Hate on Tailwind... I have seen that before!

In 2013, a company called Facebook introduced a framework called "React," which allowed developers to blend their JavaScript logic with their HTML code without having to modify the document directly.

This, my friends, was JSX, or JavaScript Extended, where you could write HTML within your JavaScript! I was just 9 years old, and I didn't know anything about programming, but the rest of the community had strong opinions about it when it was released.

"The audience was skeptical. Most people thought React was a significant step backward." - at JS ConfUS 29/05/2013

React was innovative because it challenged many established best practices of its time.

Separating JavaScript from HTML? Discarded.
Direct DOM manipulation? Discarded.
Two-way data flow? Discarded.

However, its boldness was a double-edged sword. Just as there were people excited about this innovation, others thought it didn't make sense. "Mixing JS with HTML? It will create more complexity and reduced maintainability.".

Some had been mixing JS and HTML before, such as in the MVC design pattern, which had a Model to manage logic, a View to render the UI, and a Controller to mediate between them.

Creating code that resembles React:

In computer science, there is a concept called "Separation of Concerns," which involves separating sections into their own contexts. In other words, you write your HTML in an HTML file, your CSS in a CSS file, and so on. This was a significant reason many opposed React.

Fast forward to today, React is the dominant framework. It's quite remarkable. But how does all of this relate to Tailwind? What about today?

Similar to React's history, Tailwind challenges many of the best practices we used to preach. Some people appreciate the advantages it offers, while others believe that the benefits don't outweigh the drawbacks.

Some dislike it due to its verbosity, but that doesn't make much sense (example in the image):

Verbosity - the quality of being wordy, speaking excessively, or using too many words to express oneself.

The real problem is with the "separation of concerns" mentioned earlier, and in this respect, Tailwind falls short. It's okay for CSS to have 1,000 lines if it has its own file, but a class with 1,000 words pollutes our HTML/JSX.

There are third-party libraries like Tailwind Fold, but there's no native solution that resolves this 100% and is considered a good practice (im looking to you @apply).

With this article, I didn't intend to persuade you or provide a solution, but rather to tell a story and help you understand that this kind of debate has occurred in the past.

As George Santayana said, "Those who cannot remember the past are condemned to repeat it."

Who knows, maybe in 10 years, Tailwind will also become one of the most famous ways to write CSS? Only time and its evolution will tell.

Anyway, thank you for reading!
Have a good day 😄

Comments

[Sultan]

Actually, Tailwind is the next evolution in styling. Before CSS, we used to style elements like this:

<p>
  <font color="red" face="Verdana, Geneva, sans-serif" size="+1">
    <b>Your formatted text goes here</b>
  </font>
</p>

However, this approach was deprecated with the introduction of CSS. With Tailwind, we are revisiting the concept, but instead of using attributes, we now put all styles in a className with shortened classes. This madness pushed me to come up with a new idea: importing styled React components from a CSS file, where all style manipulation can be done through component properties:

import { Title } from './styles.scss'
// crazy part, importing 👆 component from styles

<Title color="tomato" size="small">
  Hello world!
</Title>

Where style.scss file looks like this:

/**
  @tag: h1
  @component: Title
  size: small | medium | large
  color: #38383d --color
*/
.title {
  --color: #38383d;
  color: var(--color);
  font-size: 18px;

  &.small {
    font-size: 14px;
    margin: 2px 0;
  }
  &.medium {
    font-size: 18px;
    margin: 4px 0;
  }
  &.large {
    font-size: 20px;
    margin: 6px 0;
  }
}

For more information, please check out the demo and repository:

✨ Online demo
✨ GitHub repo

[Nícolas Gabriel]

That's a super interesting idea! Loved the concept!

[Joshua Amaju]

That tell's me all I need to know about your position on tech choices 😂

[Red Ochsenbein (he/him)]

You might want to take a look at PandaCSS for a similar but framework agnostic approach.

[Sultan]

There's nothing particularly unique about Panda CSS in terms of its idea or approach. Many CSS-in-JS libraries offer similar ways of styling.

[Red Ochsenbein (he/him)]

Tailwind isn't a new idea either.

