Please do not pass in the testing context - the methods in @packages/test should only provide required setup / side effects. Please check the actual assertions in the test itself.

I'd prefer reading the file list this from the tarball directly, e.g. via tar-fs.

Instead of specifying the expected bundle files manually I'd like to derive the absolute minimum from package.json entries: main: string = "index.js" and bin: string || { key: string }. This could be done via a helper fn that should live in @commitlint/test

import pkgDir from 'pkg-dir';
import {npm} from '@commitlint/test';

test('should publish required files', async t => {
  const cwd = pkgDir(__dirname);
  const expected = await npm.getPackageFiles({ cwd });
  const actual = await npm.getTarballFiles();
  t.deepEquals(actual, expected);
});

I think we could use the ignore options of tar.extract to get a file list without actually writing to disk:

const files = [];

fs.createReadStream(tarballFile)
  .pipe(zlib.createGunzip())
  .pipe(tar.extract(workDir, {
    ignore(name) {
        files.push(name);
       return true;
    }
   }))
 .once('error', err => reject(err))
 .once('finish', () => resolve());

Might be elaborate enough to factor it out to a listArchive function.

Should be obsolete if we implement listArchive

Should be obsolete if we implement listArchive

I think we can unwrap this from

try {
  // 1
} finally {
  // 2
}

to

  // 1
  // 2

now the assertions happens in the test body.

Shouldn't we put clean up code in finally clause?

try {
  // do something
finally {
  sander.rimraf(tarballFile)
  sander.rimraf(workDir)
}

I think we could use the ignore options of tar.extract to get a file list without actually writing to disk:

Sorry. Please ignore my question. We don't need writing to disk 😓

Sorry for many times...
The following code (npm pack) writes tarball file (e.g. commitlint-cli-6.0.2.tgz) to disk actually.
Should we clean-up this file in finally clause?

const tarballFilename = await execa.stdout('npm', ['pack'], {cwd});
const tarballFile = path.resolve(cwd, tarballFilename);
try {
  // do something
} finally {
  sander.unlink(tarballFile);
}

Why is this needed?

Because npm-packlist module ignores .npmignore even if it exists. 😓

We should change this to a less strict check, e.g.

// for each file we expect
t.true(includes(actual, 'some-file.js'))

I am rolling around the idea to introduce unexpected to the testing setup. Should make things like this easier.

That idea makes test code easier, because we don't need implementing code such as npm.packageFiles()! 👍
Is that idea is like the following code, correct?

  t.true(actual.includes('package.json')) //<= unnecessary? (because it is obvious)
  t.true(actual.includes('README.md'))
  t.true(actual.includes('CHANGELOG.md'))
  t.true(actual.includes('index.js'))
  t.true(actual.includes('lib/cli.js'))
  t.true(actual.includes('lib/help.js'))

[nitspick] I found typo: getArchveFiles -> getArchiveFiles.