Native CSS support

[sokra]

Current State: opt-in via experiments.css: true

Explainer (not everything is implemented yet):

• experiments.css: true enables the native css support in webpack
• This will be enabled by default in webpack 6
• There will be multiple modes depending on how you reference CSS
  • A: import "./style.css": Attaches the stylesheet to the document as side effect of importing. @import and url() are resolved.
  • B: import stylesheet from "./style.css" asset { type: "css" }: Exports the unattached stylesheet as default export. You can adopt it according to the spec. @import is not allowed (yet), but url() will be resolved.
  • C: import { main } from "./style.module.css": Like A, but renames all css classes, keyframes and css variables and exports the new names. Global selectors are only allowed with explicit :global selector prefix. @import and url() are resolved.
  • D: new URL("./style.css", import.meta.url): Gives an URL to a stylesheet including all styles. @import and url() are resolved.
  • E: import { main } from "./style.module.css": Like C, but for the node.js target. It generates a JSON module with the exports from the CSS.
• A and C will bundle all CSS of a chunk into a single CSS file (per chunk)
• splitChunks will allow to move CSS around
• output.cssFilename and output.cssChunkFilename allows to configure the filename template for css files. By default it copies filename resp. chunkFilename with extension changed to .css
• HMR will update CSS when you edit it.
• There is external CSS that will lead to @import "..." in the output css files
• There are external assets that will lead to a external url.
• class names are based on module ids, so you need to make sure that SSR and Client module ids match
  • E1: Compute module ids in a deterministic way -> DeterministicModuleIdsPlugin
  • E2: Store module ids in a file and load them on the other side -> SyncModuleIdsPlugin
• In addition to the default bindings, you can use type: "css/..." in rules to apply loader results as css. e. g. for Sass support.
• This will replace mini-css-extract-plugin and css-loader
• It will be more performant, by using a css tokenizer instead of postcss
• There is a css minimizer in production mode.

Implemented:

• A
• @import url("...");
• @import "...";
• url()
• chunking
• on demand loading
• HMR
• splitChunks
• output.cssFilename
• external type css-import
• external type asset
• :export blocks
• C (partially)
  • classes
  • ids
  • keyframes + animation
  • css variables + var()
  • var(xxx from "...")
  • var(xxx from global)
• correct css order in output file
• warnings for ordering issues
• E
• E1
• E2
• image-set()
• @import url("...") layer(test);
• @import url("...") supports(...);
• @import url("...") media query;
• move all external imports to top
• do not ignore #import and make them external
• inherit layer, supports(...) and media from parent module
• supports style resolution in package.json and test it + test support loader usage
• Implement CSS Module fetchpriority
• prefetch/preload for CSS
• external import with media/supports
• Correct publicPath for assets modules inside CSS files
• pathinfo for @import
• prefeth/preload only for CSS import() generates prefetch/preload runtime for JS
• default type: css enables CSS modules support (i.e. for CSS without CSS modules you need set type: "css/global"), which might be a bit confusing for developers, because some developer just want to use pure CSS and parsing CSS modules can affect on perf, I think we should do:
  • type: "css/auto" - enable/disable CSS modules based on path, it is css/module or css
  • type: "css" - no CSS modules and don't try to parse CSS modules things
  • type: "css/module" - CSS modules with local mode
  • type: "css/global" CSS modules with global mode
• rewrite to be align with CSS spec
• src()
• unstable ?hash generation in --webpack-main when you have multiple same modules
• improve logic in callback (more perfomance?) + better error reporting for @import
• implement /* webpackIgnore: true */ for url()/image-set()
• url() in css leaves an asset JS Module in js chunk
• :import blocks
• @value

Not implemented/tested, but Planned (ordered by priority):

• C (partially)
  • composes: xxx
  • composes: xxx from "..."
  • composes: xxx from global
• re-export for :export/@value/var()/classes and nested @import
• icss interpolation for everything
• local() and global() in declarations
• more at-rules rename in css modules
• move tests frompostcss-modules-* packages
• css/global test
• move tests from css-loader
• moove test from extract plugin
• internal refactor CSS parser + use ICSS prefix for deps
• D
• B
• CSS entry without JS file (option to do only extract without runtime + for dynamic?) and hmr
• import sheet from './styles.css' assert { type: 'css' };
• exportGlobals
• improve hashing of locals for production
• pure mode
• https://github.com/webpack/webpack/issues/17611 - allow link be before/after script
• image()
• css minimizer
• contenthashing for CSS files
• normalize media/supports/layer to lowercase?
• url("font.svg#svgFontName") created invalid filename
• support legacy browser? https://github.com/webpack/webpack/issues/16147
• wasm css tokenizer

[webpack-bot]

This issue had no activity for at least three months.

