/esm.md
Original file line number Diff line number Diff line change
Expand Up @@ -83,12 +83,24 @@ All CommonJS, JSON, and C++ modules can be used with `import`.
Modules loaded this way will only be loaded once, even if their query
or fragment string differs between `import` statements.

When loaded via `import` these modules will provide a single `default` export
representing the value of `module.exports` at the time they finished evaluating.
JSON and C++ addon modules will provide a single `default` export representing
the value of `module.exports` at the time they finish evaluating.

CommonJS modules, when imported, will be handled in one of two ways. By default
they will provide a single `default` export representing the value of
`module.exports` at the time they finish evaluating.
CJS modules may also provide a boolean `@@esModuleInterop` or `__esModule`
export indicating that the enumerable keys of `module.exports` should be used
as named exports.
In both cases, this should be thought of like a "snapshot" of the exports at
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this a snapshot of the values, or just a snapshot of the names?

The latter is necessary, but the former may not be.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I thought just saying exports was ok since its both the names and the values

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Right - I’m saying that there’s no need for the values to be snapshotted; just the names.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

technically it isn't a snapshot, it's just a useful term to describe how it becomes static when assigned in reflection with es imports.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It’s pretty important to be precise here :-) i think “a snapshot of the names of the exports”, and indicating that if the values are updated, the resulting imports will update as well (a requirement for APM-like use cases, i understand)

the time of importing; asynchronously modifying `module.exports` will not
affect the values of the exports. Builtin libraries are provided with named
exports as if they were using `@@esModuleInterop`.


```js
import fs from 'fs';
fs.readFile('./foo.txt', (err, body) => {
import { readFile } from 'fs';
readFile('./foo.txt', (err, body) => {
if (err) {
console.error(err);
} else {
Expand All @@ -97,6 +109,15 @@ fs.readFile('./foo.txt', (err, body) => {
});
```

```js
// main.mjs
import { part } from './other.js';

// other.js
exports.part = () => {};
exports[Symbol.for('esModuleInterop')] = true;
```

## Loader hooks

<!-- type=misc -->
Expand Down
33 changes: 25 additions & 8 deletions lib/internal/loader/ModuleRequest.js
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,8 @@ const search = require('internal/loader/search');
const asyncReadFile = require('util').promisify(require('fs').readFile);
const debug = require('util').debuglog('esm');

const esModuleInterop = Symbol.for('esModuleInterop');

const realpathCache = new Map();

const loaders = new Map();
Expand All @@ -36,24 +38,39 @@ loaders.set('esm', async (url) => {

// Strategy for loading a node-style CommonJS module
loaders.set('cjs', async (url) => {
return createDynamicModule(['default'], url, (reflect) => {
debug(`Loading CJSModule ${url}`);
const CJSModule = require('module');
const pathname = internalURLModule.getPathFromURL(new URL(url));
CJSModule._load(pathname);
debug(`Loading CJSModule ${url}`);
const CJSModule = require('module');
const pathname = internalURLModule.getPathFromURL(new URL(url));
const exports = CJSModule._load(pathname);
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This changes CJS to always evaluate prior to linking ESM which reorders imports in odd ways

const es = exports[esModuleInterop] !== undefined ?
exports[esModuleInterop] : exports.__esModule;
const keys = es ? Object.keys(exports) : ['default'];
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Would Object.getOwnPropertyNames(exports) make more sense?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

i chose to use Object.keys so that it only exports enumerable properties. (it says so in the esm doc)

return createDynamicModule(keys, url, (reflect) => {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks like createDynamicModule would throw if one of the exports was named executor because there would be duplicated exports? Seems like ideally it would generate an executor name that wouldn't conflict, or at least make it less likely to conflict.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It maps names to safeguard against this already:

${ArrayJoin(ArrayMap(names, (name) => `export let $${name};`), '\n')}

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah so it does, my mistake. Missed the $ on there.

if (es) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is branching like this faster? Otherwise seems like the else branch does exactly what the if branch does, just more generally.