[Sultan]

Yeah, but Tailwind made this approach popular, just as styled-components popularized CSS-in-JS. Other CSS-in-JS libraries introduced additional features, such as build-time extraction and shortened CSS classes, but in general, it is the styled component approach.

What I know is that there are several ways to style components:

• CSS module
• CSS in JS (styled-component way)
• Tailwind
• Inline styles
• Stylin with mapping style annotations

P.S.: If I had to choose between Tailwind and PandaCSS, I would pick PandaCSS.

[Ben Sinclair]

Who knows, maybe in 10 years, Tailwind will also become one of the most famous ways to write CSS

I think we're already there, unfortunately. In my experience, Tailwind is overwhelmingly loved by front-end developers.

[Adaptive Shield Matrix]

The biggest / most fair critique against tailwind I heard against is

• it makes things hard to read, by/if using multiple lines of tailwind classes
• using many divs in sequence - that makes it hard to distinguish between them
• you have to memorize that all the shortcuts mean, ex. p-1 text-sm

Here is an Article with said critique in more detail
nuejs.org/blog/tailwind-vs-semanti...

I feel like the critique comes most often from designers, who translate figma design into html/css and have less knowledge on how to write idiomatic react code.

First 2 problems are solved if

• extracting said divs into separate components
• using libraries like
  • tailwind merge - github.com/dcastil/tailwind-merge
  • and cva - github.com/joe-bell/cva
  • to merge and group class names

My biggest grip with react + tailwind is, that I wish I could shorten className to just c or class, because it is such a long word to type out and it takes to much screen / line space.

[Massimo Artizzu]

I feel like the critique comes most often from designers

The set of designers that do HTML/CSS development is definitely non-empty, but it's not the norm IME. Figma has also a decent Tailwind plugin for the matter, so they wouldn't even sweat to translate their design into Tailwind mashups.

First 2 problems are solved if

• extracting said divs into separate components
• using libraries like
  • tailwind merge - github.com/dcastil/tailwind-merge
  • and cva - github.com/joe-bell/cva
  • to merge and group class names

I fail to see how tailwind-merge could improve readability, while the others basically throw one of the main advantages of Tailwind out of the window: eliminating the necessity of naming things.

My biggest grip with react + tailwind is, that I wish I could shorten className to just c or class, because it is such a long word to type out and it takes to much screen / line space.

I personally never write className in full, because the IDE suggests it right away when I type just c, but I'm amazed to see that you consider className "too long" while the value could be something like bg-white border-slate-100 transition-all duration-500 demo-dark:bg-slate-800 transition-all duration-500 demo-dark:border-slate-500 border-b rounded-t-xl p-4 pb-6 sm:p-10 sm:pb-8 lg:p-6 xl:p-10 xl:pb-8 space-y-6 sm:space-y-8 lg:space-y-6 xl:space-y-8.

[Adaptive Shield Matrix]

while the value could be something like "long lines of tailwind"

I group similar tailwind classes together

Example:

export function CardWrapper({ children, className }: { children: React.ReactNode; className?: string }) {
  return (
    <div
      className={cn(
        "rounded-lg  shadow-lg", // card border + shadows
        "p-4 lg:p-8", // padding
        "bg-white dark:bg-gray-900", // bg
        className,
      )}
    >
      {children}
    </div>
  )
}

somewhere in ui / utils

function cn(...inputs: ClassValue[]) {
  return twMerge(clsx(inputs))
}

In the long lines of tailwind classes you provide

• why do you not group tailwind classes (like in the example above)?
• why do you not separate it into a separate component (like in the example above)?

Separating into function components like in the example

• Is the example above hard to read?
• Will it be if you add some more lines of tailwind classes in the same way?
• I can give an example with even more tailwind classes following same pattern and I find it is very readable and maintainable for myself.
• The only downside I see is you have to write many separate function components and naming them (vs similar how you have to define/name things for in css).

[Massimo Artizzu]

• why do you not group tailwind classes (like in the example above)?

Because it's not straight from the source - it's the output, taken directly from Tailwind's homepage. If I have to debug a problem, that's something the dev tools presents me.

• why do you not separate it into a separate component (like in the example above)?

Common answer: because I don't want to name things.

