fix: fs.readFile emit relative assets using cwd

[icyJoseph]

In this input file:

const fs = require('fs');

fs.readFile('./asset1.txt')

fs.readFile(require.resolve('./asset2.txt'))

const _basePath = __dirname;
const asset3 = 'asset3.txt';
fs.readFileSync(_basePath + '/' + asset3, 'utf8');

We don't emit for fs.readFile('./asset1.txt'), but we should:

Regardless of where the code is (i.e. in which chunk it ends up in after bundling), readFile is always relative to process.cwd() which is set by Next.js to always be the project root (otherwise process.cwd() + "./foo" wouldn't work either

[ijjk]

Are we matching behavior here? vercel/next.js#84155 seems like this is diverging

[icyJoseph]

Mmm - yeah, I have it wrong here, I should use cwd - and update the test

[icyJoseph]

if we have:

├── hello.json
├── index.js
├── lib
│   └── index.js
└── package.json

With lib/index.js as:

const fs = require("fs/promises")

fs.readFile('./hello.json', 'utf-8').then(console.log)

And index.js as:

require("./lib")

Then node index.js resolves hello.json in lib from the cwd, not the lib/index.js file. Also follows that node lib/index.js resolves from the root too.

[icyJoseph]

In this branch, running npm run build && node out/cli.js print test/unit/fs-emission/input.js won't output asset1.txt cuz it doesn't exist in the root.

If we place a dummy asset1.txt at the root and re-run, then we trace the file.

[ijjk]

I think we have two variations one with cwd at the root and once in the fixture

[icyJoseph]

Yes, both tests pass with that condition - I shared that as an example of how to test manually.

[styfle]

Great work, thanks! 🥇

[github-actions]

🎉 This PR is included in version 0.30.2 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