Twin – Tailwind with the flexibility of CSS-in-JS
github.com
github.com
Makes one wonder if anyone even bothers to learn CSS nowadays. Perhaps we wouldn't have Tailwind or CSS-in-JS if more people did.
Why are there trends in software development? Are we trying to solve problems or we're trying to be trendy? Is this a clothing industry?
I would have thought we should use mature technologies that have been vetted and gone through the trough of disillusionment and onto the ramp of productivity. Proven tech stack that I can rely on gives me peace. Older the tech, the more reliable it is.
The opposite of "trendy" is what we need to popularize and make it cool. It should be cool to use a 5 year old library in JS!
It is cool to use 5 year old JS projects like React. If you look at the projects listed in the report, you see the graphs going back to 2016. Which is 5 years ago. And how they have been popular for 5 years.
Not haphazardly monkey patching anything you see, add emojis to make it look cute and get 30k stars on GitHub.
I want formalization of software development. It is orthogonal to experimentation and trying out fresh ideas.
Change is good. But not towards a local optimum and getting stuck there.
If it weren't so bad, we wouldn't be here.
- css is not scoped by default. Predicting the effect of adding a new rule is very hard if you are not using scoping solutions or if you are not constently writing new class names with a strict convention
- because of the first point, css is never refactored in most places. It becomes an ever growing monster of a code base
- css selectors are build to write css that depends on html. A lot of webapp should be writing html that depends on a small set of css rules in order to get a consistent ui
- the initial "semantic html approach" is counter productive for most webapps. This blog post is a very good read about this particular point: https://adamwathan.me/css-utility-classes-and-separation-of-...
Personally, I'm more interested in HTML as a document format in the original sense. Limited as it is, I just can't fathom the blackout moment where, confronted with a very basic HTML document plus a CSS stylesheet about the same size or more written in a completely different yet redundant syntax, somebody would say "yup, that seems like a reasonable document format for exchanging most of the world's text".
Edit: regarding your scoping requirement, isn't just including the id or class of a component in CSS selectors or, when necessary, sass-like nested CSS rules enough of a scoping mechanism?
The basic approach is to only use selectors with a single class name and nothing else. No child selector or anything.
Then you use a naming convention for class names. The most popular is called BEM (for Block/Element/Modifier).
Sass can help you write BEM class names: https://medium.com/@andrew_barnes/bem-and-sass-a-perfect-mat...
You might use other libraries. Those work by generating class names for you and applying those classes at the right place. There is a whole set of possibilities here: css-modules, styled-component, jss, etc. Some of those libraries use a separate file for writing the css rules (like css modules). Some other libraries ask you to write those rules next to your react/angular/whatever component (styled-component/jss). Those are often called css-in-js.
All of this is only useful when you are making a webapp.
CSS-in-JS solves a lot of problems that people that have been writing 'real CSS' have been experiencing for a long time. Writing CSS for very large applications that deal with lots of shared components across engineering teams is a hard task. Some of those problems are alleviated by libraries like styled-components.
It's not for every use case and should only be used when there is a genuine need (rather than 'because it's shiny.') But shrugging off CSS-in-JS because you don't understand it is a bit silly.
Namely?
Overall I think opinions on HN are quite progressive, the only area where I've seen this being different so far is CSS-in-JS frameworks.
I think a lot of people already have an opinion without trying it, oh well.
1. All of a given component’s dependencies and implementation can be viewed/resolved/navigated together.
2. Point 1 also aids static analysis, allowing easier type checking & linting, refactoring, dead code elimination etc.
Like it or not, Tailwind (and things that help people use Tailwind) serves a very valid niche. Unwrapping years of legacy CSS, intertwined with CSS injection in jQuery, is no simple feat. Adding Tailwind to that project and only writing Tailwind has helped immensely.
The downside is that there are times when you have a really long list of classes on a given tag, which can be difficult to read and parse. I’m not sure if this tool is right for me or my team, but that’s one use case I could see for using it.
Are we really complaining about emojis in READMEs nowadays? Good grief.
You can absolutely do that with CSS (and some people probably did for years), but having theses classes already available kind of shows you the way.
I like most HN threads but anything frontend is laughed at by people with little frontend experience. They repeat some truth they learned 20 years ago.
Honestly HN is not a good place to discuss frontend stuff. You will only see "get off my lawn" kind of comments.
1. I have a passion for front end, I have a PhD in Human Computer Interaction.
3. I am the cofounder and CTO of a company whose main product is based on CSS-in-JS (Emotion, Chakra UI)
2. I really like Tailwind and used it on several production projects.
4. So I know very well both of these approaches. I know their advantages and drawbacks.
5. In my comment I state the sad truth about this particular project: It combines the worst of both approaches in terms of performance, without any added value as compared to these approaches. Slow build AND slow runtime, without any added modularity.
This approach is objectively bad and this is why people react like this.