• Is the example above hard to read?

Definitely harder that reading plain CSS, yes.

[Ben Sinclair]

That's a good linked article.

[Caleb Taylor]

My biggest grip with react + tailwind is, that I wish I could shorten className to just c or class

There's solidjs, lit or preact that uses just class attribute

[Sergiy Yevtushenko]

There are reasons for hate. Take a look at this article. I'd suggest read other articles from this author too. It's not a plain dislike or taste based hate. It's well grounded position. But, well, language might be somewhat offensive to some framework lovers.

[Paulo Teixeira]

I understand the reasons, but I don't like it. I know that evolution is necessary, but I also see front-end as an amusement park without supervisors, there are good ideas emerging wildly, and just because it seems cool to me doesn't mean I'll go into production and have to deal with it in 5 years.

Lots of new technologies wanting to solve problems that aren't even problems yet, that will probably cause problems.

[Nícolas Gabriel]

interesting point of view 🤔
we usually think about the now and not about the distant future, but that is something we should keep in mind as well, so, fair judgement I would say

[Dominykas Bartkus]

In the end, it is your responsibility to choose what works for you. It’s always a the matter of: “can I afford this abstraction/solution or should I choose battle tested tech?”

It’s the matter of what you think will be most stable and scalable for you software.

For instance, I regret choosing prisma as an ORM. It’s great for basic projects, but for larger projects that you need more control over prisma is awful outside of it’s schema and migration tool.

Is it bad on its own? No, just not perfect for what we’re building for multiple query related reasons.

Love its typings though.

Anyway, I’m all for new tech. Just make your judgment whether you should or shouldn’t use it. Don’t trust the hype.

But even with all the hype around tailwind, I found it useful. I don’t care about css much. Just want things to work, and tailwind provides me that out of the box without having to care about building my own set of standards. And 90% of the time it works 100% of the time.

[Camilo Micheletto]

Separation of concerns is just another opinion from what software could be and it's a lazy way to measure if a language can or cannot provide what your business needs in CSS architecture.

Tailwind provides an easy interface to CSS by dropping most of its funcionalities and abstracting it's units and APIs. This is not wrong, not good for everything and hard to measure in an industry that sees CSS architecture as a liability. When we talk about CSS we can be sentimental about it, we can talk about perceived performance in how fast we code or how we feel when coding, and care less about how it makes sense in a business model and if it truly provides an API that does not rely on template and are tightly copled to an ecosystem providing functionality.

Today tailwind future is bright and we can afford not to think about CSS at all, and abstracting even further regarding of its limitations, and it's ok, even if it does not makes sense > every time < business wise as the community makes it.

As you said the wheels always turns, it's good to see some fresheness, but I wish the discussion regarding CSS technology were deeper than the surrounding Tailwind are. Knowing it's limitations is also a power that makes you use it wisely.

[Massimo Artizzu]

we can afford not to think about CSS at all

I hope this is just an unfortunate way you expressed yourself, because that would be a terrible approach. Tailwind does not save you to learn and think of CSS.

[Camilo Micheletto]

Yes It does. Don't get me wrong, theorically I'm with you. In practice people could care less what Tailwind outputs as CSS as long as it works layout-wise - and the industry doesn't seem to care as well, CSS architecture is hard to grasp, to mantain and to enforce certain patterns.

So when I say

we can afford not to think about CSS at all

I'm saying that you can write the most verbose, repetitive, media-query full class based CSS and don't ever open the devTools again nor check the output code as long as layout works, relying only in template based reuse, JIT and compression in matters of performance, as if it were enough (it isn't).

[Massimo Artizzu]

as long as layout works

And that's the point. Because almost nothing works on the first try, you'll eventually have to debug, and understand what's going on. And devtools presents you CSS, not Tailwind - which is not, and will arguably never be - a first class citizen.

I won't even mention that Tailwind's own documentation is full of references to the corresponding CSS properties and/or mechanics, so you have to know CSS to understand what Tailwind means.

What you're imaging - an abstraction layer that renders knowing what's under the hood completely optional - is a pipe dream that will eventually collide with reality (and in the case of Tailwind it fortunately happens very soon).

