Replace clean-webpack-plugin with the core output cleaning feature

[stof]

Closes #1289

The options configurable by the configurator callback passed as second argument of cleanupOutputBeforeBuild are not the same anymore (the dry option used in the example still exists)

[stof]

The ESLint job requires #1311 to pass, because the deprecated valid-jsdoc rule does not support the import() syntax in types.

[Kocal]

Thank you for working on this!

[Kocal]

Can we use picomatch instead?

It has 0 dependencies (where multimatch has 7 dependencies), and weight much less, see graphs:

• picomatch: https://npmgraph.js.org/?q=picomatch#select=exact%3Apicomatch%404.0.2
• mulitmatch: https://npmgraph.js.org/?q=multimatch%406.0.0

picomatch is also compatible with CJS modules.

Thanks!

[stof]

no we cannot (easily). The feature I'm using is precisely the ability to pass multiple patterns including negative patterns (due to the way our cleanupOutputBeforeBuild API looks like), which is the main point of multimatch.

If we decide to break compatibility a lot more by removing support for the first argument of our current cleanupOutputBeforeBuild, I could do it without any dependency.

[stof]

and picomatch might switch to ESM in the future as well (there is a PR opened about that). So just using picomatch won't let us use CJS forever (my expectation is that more and more packages will migrate to ESM)

[stof]

note that this still reduces our dependencies compared to clean-webpack-plugin

[Kocal]

What about breaking the cleaning feature and let the user decide what it wants to keep (through output.clean.keep if I remember)?
We could directly pass its configuration to Webpack, it will be easier for us, no need for extra work nor dependencies.
WDYT ?

[stof]

We still need the config to keep the manifest.json for the dev-server (unless the issue mentioned in the code comment is not relevant anymore).

If we can remove that part, I'm fine with doing a bigger BC break dropping the first argument (and letting users configure keep through the callback if needed).

[stof]

However, as I'm not using the dev-server myself (I'm using the watch mode), I don't have a setup to test this.

[Kocal]

I can give it a try tomorrow

[Kocal]

Good news, the option output.clean.keep() seems totally ignored in dev-server:

• Just to be sure, I've removed by public/build folder,
• Commented cleanedPaths.push('!manifest.json')
• Ran the dev-server

And file public/build/manifest.json is present, with good URLs, HMR works too, etc...

So we can remove cleanedPaths.push('!manifest.json') and go for a bigger BC :)

[stof]

I updated the PR with the simpler implementation.

[Kocal]

Thanks for your modifications!

I've tried your branch locally and everything worked fine, as expected I had to update .cleanupOutputBeforeBuild(['**/*', '!amp/**']) to .cleanupOutputBeforeBuild() (because of argument check, which works fine).

We can merge after a rebase IMO

PS: in the next week, I will try to add E2E tests for the dev-server, to check if it correctly boots, if page is updated after a modification, etc... that will ease this kind of changes.