sass-site/source/blog/019-request-for-comments-module-system-proposal.html.md
Natalie Weizenbaum 193a124050 Move the blog onto sass-lang.com
Logdown seems to be basically unmaintained, its servers are
unreliable, and it's been causing some mixed-content errors lately.
This moves all blog posts to sass-lang.com itself; I'll set up
redirects from the blog as best I can once this lands.

Closes #401
Closes #402
Closes #403
2019-12-18 16:00:40 -08:00

16 KiB

title author date
Request For Comments: Module System Natalie Weizenbaum 2018-11-27 13:10 PST

Many of the most frequently-requested features for Sass have to do with its imports. The import system that we've had since the very early releases of Sass is, to put it simply, not great. It does little more than textually include one Sass file in another, which makes it hard to keep track of where mixins, functions, and variables were defined and hard to be sure that any new additions won't happen to conflict with something elsewhere in the project. To make matters worse, it overlaps with CSS's built-in @import rule, which forces us to have a bunch of heuristics to decide which is which.

Because of these problems and others, we've wanted to do a full overhaul of the way Sass files relate to one another for a long time. Over the last few years, I've been working with the Sass core team and Sass framework maintainers to create a proposal for a module system that's fit to replace @import. That proposal is now in a place that the core team is pretty happy with, at least as a starting point, so we want to open it up for community feedback.

If you want to read the full proposal, it's available on GitHub. Feel free to file issues for any feedback you have. The main body of the proposal is written as a spec, so it's very detailed, but the Goals, Summary, and FAQ sections (reproduced below) should be accessible to anyone familiar with Sass.

Goals

High-Level

These are the philosophical design goals for the module system as a whole. While they don't uniquely specify a system, they do represent the underlying motivations behind many of the lower-level design decisions.

  • Locality. The module system should make it possible to understand a Sass file by looking only at that file. An important aspect of this is that names in the file should be resolved based on the contents of the file rather than the global state of the compilation. This also applies to authoring: an author should be able to be confident that a name is safe to use as long as it doesn't conflict with any name visible in the file.

  • Encapsulation. The module system should allow authors, particularly library authors, to choose what API they expose. They should be able to define entities for internal use without making those entities available for external users to access or modify. The organization of a library's implementation into files should be flexible enough to change without changing the user-visible API.

  • Configuration. Sass is unusual among languages in that its design leads to the use of files whose entire purpose is to produce side effects—specifically, to emit CSS. There's also a broader class of libraries that may not emit CSS directly, but do define configuration variables that are used in computations, including computation of other top-level variables' values. The module system should allow the user to flexibly use and configure modules with side-effects.

Low-Level

These are goals that are based less on philosophy than on practicality. For the most part, they're derived from user feedback that we've collected about @import over the years.

  • Import once. Because @import is a literal textual inclusion, multiple @imports of the same Sass file within the scope of a compilation will compile and run that file multiple times. At best this hurts compilation time for little benefit, and it can also contribute to bloated CSS output when the styles themselves are duplicated. The new module system should only compile a file once.

  • Backwards compatibility. We want to make it as easy as possible for people to migrate to the new module system, and that means making it work in conjunction with existing stylesheets that use @import. Existing stylesheets that only use @import should have identical importing behavior to earlier versions of Sass, and stylesheets should be able to change parts to @use without changing the whole thing at once.

Non-Goals

These are potential goals that we have explicitly decided to avoid pursuing as part of this proposal for various reasons. Some of them may be on the table for future work, but we don't consider them to be blocking the module system.

  • Dynamic imports. Allowing the path to a module to be defined dynamically, whether by including variables or including it in a conditional block, moves away from being declarative. In addition to making stylesheets harder to read, this makes any sort of static analysis more difficult (and actually impossible in the general case). It also limits the possibility of future implementation optimizations.

  • Importing multiple files at once. In addition to the long-standing reason that this hasn't been supported—that it opens authors up to sneaky and difficult-to-debug ordering bugs—this violates the principle of locality by obfuscating which files are imported and thus where names come from.

  • Extend-only imports. The idea of importing a file so that the CSS it generates isn't emitted unless it's @extended is cool, but it's also a lot of extra work. This is the most likely feature to end up in a future release, but it's not central enough to include in the initial module system.