Copy link
Member Author

@devsnek devsnek Nov 7, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

i don't understand what you mean, those two blocks do different things

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sign, nevermind, just another case of misreading.

for (var i = 0; i < keys.length; i++)
reflect.exports[keys[i]].set(exports[keys[i]]);
} else {
reflect.exports.default.set(exports);
}
});
});

// Strategy for loading a node builtin CommonJS module that isn't
// through normal resolution
loaders.set('builtin', async (url) => {
return createDynamicModule(['default'], url, (reflect) => {
debug(`Loading BuiltinModule ${url}`);
const exports = NativeModule.require(url.substr(5));
debug(`Loading BuiltinModule ${url}`);
const exports = NativeModule.require(url.substr(5));
const keys = Object.keys(exports);
return createDynamicModule(['default', ...keys], url, (reflect) => {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's not documented that builtin modules still have a default export.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Perhaps for these native module keys we should add a filter - Object.keys(exports).filter(name => name.startsWith('_') === false)? This would avoid private keys such as import { _makeLong } from 'path' being on the public API that will end up in the list for type hinting systems such as TypeScript.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please do not; _ may be used to convey hopeful privacy, but it’s always fully public.

reflect.exports.default.set(exports);
for (var i = 0; i < keys.length; i++)
reflect.exports[keys[i]].set(exports[keys[i]]);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

One of the worries with doing this for native modules was that any updates to native module exports won't be seen. Specifically I think this was an issue when considering APM use cases that monkey patch any native internals.

That said, if such issues do come up, setter detection on native modules could possibly be configured to update named exports.

});
});

// Strategy for loading a native addon module
// Named exports will not be parsed from these - see
// https://github.com/nodejs/abi-stable-node/issues/256#issuecomment-325138872
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fwiw, another thing I’d like to see (not necessarily in this PR) is keeping parity between existing addons and CJS modules. I agree with the linked comment in that we don’t need any extra wiring for ESM, at least for now, but I don’t see any good reason to let wrapping for CJS and native addons diverge.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

i read this as napi wanting to keep a single export point such that they don't need to worry about if its default or module.exports or whatever else it may be, and changing how that gets exposed on the node side would be kinda weird imo

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It’s certainly not that important for addons, but … if require('foo') and require('foo.node') do the same thing, then I think import 'foo'; and import 'foo.node'; should also do the same thing

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@addaleax the foo.node package can create an ESM module wrapper that sets up named imports.

loaders.set('addon', async (url) => {
const ctx = createDynamicModule(['default'], url, (reflect) => {
debug(`Loading NativeModule ${url}`);
Expand Down
2 changes: 2 additions & 0 deletions test/es-module/es-module.status
Original file line number Diff line number Diff line change
Expand Up @@ -5,3 +5,5 @@ prefix es-module
# sample-test : PASS,FLAKY

[true] # This section applies to all platforms

test-esm-esmoduleinterop-override : FAIL
9 changes: 9 additions & 0 deletions test/es-module/test-esm-cjs-esmodule.mjs
Original file line number Diff line number Diff line change
@@ -0,0 +1,9 @@
// Flags: --experimental-modules
/* eslint-disable required-modules */

import assert from 'assert';
import eightyfour, { fourtytwo } from
'../fixtures/es-module-loaders/babel-to-esm.js';

assert.strictEqual(eightyfour, 84);
assert.strictEqual(fourtytwo, 42);
6 changes: 6 additions & 0 deletions test/es-module/test-esm-esmoduleinterop-override.mjs
Original file line number Diff line number Diff line change
@@ -0,0 +1,6 @@
// Flags: --experimental-modules
/* eslint-disable required-modules */

