--- layout: post title: "More Transparent UI Code with Namespaces" date: 2015-03-06 14:11:00 categories: Web Development meta: "Employing a suite of namespaces to make your UI code more readable and transparent" --- When we work at scale, we often find that we spend a large amount of our time reading, maintaining, and refactoring existing code, rather than writing and adding new features. This is the reason we focus so much on things like architectures, naming conventions, methodologies, preprocessors, scalability, etc.: because writing CSS is easy; looking after it is not. What we want is to be able to write code that is as transparent and self-documenting as possible. Transparency means that it is clear and obvious (to others) in its intent; self-documenting means that we don’t have to lose time to writing and reading lengthy, supplementary documentation. The need for this is particularly true when working with languages like HTML and CSS. Their declarative nature means there is no control flow to give clues as to the state or shape of the project, and the fact that the two languages are written separately but exist so closely often provides a large disconnect between some CSS’ source and where it is implemented. That is to say, we may see classes all throughout our markup, but that is only one very small part of the picture: somewhere else there is the corresponding CSS that completes the other half of the story. Cross-referencing these classes to ensure their proper treatment (reusing them elsewhere in the DOM, binding onto them to make modifications, deleting them to remove styling, etc.) requires a very diligent developer, and consumes a lot of time. How many times have you looked at a piece of HTML only to wonder which classes do what, which classes are related to each other (if at all), which classes are optional, which classes are recyclable, which classes can you delete, and so on? A lot of times, I’m willing to bet. Naming conventions like [BEM](/2013/01/mindbemding-getting-your-head-round-bem-syntax/) do a fantastic job to help communicate the roles and responsibilities of the classes we find in our HTML, and if you’re not yet using BEM then I urge you to stop reading this article right now and to start with that instead—this post will be levelling BEM up a notch. To quickly recap, BEM gives us two very useful suffixes—`__element` and `--modifier`—that we append onto our classes in order to tell us the role of certain bits of UI, for example: /** * The top-level ‘Block’ of a component. */ .modal {} /** * An ‘Element’ that is a part of the larger Block. */ .modal__title {} /** * A ‘Modifier’ of the Block. */ .modal--large {} In our CSS, this naming isn’t all that useful, but when we see it in out HTML we get a much better view of what’s going on:

Sign into your account

We can see from this that we have a number of classes all relating to our `.modal`, and a class of `.form-login` which begins a brand new context. Being able to glean this level of information from our classes in our markup actually tells us quite a lot about the corresponding CSS, and also about how and why they interact with each other in the way they do. It also tells us about how we should (or should not) reuse these classes elsewhere in the DOM: `.modal--large`, `.modal__title`, and `.modal__content` all have a dependency on `.modal`, and therefore cannot be used without that `.modal` class also being present. This gives us some great transparency and—because it exists right there in our classes—it is also fairly self-documenting. This is a naming *convention*. One thing I’ve been researching and implementing a lot with my clients lately is the idea of taking naming conventions a step further by adding _namespaces_. A naming convention tells us how classes within a component relate to one another, but a namespace will tell us exactly how classes behave in a more global sense. A namespace tells us exactly what a class (or suite of classes) does in non-relative terms. --- There are a number of common problems when working with CSS at scale, but the major two that namespacing aims to solve are clarity and confidence: * **Clarity:** How much information can we glean from the smallest possible source? Is our code self-documenting? Can we make safe assumptions from a single context? How much do we have to rely on external or supplementary information in order to learn about a system? * **Confidence:** Do we have enough knowledge about a system to be able to safely interface with it? Do we know enough about our code to be able to confidently make changes? Do we have a way of knowing the potential side effects of making a change? Do we have a way of knowing what we might be able to remove? Usually, unfortunately, the answer to most of these questions is no. This is the main reason we end up with bloated codebases, full of legacy and unknown CSS that we daren’t touch. We lack the confidence to be able to work with and modify existing styles because we fear the consequences