  • Context-independent modules. It's tempting to try to make the loaded form of a module, including the CSS it generates and the resolved values of all its variables, totally independent of the entrypoint that cause it to be loaded. This would make it possible to share loaded modules across multiple compilations and potentially even serialize them to the filesystem for incremental compilation.

    However, it's not feasible in practice. Modules that generate CSS almost always do so based on some configuration, which may be changed by different entrypoints rendering caching useless. What's more, multiple modules may depend on the same shared module, and one may modify its configuration before the other uses it. Forbidding this case in general would effectively amount to forbidding modules from generating CSS based on variables.

    Fortunately, implementations have a lot of leeway to cache information that the can statically determine to be context-independent, including source trees and potentially even constant-folded variable values and CSS trees. Full context independence isn't likely to provide much value in addition to that.

  • Increased strictness. Large teams with many people often want stricter rules around how Sass stylesheets are written, to enforce best practices and quickly catch mistakes. It's tempting to use a new module system as a lever to push strictness further; for example, we could make it harder to have partials directly generate CSS, or we could decline to move functions we'd prefer people avoid to the new built-in modules.

    As tempting as it is, though, we want to make all existing use-cases as easy as possible in the new system, even if we think they should be avoided. This module system is already a major departure from the existing behavior, and will require a substantial amount of work from Sass users to support. We want to make this transition as easy as possible, and part of that is avoiding adding any unnecessary hoops users have to jump through to get their existing stylesheets working in the new module system.

    Once @use is thoroughly adopted in the ecosystem, we can start thinking about increased strictness in the form of lints or TypeScript-style --strict-* flags.

  • Code splitting. The ability to split monolithic CSS into separate chunks that can be served lazily is important for maintaining quick load times for very large applications. However, it's orthogonal to the problems that this module system is trying to solve. This system is primarily concerned with scoping Sass APIs (mixins, functions, and placeholders) rather than declaring dependencies between chunks of generated CSS.

    We believe that this module system can work in concert with external code-splitting systems. For example, the module system can be used to load libraries that are used to style individual components, each of which is compiled to its own CSS file. These CSS files could then declare dependencies on one another using special comments or custom at-rules and be stitched together by a code-splitting post-processor.

Summary

This proposal adds two at-rules, @use and @forward, which may only appear at the top level of stylesheets before any rules (other than @charset). Together, they're intended to completely replace @import, which will eventually be deprecated and even more eventually removed from the language.

@use

@use makes CSS, variables, mixins, and functions from another stylesheet accessible in the current stylesheet. By default, variables, mixins, and functions are available in a namespace based on the basename of the URL.

@use "bootstrap";

.element {
  @include bootstrap.float-left;
}

In addition to namespacing, there are a few important differences between @use and @import:

  • @use only executes a stylesheet and includes its CSS once, no matter how many times that stylesheet is used.
  • @use only makes names available in the current stylesheet, as opposed to globally.
  • Members whose names begin with - or _ are private to the current stylesheet with @use.
  • If a stylesheet includes @extend, that extension is only applied to stylesheets it imports, not stylesheets that import it.

Note that placeholder selectors are not namespaced, but they do respect privacy.

Controlling Namespaces

Although a @use rule's default namespace is determined by the basename of its URL, it can also be set explicitly using as.

@use "bootstrap" as b;

.element {
  @include b.float-left;
}

The special construct as * can also be used to include everything in the top-level namespace. Note that if multiple modules expose members with the same name and are used with as *, Sass will produce an error.

@use "bootstrap" as *;

.element {
  @include float-left;
}

Configuring Libraries

With @import, libraries are often configured by setting global variables that override !default variables defined by those libraries. Because variables are no longer global with @use, it supports a more explicit way of configuring libraries: the with clause.

// bootstrap.scss
$paragraph-margin-bottom: 1rem !default;

p {
  margin-top: 0;
  margin-bottom: $paragraph-margin-bottom;
}
@use "bootstrap" with (
  $paragraph-margin-bottom: 1.2rem
);

This sets bootstrap's $paragraph-margin-bottom variable to 1.2rem before evaluating it. The with clause only allows variables defined in (or forwarded by) the module being imported, and only if they're defined with !default, so users are protected against typos.

