Running a .mjs-script through symlink with early exit creates copy of the script

[ath88]

To follow up on #579 - I don't know why it was considered completed, as there is no resolution mentioned in the issue.

I still experience this issue with zx 8.2.4.

To replicate, make a file called test.mjs with one line:

process.exit();

.. then make a symlink to the file with ln test.mjs test, and run with zx test.

You'll start seeing a copy of test.mjs for every execution, showing up in the directory with names like:

• test-zz5yo0hugt.mjs
• test-ysdh5sy58g.mjs

[ath88]

After some digging, I notice that its a combination of a few pieces of code. Namely these:

• https://github.com/google/zx/blob/8.2.4/src/cli.ts#L186

  if (ext === '') {
    const tmpFilename = fs.existsSync(`${filepath}.mjs`)
      ? `${base}-${randomId()}.mjs`
      : `${base}.mjs`

    return writeAndImport(
      await fs.readFile(filepath),
      path.join(dir, tmpFilename),
      origin
    )
  }

Here - if there is no extension, we make a copy, now with an extension, and with a random ID in my case, to avoid collisions.

• https://github.com/google/zx/blob/8.2.4/src/cli.ts#L168

export async function writeAndImport(
  script: string | Buffer,
  filepath: string,
  origin = filepath
) {
  await fs.writeFile(filepath, script.toString())
  try {
    await importPath(filepath, origin)
  } finally {
    await fs.rm(filepath)
  }
}

Here - we write the new filename, and once we have executed the script, we remove the file again with a finally block.

• https://github.com/google/zx/blob/8.2.4/src/cli.ts#L209

  await import(url.pathToFileURL(filepath).toString())

Here there actual execution happens. Though - if this executes a process.exit() we never reach the finally block and the temporary file is never deleted.

A fix - in my case - could be to execute the file without the extension, which works just fine. Removing the first block of code above does it. I don't know why the block was added in the first place - I don't see any comments or information in the git commit message.

However - there is still an issue with .md files executing process.exit(). Also - should a finally block be added elsewhere, this will still not be executed.

Another solution would be to execute the script in another thread, so we avoid having process.exit() terminate the parent process. I bet there will be another host of issues cropping up by doing this.

Feedback will be appreciated. :)

[antongolub]

* Thinking about replacing randomId with some sort of file hash...

[ath88]

* Thinking about replacing randomId with some sort of file hash...

At least there will only be one file then, but it still doesn't fix the issue, right?

[antongolub]

@ath88,

Check this out: v8.2.4-dev.c8ca866

[ath88]

@ath88,

Check this out: v8.2.4-dev.c8ca866

Works as intended. :) Thanks!