It's not something we haven't seen in the frontend community, mind you. Back in the day, there were jQuery developers who hardly knew anything about the DOM or even JavaScript. More recently, we have React developers who get lost outside of it. Accumulating knowledge debt isn't a route with a good outlook, although it seems sufficient in the short term.

[Camilo Micheletto]

I really hope you are right, because when there is an error with jQuery, it'll certainly break the application. CSS errors, verbosity or misconceptions are way more cheaper and overlooked, even if the layout breaks at some point. I had this problem working with CSS, SCSS and even Styled codebases, but with Tailwind is worse, not only because the thicker abstraction layer, but the community around it.

[András Tóth]

The same idea is repeated with ORMs which in my opinion comes from a very greedy point of view: we don't have time to understand something in depth, we want it immediately.

This is not a bad thing, on its own, when you work on prototypes and do market-research and you try to land what exact features bring value to people.

But then, eventually, successful companies need to scale; they start serving more people, the amount of data grows very large under them and so on.

You can't afford not understanding how SQL works, how the specific database you use work. You would do tons of ineffective queries with an ORM that can be into 1 query 1 round-trip with a mid-level SQL knowledge.

Same is true with CSS: you can design a set of rules that are easy to change, but if you don't understand it, and with the magic framework for hiding it for you, you won't be able to achieve certain effects, you will struggle a lot with a design change, your code will be brittle and ineffective, because there were much simpler ways that even a mid-level CSS developer would have created for you.

Why I am angry with bootstrap and tailwind and the rest, that they are no longer sold as a great tool for rapid prototype; we are told that it is "evolution", it is the logic next step.

It is not! Neither are ORMs. You can't skip learning your craft. You will need it eventually.

[András Tóth]

The false dichotomy of the day is this:

// whoa, so little!
<div class="px-0 pt-1 text-sm" />

vs.

/* Ew, too long */
.my-component {
  padding: 0.25rem 0 0 0;
  font-size: 0.75rem;
  line-height: 1rem;
}

For a more eloquent great case against using utility classes please read this: every-layout.dev/rudiments/global-...

Now my short opinion. Locally styling each and every component is bad CSS practice. You should have a lean vocabulary of well-defined bigger components and shapes. You see, CSS is an exception-based language: you create a set of general rules, and then only write further ones, when you have to differ from those.

Tailwind on the other hand, went full in with local styling.

When you have globally set general rules, you will have less changes required to change the appearance of the page. If anyone here remember CSS Zen garden, it was possible only because the classes on the items were semantic. CSS Zen garden would have never been possible using the tailwind approach.

Generating another style would instead mean to rewrite each and every component. Let's say: screen sizes grew and we can afford thicker margins? With tailwind approach now you have to go into each items and change those values. Because you have eliminated general rules from the equation, now everything is an exception.

This line: <div class="px-0 pt-1 text-sm" /> means that you want an average <div> except that it has no horizontal padding, it has 1 amount of top padding and it also has a special small set of font.

If you are lucky enough someone in the team will have a brain to move it into "presentational" React component, that does nothing just the same thing class="something" would do.

[Eckehard]

Maybe we should look at the topic a little more generally...

HTML is not the first language to build User interfaces. If you write an application for windows, linux or apple, things are very similar. If you need an input field, you will define it in a similar manner you do in HTML. But in traditional applications, you do not have so much control over the appearance, in fact, most visual components are styled by the OS, not the application. This may limit the developer, but gives a better user experience. Companies like Apple have their Human Interface Guidelines to ensure a uniform user experice - something we would whish we had for web developers today.

HTML / CSS was initially not designed as a UI toolkit, it was designed to apply better styling to static documents. The more the web is used to build applications, the more the conceptual issuses of CSS/HTML will be visible. The web is evolving from a presentation platform to an application platform. Any modern CSS toolkit should support this evolution, so future will show, if tailywind does the job.