// eslint-disable-next-line no-unused-vars
import eightyfour, { fourtytwo }
from '../fixtures/es-module-loaders/babel-to-esm-override.js';
11 changes: 9 additions & 2 deletions test/es-module/test-esm-namespace.mjs
Original file line number Diff line number Diff line change
@@ -1,7 +1,14 @@
// Flags: --experimental-modules
/* eslint-disable required-modules */

import * as fs from 'fs';
import assert from 'assert';
import fs, { readFile } from 'fs';
import main, { named } from
'../fixtures/es-module-loaders/cjs-to-es-namespace.js';

assert.deepStrictEqual(Object.keys(fs), ['default']);
assert(fs);
assert(fs.readFile);
assert.strictEqual(fs.readFile, readFile);

assert.strictEqual(main, 'default');
assert.strictEqual(named, 'named');
8 changes: 8 additions & 0 deletions test/es-module/test-reserved-keywords.mjs
Original file line number Diff line number Diff line change
@@ -0,0 +1,8 @@
// Flags: --experimental-modules
/* eslint-disable required-modules */

import assert from 'assert';
import { enum as e } from
'../fixtures/es-module-loaders/reserved-keywords.js';

assert(e);
18 changes: 18 additions & 0 deletions test/fixtures/es-module-loaders/babel-to-esm-override.js
Original file line number Diff line number Diff line change
@@ -0,0 +1,18 @@
"use strict";

/*
created by babel with es2015 preset
```
export const fourtytwo = 42;
export default 84;
```
*/

Object.defineProperty(exports, "__esModule", {
value: true
});
var fourtytwo = exports.fourtytwo = 42;
exports.default = 84;

// added after babel compile
exports[Symbol.for('esModuleInterop')] = false
15 changes: 15 additions & 0 deletions test/fixtures/es-module-loaders/babel-to-esm.js
Original file line number Diff line number Diff line change
@@ -0,0 +1,15 @@
"use strict";

/*
created by babel with es2015 preset
```
export const fourtytwo = 42;
export default 84;
```
*/

Object.defineProperty(exports, "__esModule", {
value: true
});
var fourtytwo = exports.fourtytwo = 42;
exports.default = 84;
4 changes: 4 additions & 0 deletions test/fixtures/es-module-loaders/cjs-to-es-namespace.js
Original file line number Diff line number Diff line change
@@ -0,0 +1,4 @@
exports.named = 'named';
exports.default = 'default';

exports[Symbol.for('esModuleInterop')] = true;
6 changes: 6 additions & 0 deletions test/fixtures/es-module-loaders/reserved-keywords.js
Original file line number Diff line number Diff line change
@@ -0,0 +1,6 @@
module.exports = {
enum: 'enum',
class: 'class',
delete: 'delete',
[Symbol.for('esModuleInterop')]: true,
Copy link
Member

@jdalton jdalton Nov 7, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Babel folks: Any opinions on the symbol name esModuleInterop?
@hzoo @danez @loganfsmyth

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

there's no need to force camelCase. ES Module Interop would work too.

Copy link
Member Author

@devsnek devsnek Nov 7, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

most symbols i see throughout the js world use camelCase or PascalCase, and __esModule is still supported (as it says in the docs)

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Variable names do; Symbols haven't been around long enough to have an established convention around that. The string in Symbol() and Symbol.for() is a description - there's zero reason it needs to be limited to being an identifier.

Copy link
Member

@jdalton jdalton Nov 7, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I did not see the camel case debate coming... was thinking more along the lines of whether it should be Symbol.for('esModule') (so just remove the double underbar), but sure. I also wanted to pull Babel folks into this thread since it's super relevant to them.

Node is currently a bit all over the place with its symbol labels. Examples:

Symbol('util.promisify.custom')
Symbol('customPromisifyArgs')

I'm curious if Babel is transitioning to a symbol key as well and if they already have a name.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Definitely appreciate the ping. I personally don't see that it'd be feasible for us to use a Symbol because the code could be running on a platform that doesn't support Symbols, assuming I'm following all this properly, but it probably does make sense to use a Symbol in the long run for loaders, assuming Symbol support is known to exist.

};