Aleksey Kulikov

• I'm from Russia, Perm
• I live in Ljubljana since Jan 2015
• I like travel and hiking
• I have a wife and son

• I write web applications since 2009.
• I contributed to many open source projects like rails and component

• I developed a few successful open source libraries, like treo.js, backbone-offline, and browserify-test

• I bootstrapped Gingko App, than sold my part and relocated to Ljubljana

• I was first employee of butterfly.ai (now we are 19 people team around the world and keep growing)

Part 1.

The problem

Documents vs Web Apps

• CSS was designed for web documents:
  • separate HTML + CSS and JS files
  • stateless communication by URL
• Now we create web applications using components
  • rich, interactive UX
  • complex client-side logic and routing

Different separation of concerns

https://speakerdeck.com/didoo/let-there-be-peace-on-css

Scale

• one developer & one page - css is perfect

• many developers & 100s of components - css sucks

Why CSS does not scale for web apps?

Everything is global

• global variables are bad
• hard to reuse external libraries
  • new external library -> bunch of global variables
  • for example bootstrap.css -> 600 globals
• no isolation, easy to rewrite rule

Unspecified dependencies

• not clear how different parts of application are connected
• hard to refactor and delete legacy code

Part 2.

CSS still works

With good naming convention CSS works pretty well

https://gist.github.com/alekseykulikov/68a5d6ddae569f6d0456b0e9d603e892

http://slides.com/alekseykulikov/5-principles-for-modern-css

On a medium "scale"

• 6 developers team

• 9270 lines of CSS and 0 !important

• 0 issues with CSS during 1.5 year

1. BEM naming convention

• BEM 101| CSS-Tricks
• SUIT CSS + SUIT CSS design principles

2. CSS Preprocessor: PostCSS + CSSNext

• Future CSS syntax like variables & nesting
• Autoprefixer
• Performant and extensible

3. Flexbox

• No need for grid framework
• Good browser support

4. normalize.css

• solid foundation for all components
• no need for frameworks
• css is really expressive language

5. No DRY

• Focus on better components
• Avoid wrong abstractions in the beginning 
• Write code that is easy to delete, not easy to extend

6. Stylelint

• stylelint + stylelint-config-standard
• stylelint-selector-bem-pattern to enforce BEM rules

* but:

• no connection with JS code
• hard to start and requires review for new developers
• a lot of repetition when refer on classes
• hard to setup rules validation on existing code base
• easy to break

*

Part 3.

CSS-in-JS

Beginning of CSS-in-JS

Christopher "vjeux" Chedeau
https://speakerdeck.com/vjeux/react-css-in-js

Facebook level of Scale

Problem with inline styles

• it's only subset of CSS
  • no @media queries
  • no pseudo-selectors, like :hover or :visited
  • no css tools, like autoperfixer
  • no custom fonts

Experimentation begins

CSS in JS techniques comparison.

aphrodite

• inline styles that work
• easy to add to existing project, because every rule marked as !important
• problems: client-side processing, 20kb size, odd syntax for native css features

styled-jsx

• works similar to <style scoped> or vue.js
• full CSS support
• minimal and performant
• problems: jsx (react) only, no nesting and custom CSS syntax

styled-components

• use ES6 template literal to define styles
• components instead of classes and tags
• dynamic properties
• problems: react only, slow (at least v1), not necessary abstraction on html tags

styletron

• atomic css
• critical rendering path optimization
• high-performance
• problems: low-level, object syntax for css declarations

Different approaches and no silver bullet

Part 4.

Ideal approach

A bit of theory

• CSS rule: selector + declaration(s)
• Declaration: key + value

key+value is fine

• no one says we need better background property
• concept of declarations is simple and robust
• many css guides focus mostly on it, leaving selector part up to developers

Selectors == Cascade

Cascade is broken

• Who can explain how cascade works?
• CSS selectors may be a good idea to traverse DOM (jquery), but not for styling.
• > (parent selector), may be the only one adequate selector, which brings the glory of old days CSS to modern web apps.

<link rel=”stylesheet” href=”file.css” >

== Sheet

• it loads text file with rules (it's hard to use this concept in components world)
• it blocks render
• for apps with client-side routing, this is not an option

How ideal styles library should look like?

• supports all CSS features and keeps flexibility and power of the language
• has 0 size and minimal initialisation overhead
• injects only actually used styles
• eliminates concept of selector
• scales in teams with different experience (not css gurus)
• connects with JS world (similar to jsx)

That's it!

injected-css

https://github.com/alekseykulikov/injected-css

Usage

Install the package:

npm install injected-css

Add "injected-css/babel" to plugins section

in your babel config:

                                     {
                                         "plugins": [
                                             "injected-css/babel"
                                         ]
                                     }

Main features

matches "ideal" points

• Write real CSS (copy-paste from stack-overflow)
• Use powerful tools of JS world
  • ES modules to manage dependencies between styles and JS components
  • Eslint to ensure, it's hard to write bad CSS
  • Variables and functions
  • Webpack 2, Hot Module Replacement, code splitting, and server side render
• Overhead is minimal - 300 byte runtime (think about inject function as styles version of require)

Demo

Path to v1.0

• implement eslint plugin
• server side rendering
• support BEM's __
• feedback from community
• experiments: atomic classes and development hints

https://github.com/alekseykulikov/injected-css/issues/1