And to be constructive: If you want to use tailwind, use tailwind and you will have performant runtime but less customizability. If you want to use CSS-in-JS, use it and you will have customizability, but slower runtime. Engineering is trade-offs, but twin is still the worst of both worlds.
The runtime cost of css in js is real, and becoming a real problem for my app. Seeing like 2x slowdown as compared to classnames like in https://calendar.perfplanet.com/2019/the-unseen-performance-...
Well, something needs to justify all those latte orders while you patiently chill in front of your Mac for over two hours in the store.
I'm not ready to dismiss new initiatives off the bat (if anything they can at least be a good learning experience), but I'd like more concrete terms than "magic" and "flexibility" to help me understand the potential benefits.
[1] https://twitter.com/VincentTunru/status/1330103979762651138
So they're IDs, not classes?
(Ninja edit:) So to clarify: if I want to add some padding to an element with plain CSS, I have to think of a class name for that element, and then define styles for that class. However, if I happen to use that class name elsewhere in my code, then the styles I've defined there suddenly apply to that element as well. And vice versa: before I can delete a class from my code base, I have to be aware of every element that uses it.
CSS-in-JS solves that: all styles defined for a single component will only ever apply to that component.
The same holds true for Tailwind: you define the styles through the classes you apply, which will not affect other components in your codebase, and likewise, removing them (the Tailwind classes) is safe as well.
(And potentially also: what are their downsides, and does this project suffer from them as well?)
One of the nice things about Tailwind is that it’s completely portable, because it’s ”just” CSS. You can use the exact same design system regardless of how your markup ends up on the screen, as long as you can set a class attribute.
Why not just simple use Svelte?
Emojis in docs? Check
Piles of abstraction layers? Check
No tests? Check
Already at version 2, but has breaking changes? Check
Stacking up on yet-to-be-proven-framework (tailwind)? Check
We just need a BLM support message, donation link in the npm install stdout and some more JS shenanigans like support for yarn, gulp, webpack, npx, degit, and the next gen packaging tool soon™ to be released.
Sorry if this is offensive, but this ecosystem offends me and my sanity. This project is going to last 6 months and they'll move onto something else. I think I am finally ready to jump on React since it has been a few years and appears to be stable.
Edit: My ramblings are unfair. There are tests.
There's been many such trends to name a few - discarding years worth of stable codebases to try out some shiny new framework, throw away proven technologies to trade for some perceived developer convenience (Eg. PostgreSQL to MongoDB), labeling known terminologies to sound cool (Hacker vs Growth 'Hacker'). All these hipster moves come with a huge cost - they cost morale, time, money, teams, relationships. It just feels so unnatural to accept this as the norm.
In comparison, I just feel like some old dude from the 1950s rambling though these things have started to happen only the last half of the decade or so. Where did we go wrong?
Edited for OT
They come to expect that all development is as slow as theirs, including frontend <--> backend apis (graphql), huge js complexity (webpack), breaking changes (react-router), etc.
I see lots of companies nowadays wanting to hire "React devs". It seems we're collectively forgetting how fast you can ship stuff if you just use proven technology. Makes me wonder where we're headed.
Maybe it'll be a competitive advantage to use old (= proven) technology in the future?
My impression is, that began, when highly political people started to join the JS community (and Web engineering in general).
Of course, both you and me are going to be downvoted because these types oppose (downvote, "cancel") anyone who mentions their presence.
Speaking of piles of abstractions, please enlighten me. What is not a pile of abstraction today? Even x86 instructions are implemented with microcode.
Speaking of tests, anyone with some frontend experience will recognize that jest is a testing library.
I would say that your comment is the epitom of a HN frontend thread ie mean criticism by someone with 0 experience in the field.
Tailwind’s entire premise stands on these pillars which are shaky at best:
- Developers don’t want to write css so they add insane amounts of nuisance in templates. Impossible to read the design language.
- The original author of Tailwind complains about naming classes in the css but Tailwind ironically names things that are “tx—-bb-4-2” or whatever the hell that means. No thanks, I’ll write “customer-support-card” or something useful. Maybe even more generic - “card”.
- Tailwind’s front page is everything that’s wrong with this. Moving around components with “mr-4” tags and shifting it up. Left. Right. Making it larger. That means that there is no standardized design language and you’ll create design that will have no cohesive unity. One component will have “mr-4” margin and another “mr-3”.
We have the technology. Sass can create compositional components with reusability and logical hierarchy. There is no need for this Tailwind nonsense.
Tailwind is a controversial project, not mature. It’s proponents think that others just don’t get it. No, we do. We’ve used it and it makes it very fast to get something working but that’s more telling about one’s inability to take the time to think about layout/design and writing proper sass.
Fast is not always better. It makes spaghetti code with shit design that lacks an architect behind it.
There are many threads on HN about Tailwind and it’s always very controversial.
