Odd behaviour: createRequireFromPath('.')

[SMotaal]

• Version: v12.0.0
• Platform: macOS 10.14.4
• Subsystem: module

Calling createRequireFromPath('.') does not honour process.cwd() when called a second time after an effective call to process.chdir(…).

Best explained by this: https://github.com/SMotaal/esm/blob/master/tests/create-require.mjs#L8

import {createRequireFromPath} from 'module';
import {fileURLToPath} from 'url';

const specifier = './package.json';
const [scope, ...paths] = ['../../', '../', './'].map(path => fileURLToPath(new URL(path, import.meta.url)));
const sanitize = path => path.replace(scope, '‹scope›/');

for (const path of process.argv.includes('-t2') ? paths.reverse() : paths) {
  process.chdir(path);
  console.group('\nprocess.chdir(%o)', sanitize(process.cwd()));
  console.log();
  try {
    const require = createRequireFromPath('.');
    console.log(
      `createRequireFromPath(%o)\n  .resolve(%o)\n    => %o`,
      '.',
      specifier,
      sanitize(require.resolve(specifier)),
    );
  } catch (exception) {
    console.warn(
      `createRequireFromPath(%o)\n  .resolve(%o)\n    => %o`,
      '.',
      specifier,
      `${exception}`.split('\n', 1)[0],
    );
  }
  console.log();
  console.groupEnd();
}

/*******************************************************************************
 * $ node --experimental-modules esm/tests/create-require.mjs
 *
 *   process.chdir('‹scope›/esm/')
 *
 *     createRequireFromPath('.')
 *       .resolve('./package.json')
 *         => '‹scope›/esm/package.json'
 *
 *   process.chdir('‹scope›/esm/tests/')
 *
 *     createRequireFromPath('.')
 *       .resolve('./package.json')
 *         => '‹scope›/esm/package.json'
 *
 ******************************************************************************
 * $ node --experimental-modules esm/tests/create-require.mjs -t2
 *
 *   process.chdir('‹scope›/esm/tests/')
 *
 *     createRequireFromPath('.')
 *       .resolve('./package.json')
 *         => '‹scope›/esm/tests/package.json'
 *
 *   process.chdir('‹scope›/esm/')
 *
 *     createRequireFromPath('.')
 *       .resolve('./package.json')
 *         => '‹scope›/esm/tests/package.json'
 *
 ******************************************************************************/

[BridgeAR]

This is actually working as intended after looking deeper into this. To path provided is the relative path to the current file or an absolute path. AFAIK this should not honor changing the working directory as it's independent.

[MylesBorins]

I think there is a slightly more subtle bug behavior here... in that createRequireFromPath('.') seems to always use process.cwd() as the base 😅

indirection/lol.js

const {createRequireFromPath} = require('module');
const req = createRequireFromPath('.');

console.log(process.cwd());
console.log(require.resolve('./package.json'));
console.log(req.resolve('./package.json'));

$ node indirection/lol.js
/Users/mylesborins/code/makeRequireFromPath
/Users/mylesborins/code/makeRequireFromPath/indirection/package.json
/Users/mylesborins/code/makeRequireFromPath/package.json
$ cd indirection
$ node lol.js
/Users/mylesborins/code/makeRequireFromPath/indirection
/Users/mylesborins/code/makeRequireFromPath/indirection/package.json
/Users/mylesborins/code/makeRequireFromPath/indirection/package.json

edit: this is because createRequireFromPath(path) calls new Module(path) which is turn calls path.dirname(path) which when called is relative to the CWD, not the filename of the callee

[devsnek]

I don't think there should be any "callee" magic here. my intention when I originally proposed this feature was something along the lines of makeRequire(import.meta.url).

[MylesBorins]

@devsnek but as the feature is documented it doesn't really work for two reasons

from docs

const { createRequireFromPath } = require('module');
const requireUtil = createRequireFromPath('../src/utils');

// Require `../src/utils/some-tool`
requireUtil('./some-tool');

issue 1) if you create a require from a relative path it is going to be different depending on cwd(), likely unexpected

issue 2) directories don't actually work like that

• open issue: module: createRequireFromPath documentation is incorrect #23710
• open fix: module: allow passing a directory to createRequireFromPath #23818

Seems like we need to revisit some of the behavior here imho

[SMotaal]

@BridgeAR I just wanted to clarify the behaviour:

1. running node … esm/tests/create-require.mjs from '‹scope›'/

   • it calls process.chdir(`‹scope›/esm/`) first
   • it calls createRequireFromPath('.').resolve('./package.json')
   • this returns '‹scope›/esm/package.json'
   • it then calls process.chdir(`‹scope›/esm/test/`) second
   • it calls createRequireFromPath('.').resolve('./package.json')
   • this returns '‹scope›/esm/package.json' also

2. running node … esm/tests/create-require.mjs -t2 from '‹scope›'/

   • it calls process.chdir(`‹scope›/esm/test/`) first
   • it calls createRequireFromPath('.').resolve('./package.json')
   • this returns '‹scope›/esm/test/package.json'
   • it then calls process.chdir(`‹scope›/esm/`) second
   • it calls createRequireFromPath('.').resolve('./package.json')
   • this returns '‹scope›/esm/test/package.json' also

There is clearly a cached aspect which:

1. does not get initialized until first call to createRequireFromPath
2. picks up process.cwd() at that time (at least with '.' as the arg).
3. since 2 happens, then process.chdir(…) not affecting next call to a new createRequireFromPath - irrespective of why the user wanted 2 of them — is not a behaviour that is clear from a user perspective.
4. if 2 is not intended, then '.' not resolving to the current module — irrespective of this being CJS/ESM — is not a behaviour that is helpful from a user perspective.

So I was really hoping we can clarify the behaviour of '.' here.

Hope this was helpful 😄

Side note: The few times, I wanted to chdir in node, I always need to remind myself that process.chdir(…) not process.cwd(…) — ie I maybe do this once every 1-2 years and usually to test odd behaviour of resolved paths seemingly tied to process.cwd().

So, I would be favourable of module related utilities being more aware of module locations than process.cwd is my intuitive feel for moving forward here.

[SMotaal]

The cached behaviour in your opinion, does that qualify as a bug?

[MylesBorins]

Closing as I believe this has been fixed