Add support for more complex selectors for sideEffects

[TheLarkInn]

Do you want to request a feature or report a bug?
Feature

What is the current behavior?
webpack searched for "sideEffects": false in package.json to perform the extra optimizations for esm modules. However if false is a string, the optimization isn't detected.

We should probably (in the case that package authors might mistake this as "false" vs false) put the wrong value in, and theres no simple way to detect (as a dev) why the 'deopt' occurs.
If the current behavior is a bug, please provide the steps to reproduce.

What is the expected behavior?
"sideEffects": "false" in package.json works the same as "sideEffects": false

If this is a feature request, what is motivation or use case for changing the behavior?
Motivation described above.

Please mention other relevant information such as the browser version, Node.js version, webpack version and Operating System.
webpack 4.0.0-alpha

[codywl]

I'm not quite sure how to architect the test case(s) for this issue. However, I am looking into existing tests to see if I can reuse anything.

[sokra]

"sideEffects": "false" should mark every module as side effect free except a file named false

[codywl]

Oh, I may have completely misunderstood the issue, then. Am I just to implement the case where specifying a String instead of a Boolean value tells webpack which files are known to have effects? If so, should we allow Arrays of strings as well?

[sokra]

yep Array of string should be allowed too.

It should support glob expressions too.

"sideEffects": [
  "src/**/*.js"
  "./lib/abc.js"
]

[codywl]

Got it - I'll work on that, and inquire further if I have any difficulties.

[codywl]

I've been trying to decipher some segments of this particular Plugin instance, but I might be lacking some context. The first ambiguous thing in here (to me) is the usage of harmony, so I ran off to check out examples. As far as I can tell, this is part of how webpack keeps collisions out of imports. I can tell that it will take some time before I grasp the larger web of interaction, so I'll proceed with that assumption (unless somebody corrects me, which would be welcome anyways).

In any case, do I have access to the name of the module or the filename when normal-module-factory returns the data object? I'm assuming that's how I will be comparing it to the array of strings the user provides. It looks like I may be able to call __filename from the Node API, but it's unclear to me whether that would report what I want inside of this context.

[sokra]

https://github.com/webpack/webpack/blob/next/lib/optimize/SideEffectsFlagPlugin.js#L19

resolveData.relativePath is the relative path of the file relative to the package.json.

[codywl]

Oh, thanks for that. This will make things much easier.

[codywl]

Apologies for the gap!

Here's where I am at:

• glob expansion via glob (run against the user strings if they both contain * and there was no prior match for the current module);
• matches for relative paths.

I need to make tests for:

• arbitrary relative file paths
• glob paths of arbitrary depths and file extensions

I guess there might be some potential slowdown if there's a significantly larger glob than normal. I'll measure performance if it's wanted. Otherwise, I'll continue immersing myself to learn webpack contribution knowledge. 😃

[sokra]

Instead of using glob: Convert the glob expression into a RegExp and just test the relative path with it.

src/**/abc/*.js => /\/src\/.+\/abc\/[^\/]+\.js($|\/)/

./src/def => /^\./src\/def($|\/)/

[codywl]

Oh, interesting. I'd not thought of that. I'll convert it over accordingly!

Quick question before I go too far: How complete should the conversion be? Are we only looking for **, *.extension and *? Looks like you also provided a case for just a whole directory, so I assume that's one too. (Thanks for the help!)

[sokra]

Maybe just use https://github.com/fitzgen/glob-to-regexp

[abiduzz420]

Hi @CWelshE I want to know if you are still working on it ?

[reergymerej]

@abiduzz420 I've got a PR that needs a modification to the glob feature. It's a pretty small change, but I'm not able to follow up on it right now.

[TheLarkInn]

@reergymerej is there any notes you could provide to assist others? Happy to help find another contributor.

[reergymerej]

The PR is #6334. I just needed to switch the library doing the glob checking to minimatch and update tests. The test structure is already there. It's should be as easy as changing the library in the source and the assertions to go along with those in the PR review notes.

[jevancc]

Hi, may I take up this issue and continue what @reergymerej has done?

[reergymerej]

@jevan0307, please do. I pushed the work in progress I had. 2b45e65

[massicattaneo]

I'm trying to use the tree shaking on 2 different bundles that imports different fn from the same code.
imaging a "shared.js" like this:

export function square(x) {
    return x * x;
}

export function cube(x) {
    return x * x * x;
}

And 2 bundles 1 importing {square} and the other importing {cube} I was expecting to see the tree shake working but it happen to me that the both bundles contains the whole "share.js" exports. Am I doing something wrong?

[srq18211]

Maybe other options have a higher priority than this option .
For example :

webpack.config.js

//Bad
module.exports = {
    devtool: 'inline-source-map'
}
//Good
const env = process.env.NODE_ENV;
module.exports = {
    devtool: env == 'development' ? 'inline-source-map' : false,
}

package.json

{
  "name": "parcel",
  "sideEffects": false,
  ... ...
}