I wouldn’t call this thing mature when there is so much controversy.
One component having mr-4 and another having mr-3 is chill if that was the designer’s intention. If not, then it should be picked up during design review and easily fixed (since you end up really only having to pick from like a few values).
And put constraints as well! Comment them! Write for loops to create a full spectrum of tokens!
Then use those tokens to create a card class. Extend it for customer-support-card class!!!
What’s missing here?
Some comments:
- It's very easy to integrate Sass with Tailwind. You can import the Tailwind classes just like you would Bootstrap: https://tailwindcss.com/docs/using-with-preprocessors#using-...
- You are also allowed to write semantic classes as well: https://tailwindcss.com/docs/extracting-components
I would argue the main point would be you can rapidly build interfaces... like really fast. I worked on a proposal for a client using Tailwind a month ago and we smashed out a whole website in a week. We ended up "extracting components" like buttons and cards and stuff that we found was being reused a lot (kinda like Sass).
As opposed to Styled-Components, the Tailwind classes are very valuable for designers who want to prototype live in the browser.
Also, you were complaining before about design consistency but I would argue that Tailwind allows for very high design consistency. I work with armies of outsourced developers of varying skillsets and I honestly don't trust a lot of them to write clean CSS at all. With Tailwind and Tailwind design tokens, the army of outsource developers are encouraged to work with the design tokens (as opposed to yolo'ing 3px !important everywhere).
I am not sold.
I don't think this premise is very shaky, unfortunately, if it was shaky I doubt we would be seeing all these solutions coming around trying to help developers avoid writing css, solutions which seem to have uptake in the market.
- Unless the same class can be used in multiple HTML templates, there’s no point in giving the class a name.
- Only very general CSS can be used in multiple templates because anything specific will have the brittle base class problem.
PS mr-3 is .75rem and mr-4 is 1rem, so you can see with your eyeballs if you’re using the wrong one.
With tailwind you can create a very small configuration files where you pick a small set of available options for spacing and colors. This is actually a big help to implement a consistent design system. Many people see tailwindcss as a "design system builder".
Please show me the doc for "tx—-bb-4-2". Does it exist or are you just inventing it for some reason? Is "mr-2" unclear?
Sass has nothing to do with tailwind. It would be like comparing typescript and react.
Tailwind can be controversial and mature at the same time. Both things are not exclusive. PHP is controversial and mature.
I don't know about other people but in your particular case I would say that you are showing sign of a knee-jerk reactions and bad faith justifications.
I love me some sass scoped to the component. You don't need classes and the structure matches the markup, making it trivial to see what styles apply.
So you’re just writing your HTML twice, essentially. You can’t change your HTML without changing your CSS, so at that point, what’s the point in writing the CSS? May as well just add the styles right on the HTML.
They’re both fundamentally just part of how it looks, and I don’t see a lot of benefit in trying to keep that separate.
> - Developers don’t want to write css so they add insane amounts of nuisance in templates. Impossible to read the design language.
It's not so much that developers don't want to write CSS, but that it's largely impossible to delete lines of CSS.
> - The original author of Tailwind complains about naming classes in the css but Tailwind ironically names things that are “tx—-bb-4-2” or whatever the hell that means. No thanks, I’ll write “customer-support-card” or something useful. Maybe even more generic - “card”.
So those are different complaints about naming. Your complain is that names are hard to understand. Tailwind's complaint is that trying to define "generic" (i.e. reusable) classes is practically impossible. Is "card" the right abstraction? Is "customer-support-card"?
> - Tailwind’s front page is everything that’s wrong with this. Moving around components with “mr-4” tags and shifting it up. Left. Right. Making it larger. That means that there is no standardized design language and you’ll create design that will have no cohesive unity. One component will have “mr-4” margin and another “mr-3”.
The idea is that all those margins are part of a set of "constraints" that all compose well together to form a cohesive unity. More about this is written here: https://styled-system.com/guides/why-powers-of-two
> We have the technology. Sass can create compositional components with reusability and logical hierarchy.
This might be the main point: Tailwind means you're not defining reusable styles. All the styles are specific to the component, and the component is the thing you reuse - which we were defining already. So it actually gets rid of yet another tool to create compositional components with reusability and logical hierarchy.
(It's less fast for me to get to something working than e.g. using Bulma is, but I'm expecting that it will rid me of a number of the pains I run to when I'm hitting the limits of its built-in component styles.)
However, for production, I always come back and refactor everything into a beautiful Sass file that explains with comments what all the stuff is doing.
Now, the designer can come in and change something in the sass without having to worry about the rest of the app. They can easily debug too 'customer-support-class' straight from chrome browser.
I feel like Tailwind is a prototyping tool that's then shoehorned into production environment and in large applications where the design language (despite of tokens and standards) breaks down.
All of which is to say: just because it's not the best solution in your use case, that doesn't mean it doesn't solve actual problems for others in production.