[Peter Vivo]

Imho Tailwind greatest strength is eliminate the abstraction layer between layout of HTML and CSS. Before Tailwind I need to be use a verbose and class link between HTML DOM and CSS, with boring name giving process. With Tailwind mandatory layout information is stay here where it is belong.

export default component$(({contentList}) => { 
  return (
    <main class="grid place-items-center min-h-screen">
      <article class="grid gap-2 w-2/3">
         { contentList.map(({content, id, user}) => 
            <Paragraph id={id} user={user} content={content} />
         )}
      </article>
    </main>
  )
};

In this example easy to understand which layut setup are we build. Does not need to futher explanation.

The JSX format MVC concept is really simple: The return object is the View, the props or useContext is the Model, and the hooks is the Controller.

[Massimo Artizzu]

You're forced to name things only because of the limitations of your framework of choice. How about this:

main {
  display: grid;
  place-items: center;
  min-height: 100vh;
}
article {
  display: grid;
  gap: 0.5rem;
  width: calc(100% * 2 / 3);
}

If you have a truly encapsulated stylesheet - even emulated - you could do this.

[Peter Vivo]

You are right this is the CSS solution but the problem is the selector: main and article which is to global and it is affect all main and article. But the tailwind does not need to create instance name, and recreacte DOM selector in CSS side.

[Massimo Artizzu]

I mentioned style encapsulation for a reason: those selectors would have a local scope only.

This has been possible for a while, even natively.

[Peter Vivo]

My example does not framework dependent, that DOM part can be written in plain HTML too.

[Massimo Artizzu]

I never mentioned any framework either. I said it's possible to do that natively.

[Peter Vivo]

Sorry I missunderstand you, In my point of view simple:

<foo class="layout by tw">
  <content />
  <content />
  <section class="another layout by tw">
     <input />
     <button />
  </section>
</foo>

Is much easier mental model than

<foo class="layoutCSSClassName">
  <content />
  <content />
  <section class="anotherLayoutCSSClassName">
     <input />
     <button />
  </section>
</foo>

.layoutCSSClassName {
 //
}

.anotherLayoutCSSClassName {
  //
}

Because In first case you can immediatly know everything about your layout conception, elements and layouts directive with a well .

[Massimo Artizzu]

Why should you know what's the layout from the markup? The markup is for the content.

What I suggest is something like this:

<foo>
  ...
</foo>

with a corresponding CSS:

foo {
  /* layout styles */
}

[Gabriel Onezio Ferreira]

10 years? I believe that as the technological evolution is exponential, tailwind will evolve in less than 2 years for this to become an interesting factor!

v1.tailwindcss.com/docs/upcoming-c...
On the official Tailwind CSS website, there is a section called Upcoming Changes, which shows the updates and deprecations coming in the future and how to prepare for them. For example, they plan to remove the deprecated gap classes, change the purge mode for layers by default, and add default line heights to the font size classes. These changes are intended to simplify and improve the use of Tailwind CSS.

[Søren Kottal]

Tailwind is at version 3.3, and the changes you mention has long been done.

[LearnCodeProfessor]

Great post!

[zKriguer]

Thats a good approach, thanks for sharing your experience with both ways to do CSS.

[Massimo Artizzu]

I don't think I'll ever stop thinking Tailwind is a footgun, generally speaking.

Other than polluting the markup, my main concern is that it's another cognitive layer on top of CSS. Meaning you have to learn Tailwind's "syntax" in order to use it. At that point, I'd be happy using CSS straight away.

Not to mention that Tailwind's syntax for some more-than-dead-simple cases (like a calc, a custom property or simply targetting descendants) is clunky at best:

<div class="w-[calc(50%-80px)] text-[color:var(--text-color)] [&>*]:p-4">

I understand the nicety to use utility classes, but definitely not to make a mess out of your markup - that is now more like a <div>-soup:

