I've been fighting with CSS long enough in production now that I don’t enjoy working with it anymore unfortunately, so at least for now this my formal farewell! *
CSS has been designed to style documents, not applications. Anyone that has tried to use CSS at scale knows how it requires too many restrictions and rules to make it work. In short, it's painful and not really a pleasure.
CSS selectors are the main problem we have to face with CSS. It's like global mutable variables. Because of how selectors works, it's very hard to know if removing a piece of CSS is ok or not. There is still no way to write trustable scoped CSS. So yeah, you can use BEM, CSS modules and other fancy tools & methodologies, but at the end of the day, those are just patches that require strict methodologies and tooling. Not solution for the future.
Many have already talked & written about the limits of CSS for apps and the most accurate reference to me is still one of the first (from 2014).
To refresh your memory, here is the list of problems that you encounter when working with CSS on apps
No local variables.
No dead code elimination.
No code minification.
No sharing of constants.
The way CSS is currently evolving won't fix anything.
Meanwhile, CSS-in-JS is becoming clearly more and more popular as it can help to avoid all those problems and also has some benefits:
Generates the minimum required CSS.
Offers good runtime performance.
Supports static and dynamic styles.
Helps to pre-render critical CSS.
At the end, at least for now on the web, we still use CSS, but as a low-level primitive (like React is using the DOM). Just like any code we write ends up being compiled to a language that our computer will understand.
When relying on a good abstraction, you can write code similar to CSS (without selectors) that will work in other places, thanks to Yoga. Yoga is the solution used in React-Native and some others engines. Yoga is just an example and you can probably find other alternatives like flex (a Yoga reimplementation in Reason).
In addition to previous benefits, using this kind of solution also adds
Simple API and expressive subset of CSS.
Support for RTL layouts.
Easy pre-rendering of critical CSS (if your target compilation is the web)
Statically type-checked styles ♥️.
For those that are going to say something like
But if you are smart, you will use cascading to your advantage
Before saying anything like this, be sure that:
you haven't shipped one-shot projects only without having to maintain those on the long run
you actually had more than 1000 CSS selectors in production for a single project
Using CSS for small projects is probably ok, but it's not for medium and large websites/apps. Please watch and read about the motivations of CSS-in-JS until we are convinced. And if you are not... 🙃
For these reasons, I will only use CSS when I will use HTML directly, so that means 2 times per year when I do a real quick landing page.
I really believe that CSS will become a low level primitive, a target for compilation, and nothing more in the near future. And for a far future, it may even become obsolete and not used.
This is why I am deprecating cssnext as of today.
(If you ask me "what about media queries and pseudo elements?", I don't need those. I think media queries are not good enough (element queries could be a thing I would use tho) and pseudo elements are clearly oriented for document styling. See also react-native-web Style FAQ.
I also look forward to react-native-dom.