@forward

The @forward rule includes another module's variables, mixins, and functions as part of the API exposed by the current module, without making them visible to code within the current module. It allows library authors to be able to split up their library among many different source files without sacrificing locality within those files. Unlike @use, forward doesn't add any namespaces to names.

// bootstrap.scss
@forward "functions";
@forward "variables";
@forward "mixins";

Visibility Controls

A @forward rule can choose to show only specific names:

@forward "functions" show color-yiq;

It can also hide names that are intended to be library-private:

@forward "functions" hide assert-ascending;

Extra Prefixing

If you forward a child module through an all-in-one module, you may want to add some manual namespacing to that module. You can do what with the as clause, which adds a prefix to every member name that's forwarded:

// material/_index.scss
@forward "theme" as theme-*;

This way users can use the all-in-one module with well-scoped names for theme variables:

@use "material" with ($theme-primary: blue);

or they can use the child module with simpler names:

@use "material/theme" with ($primary: blue);

@import Compatibility

The Sass ecosystem won't switch to @use overnight, so in the meantime it needs to interoperate well with @import. This is supported in both directions:

  • When a file that contains @imports is @used, everything in its global namespace is treated as a single module. This module's members are then referred to using its namespace as normal.

  • When a file that contains @uses is @imported, everything in its public API is added to the importing stylesheet's global scope. This allows a library to control what specific names it exports, even for users who @import it rather than @use it.

In order to allow libraries to maintain their existing @import-oriented API, with explicit namespacing where necessary, this proposal also adds support for files that are only visible to @import, not to @use. They're written "file.import.scss", and imported when the user writes @import "file".

Built-In Modules

The new module system will also add seven built-in modules: math, color, string, list, map, selector, and meta. These will hold all the existing built-in Sass functions. Because these modules will (typically) be imported with a namespace, it will be much easier to use Sass functions without running into conflicts with plain CSS functions.

This in turn will make it much safer for Sass to add new functions. We expect to add a number of convenience functions to these modules in the future.

meta.load-css()

This proposal also adds a new built-in mixin, meta.load-css($url, $with: ()). This mixin dynamically loads the module with the given URL and includes its CSS (although its functions, variables, and mixins are not made available). This is a replacement for nested imports, and it helps address some use-cases of dynamic imports without many of the problems that would arise if new members could be loaded dynamically.

Frequently Asked Questions

  • Why this privacy model? We considered a number of models for declaring members to be private, including a JS-like model where only members that were explicitly exported from a module were visible and a C#-like model with an explicit @private keyword. These models involve a lot more boilerplate, though, and they work particularly poorly for placeholder selectors where privacy may be mixed within a single style rule. Name-based privacy also provides a degree of compatibility with conventions libraries are already using.

  • Can I make a member library-private? There's no language-level notion of a "library", so library-privacy isn't built in either. However, members used by one module aren't automatically visible to downstream modules. If a module isn't @forwarded through a library's main stylesheet, it won't be visible to downstream consumers and thus is effectively library-private.

    As a convention, we recommend that libraries write library-private stylesheets that aren't intended to be used directly by their users in a directory named src.

  • How do I make my library configurable? If you have a large library made up of many source files that all share some core !default-based configuration, we recommend that you define that configuration in a file that gets forwarded from your library's entrypoint and used by your library's files. For example:

// bootstrap.scss
@forward "variables";
@use "reboot";
// _variables.scss
$paragraph-margin-bottom: 1rem !default;
// _reboot.scss
@use "variables" as *;

p {
  margin-top: 0;
  margin-bottom: $paragraph-margin-bottom;
}
// User's stylesheet
@use "bootstrap" with (
  $paragraph-margin-bottom: 1.2rem
);

Sending Feedback

This is still just a proposal. We're pretty happy with the overall shape of the module system, but it's not at all set in stone, and anything can change with enough feedback provided by users like you. If you have opinions, please file an issue on GitHub or just tweet at @SassCSS. We'll take anything from "it looks awesome" to "it looks awful", although the more specific you can be the more information we have to work with!