<div class="relative pt-10 xl:pt-0 mt-10 xl:mt-2">
  <div class="max-w-7xl mx-auto px-4 sm:px-6 md:px-8 lg:grid lg:grid-cols-12 lg:gap-8">
    <div class="lg:col-span-5 xl:col-span-6 flex flex-col">
      <div class="relative xl:mt-18">
        <div class="mt-6 sm:mt-10 relative z-10 rounded-xl shadow-xl">
          <div class="bg-white border-slate-100 transition-all duration-500 demo-dark:bg-slate-800 transition-all duration-500 demo-dark:border-slate-500 border-b rounded-t-xl p-4 pb-6 sm:p-10 sm:pb-8 lg:p-6 xl:p-10 xl:pb-8 space-y-6 sm:space-y-8 lg:space-y-6 xl:space-y-8">
            <div class="space-y-2">
              <div class="relative">
                <div class="ring-cyan-500 transition-all duration-500 demo-dark:ring-cyan-400 ring-2 absolute left-1/2 top-1/2 w-4 h-4 -mt-2 -ml-2 flex items-center justify-center bg-white rounded-full shadow">
                  <div class="w-1.5 h-1.5 bg-cyan-500 transition-all duration-500 demo-dark:bg-cyan-400 rounded-full ring-1 ring-inset ring-slate-900/5">

This is taken straight from Tailwind's homepage, by the way. When you forfeit control of the meaning of your DOM, this is usually the result.

The main alleged advantages of Tailwind all seem very weak to me:

• the size of the generated CSS is much smaller - but CSS has hardly ever been a problem for size and CPU load;
• HTML can be well compressed with gzip and others - but CSS is compressed even better and it's usually cached, while HTML is not;
• it saves you from the burden of "naming things" - but naming is a developer skill, and keeps track of why you're doing what you're doing;
• it lets you forget about the cascade - but the problem with the cascade is when it becomes too big to be easily handled, while there are many other native solutions for that (style encapsulation, style scoping, cascade layers).

Yes, the last point means that you need to learn and understand some advanced mechanics of CSS that need time and effort to be fully internalized. Tailwind wins when the team has developers with less experience in CSS, because it's an easy-enough solution for all. But those less-experienced devs will also have the worst time when it comes to maintain their own code.

[Dusan Petkovic]

I wonder what is the best practice in Tailwind for grouping styles, in normal css you can do that with classes, and then create a special class to indicate a specific button style.

What I don't like is having to copy and paste the same classes over and over... I've used the @apply directive but not sure if its the best practice..

[András Tóth]

That's where the design mistake shows. In a well-architected CSS page, you set sensible defaults for main elements. You look at your HTML tags, you create sane defaults for them that already gets you 80% there. Then you create special HTML classes that will represent your main layout components: is it a main panel, is this a header, is this a label? And so on. You are 95% there.

Finally you look at the components which need adjustment and you define those rules. This is a top-down approach, requiring knowledge of CSS and HTML, also you should have a lot of certainty of what kind of application you want to build.

Tailwind's approach is to work in isolation, bottom-up: you get a design, you add HTML tags until it matches closely what they gave you and you move on. You don't discuss decisions, you don't factor in future changes, you go and get productive now. Perfect for companies where people want to work hard, but smart, scaling or future-proof are not requirements.

In my opinion, if you wish to write code for long-term, you have to understand good CSS architecture.

If you want a band-aid, you can also make presentational components, that now nothing about anything, they just accept children and that's it.

export function TheSameComponentUsedEverwhere({ children }: props) {
  return <div className="bg-white border-slate-100 transition-all duration-500 demo-dark:bg-slate-800 transition-all duration-500 demo-dark:border-slate-500 border-b rounded-t-xl p-4 pb-6 sm:p-10 sm:pb-8 lg:p-6 xl:p-10 xl:pb-8 space-y-6 sm:space-y-8 lg:space-y-6 xl:space-y-8">{children}</div>
}

Which you can use as:

function MyComponent(props) {
   const { posts } = useExternalStore(); 
   return <TheSameComponentUsedEverywhere>
    {posts.map(({url, title}, index) => (<a href={url} key={index + '_' + title}>{title}</a>)}
   </TheSameComponentUsedEverywhere>
}

[Dusan Petkovic]

Its certainly the opposite of how I am normally used to write CSS, but maybe it plays well with the component based architecture...

[Massimo Artizzu]

It is not according to Adam Wathan himself.

[Dusan Petkovic]

So the @apply directive is the best way to group?

[Massimo Artizzu]

No, I mean Wathan discourages the use of @apply.
He provided it to ease the adoption of Tailwind, but he personally just chugs classes into the markup.

[Victor Dolsan]

Nice post! I think Tailwind will become the standard way in a very short time