▇▇▇▇▇▇▇▇▇▇▇▇
▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
▇▇▇▇▇▇▇▇▇▇
▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇
So a few days ago I found myself looking into a rabbit-whole I've been in quite a few times...
About 7 years ago when component libraries were just about to become a thing I was part of a small team that drastically increased development speed and unified the UI/UX of Jimdo by creating a component library along with a strategy on how to get adoption and support from all product teams.
After that I was able to take what we've learned at Jimdo and apply it to new and existing design systems at various other companies and projects.
After that I've been more on the product side of things until my employer musicube was acquired by songtradr and the design and development teams there were just on the verge of creating a component library.
That's the reason I thought it might be cool to take everything I've learned about these systems, mix it with my current assessment of the web platform, sprinkle a little flat-hierarchy ideology in there and write down how I'd create a component library today, mid 2022.
Ready? Ok let's go! 🚀
Why should I build a component library in 2022?
Communication between designers and developers is essential when building software products in a bigger company. And a component library is on of the most powerful tools they can utilize to make communication as easy as possible.
Each component documents all design and development decisions leading to it.
The creation process may vary, but in the end there is one component that has been approved by both development and design. Now it can confidently being used, knowing that it's the latest state and when needed also can find out which decisions lead to it.
When someone is not entirely happy with a component they can use the current state of the library as a reference point and propose improvements based on that.
Users of your product will subconsciously learn its UX patterns and apply it to other parts of the product. When these patterns match, the user is effective and happy.
A widely adopted component library is the best way to get there.
If you fail to deliver a coherent experience, for example when the colors of your destructive buttons vary a lot across the system, navigating requires more cognitive affords from the user which can be exhausting and lead to deep sadness 😥.
Users of the library should be faster when using it then they'd be without it.
..."Fast" is a pretty diffuse term in product development... what I mean is:
How should I approach building a library to be effective?
I saw the greatest success when people maintain, contribute and use a company internal library as if were an open source project (only that this one is "just" open to the whole company instead of the whole world).
With that in mind it comes natural to document everything properly, to be inclusive to everyone, to work with issues and pull requests, to be mindful with API changes, etc...
🧑🏫 Don't worry in case you never maintained or contributed to an open source project, that's fine.
I cover the most important parts later in this article. Just behave sensible, remember why the project exists and be respectful to users and co-maintainers.
The most used methodology for design systems is probably Atomic Design. And though I think there is a great amount of guidance you can get out of that remember that a component library is a living system.
Everything can be changed, especially in the beginning.
Later on be careful when renaming API in order to keep the amount of breaking changes low*
Dare giving components a suboptimal name or put them in the wrong category. The most important thing is that everybody (design and development) is on the same page and uses a common language.
You want to allow projects to have their own components and maybe even variants of elements in the library. In a living system, evolution must also occur. And a process that is too strict hinders product development instead of helping it.
I still like to call these one-of components "snowflakes", a label coined by the team behind XINGs ui system.
In order to support rapid prototyping and to prevent locking your company in a narrow tech choice I strongly recommend maintaining different integration points with the library.
When you have a common css preprocessor (sass for example) that is widely adopted in your development, this makes for a great low level integration.
Provide Colors, Spacings, Typography settings and maybe also mixins on this level
🤔 Think plain Bootstrap.
Level 2 uses the values provided by the CSS Preprocessor and provides static components that implement the look of your UI design.
The web platform uses CSS and HTML for UI, there's no way around it and therefore integrating here is tremendously valuable.
Create a stylesheet that unlocks your components in any page by simply adding a <link />.
Even when your development writes no single line of plain/html and uses ReactJS only, Marketing & Support teams use low-/no-code solutions, Growth teams want to try out some crazy new stuff, Companies get acquired that had build their
awesome product with Rails and now should adopt your look fast...
All of these use HTML/CSS in the end and most likely support serving a css file.
This level abstracts specific html structure and detailed class names and should get really close to the language that is actually spoken between
design and development.
When someone says "Can you please create a centered modal with a destructive button saying 'Delete!'?", on the framework integration level it should boil down to something like
<Modal align="modal-center">
<Button variant="button-destructive">Delete!</Button>
</Modal>
It uses the static components provided by the HTML + CSS level and arguments them with behaviour, makes them accessible and easy to work with in your most common product development environments.
Make sure that every state is accessible and changeable from outside the component. Most state is best managed in the products using the library rather then in the components even when it requires a little more boilerplate on the consumer side.
routing is also a state
There is this great article from Brad Frost about the difference between storefront and workshop documentation.
Let's comparing the most popular UI library tools
ℹ️ Figma is also a workshop
So with that in mind, and the fact that the library is of little worth when nobody uses it I advise to have an up to date storefront. (Doesn't mean you can't also have a workshop).
I personally feel like managing external dependencies is one of the most painful and challenging tasks in modern web development.
Therefore it's of high worth when your library is sleek, doesn't disrupt the dependency choice
of consumers and has little to ask when dependants want to bring their packages up to date.
There is no golden rule here but I'd recommend using as little dependencies that might interfere with consuming projects as possible.
Users of your library will inevitably need custom styling. They should be free to choose if they use inline styes, css in js or tailwind.css for that.
The libraries job is to provide component styles, not to serve all css needs.
Ideally whenever you release an update to the library, consumers can simply install the latest version and ship it to users.
In order to achieve that be aware of the APIs consumers rely on.
When you went with the integration points mentioned earlier your APIs are now:
Whenever one of these is modified or the look / behavior is changed significantly you should consider it a breaking change – meaning users of the library must be informed and provided with straight forward instructions on how to migrate.
Your library will most likely include more components and helpers than any single project will need to use. So there should be tooling at hand to either only pick the parts needed for a project or to remove code that is not required.
PurgeCSS needs to statically find the css classes you're using in your sourcefiles in order to know what css to keep and what to throw away 🚮
so its usually more effective to use class names in api's instead of magically creating them, at least for less common variants.
// Good:
<Button color="button-blue" />
// => <button class="button button-blue">
// Problematic:
<Button blue />
// => <button class="button button-blue">
What tech stack should I go with today?
Here's my personal list of solutions that stood the test of time, have huge adoption rates and support the requirements I've set out above.
There are lots of other puzzle pieces required but I think these are the most important ones to get right in the beginning.
Alright, now let's go build a library! 📚