It's subject to automatic issue closing if there is no activity in the next 15 days.

[alexander-akait]

bump

[webpack-bot]

Issue was closed because of inactivity.

If you think this is still a valid issue, please file a new issue with additional information.

[alexander-akait]

/cc @TheLarkInn We have this, so I just will add more things from css-loader/etc so we can see our progress

[alexander-akait]

Broken URL - https://github.com/webpack/webpack/issues/16969#issue-1664483188

[hardfist]

@alexander-akait I'm wondering how experiments.css working with the current css-loader ecosystem,for example https://github.com/seek-oss/css-modules-typescript-loader which consume the result of css-loader, but since experiments.css handle css after loader there're no chances for other loader to get css-module intermediate result, will experiments.css expose some hooks for this scenarios?

[alexander-akait]

@alexander-akait I'm wondering how experiments.css working with the current css-loader ecosystem,for example https://github.com/seek-oss/css-modules-typescript-loader which consume the result of css-loader, but since experiments.css handle css after loader there're no chances for other loader to get css-module intermediate result, will experiments.css expose some hooks for this scenarios?

It doesn't work, css-loader and etc output a warning and do nothing. Anyway you can set type: ""javascript/auto" and css-loader will work as before.

css-modules-typescript-loader should be rewritten, it should be a plugin which takes CSS dependecies and modify generated code, it is no hard to implement, we can't adopt it and make compatibility, because css-modules-typescript-loader uses generated JS by css-loader, I don't know why it was implemented in such way, because it should be just an option for css-loader (just a custom function that postprocess exports).

[alexander-akait]

The main idea of the new CSS pipeline - avoid generate JS and CSS multiple times, like we have right now

[hardfist]

i do agree this should be implemented as plugin and no need to be compatible，how this plugin should be implemented becomes a problem，should this plugin consume css dep directly and do the same codegen logic as internal css codegen or should it just consume the intermediate result of intetnal css module processing

[alexander-akait]

@hardfist

i do agree this should be implemented as plugin and no need to be compatible，how this plugin should be implemented becomes a problem，should this plugin consume css dep directly and do the same codegen logic as internal css codegen or should it just consume the intermediate result of intetnal css module processing

Yeah, we need to think about it, maybe we can introduce extra hooks to simplify this process, let's keep this here, We wanted to finish normal CSS support firslty and then I will think about it and put in our test cases a simplified example

[hardfist]

I'm not sure whether all the options in css-loader and mini-css-extract-plugin will be supported in experiments.css, and we're migrating some applications from css-loader & mini-css-extract-plugin to experiments.css and missing some functionalities, so maybe I can track these in this issue or file separate issue?

• https://webpack.js.org/loaders/css-loader/#sourcemap user want to ignore css sourcemap but keep js sourcemap, so an separate devtool option for css may be needed

if these functionalities are accepted we're willing to implement it for webpack

[fregante]

from https://github.com/webpack/webpack/issues/17700

It's unclear how to get the old extract + css-loader setup: generate CSS files and do nothing else.

Currently it appears that experiments.css: true generates the CSS files but also tries to load them, causing this error in my environment (web extension)

Uncaught TypeError: Failed to execute 'getComputedStyle' on 'Window': parameter 1 is not of type 'Element'.
    at loadCssChunkData (refined-github.js:28644:21)

Which wasn't the case with the following configuration:

module: {
	rules: [
		{
			test: /\.css$/,
			use: [
				MiniCssExtractPlugin.loader,
				'css-loader',
			],
		},
	]
},
plugins: [
	new MiniCssExtractPlugin(),
],

[jeromehan]

getComputedStyle

@fregante This might help you https://github.com/webpack/webpack/issues/16147#issuecomment-1583612714

[fregante]

No, webpack should not run any JavaScript if I ask it to generate a CSS file, so that isn't useful for this scenario.

[Sushil642]

Dear @sokra ,

I am Sushil SIngh. I'm writing to express my interest in contributing to the ongoing enhancement of Webpack's CSS capabilities, specifically Issue #14893 . I have reviewed the provided information and believe that I possess the necessary skills to make meaningful contributions to the project.