of CSS’ globally operating and leaky nature. Almost all problems with CSS at scale boil down to confidence (or lack thereof): People don’t know what things do any more. People daren’t make changes because they don’t know how far reaching the effects will be. Old CSS never gets deleted because it’s hard to tell where things might be being used. As a result, we pile on new CSS, using new selectors, in order to avoid having to touch anything that exists already. Our CSS gets increasingly hard to manage, new styles get added where they might not be needed, legacy CSS remains a part of the core codebase, and then the only option is to do a complete teardown and rewrite every few years. Expensive. With the nature of maintaining a large project like this, we often find that we spend more time reading markup and its styling through developer tools than we might do reading source CSS files. This means that meaningful class names become invaluable for communicating rich information to other developers. We need to say exactly what a class does, why it exists, where (else) it might already occur, whether or not we can reuse it elsewhere, and how safe it is to bind onto or modify. This means that the names of the classes become documentation, and we can read all of that documentation right there in the view. Wouldn’t it be nice to know the exact scope and reach of a selector from its name alone? Read on… ## The Namespaces In no particular order, here are the individual namespaces and a brief description. We’ll look at each in more detail in a moment, but the following list should acquaint you with the kinds of thing we’re hoping to achieve. * `o-`: Signify that something is an Object, and that it may be used in any number of unrelated contexts to the one you can currently see it in. Making modifications to these types of class could potentially have knock-on effects in a lot of other unrelated places. Tread carefully. * `c-`: Signify that something is a Component. This is a concrete, implementation-specific piece of UI. All of the changes you make to its styles should be detectable in the context you’re currently looking at. Modifying these styles should be safe and have no side effects. * `u-`: Signify that this class is a Utility class. It has a very specific role (often providing only one declaration) and should not be bound onto or changed. It can be reused and is not tied to any specific piece of UI. You will probably recognise this namespace from libraries and methodologies like [SUIT](https://suitcss.github.io/). * `t-`: Signify that a class is responsible for adding a Theme to a view. It lets us know that UI Components’ current cosmetic appearance may be due to the presence of a theme. * `s-`: Signify that a class creates a new styling context or Scope. Similar to a Theme, but not necessarily cosmetic, these should be used sparingly—they can be open to abuse and lead to poor CSS if not used wisely. * `is-`, `has-`: Signify that the piece of UI in question is currently styled a certain way because of a state or condition. This stateful namespace is gorgeous, and comes from [SMACSS](https://smacss.com/). It tells us that the DOM currently has a temporary, optional, or short-lived style applied to it due to a certain state being invoked. * `_`: Signify that this class is the worst of the worst—a hack! Sometimes, although incredibly rarely, we need to add a class in our markup in order to force something to work. If we do this, we need to let others know that this class is less than ideal, and hopefully temporary (i.e. do not bind onto this). * `js-`: Signify that this piece of the DOM has some behaviour acting upon it, and that JavaScript binds onto it to provide that behaviour. If you’re not a developer working with JavaScript, leave these well alone. * `qa-`: Signify that a QA or Test Engineering team is running an automated UI test which needs to find or bind onto these parts of the DOM. Like the JavaScript namespace, this basically just reserves hooks in the DOM for non-CSS purposes. Even from this short list alone, we can see just how much more information we can communicate to developers simply by placing a character or two at the front of our existing classes. It is probably worth noting at this point that these namespaces do not exist for encapsulation and sandboxing of styles, but for clarity and informative reasons. [Ben Frain](https://twitter.com/benfrain)’s [FUN](http://benfrain.com/fun-css-naming-convention-explained/) convention utilises namespacing as a means of soft encapsulation. ## Object Namespaces: `o-` Format: .o-object-name[|] {} Example: .o-layout {} .o-layout__item {} .o-layout--fixed {} The `o-` namespace for Objects is a very useful one for any teams who use Object-Oriented CSS. OOCSS is fantastic in that it teaches us to abstract out the repetitive, shared, and purely structural aspects of a UI into reusable _objects_. This means that things like layout, wrappers and containers, the [Media Object](http://www.stubbornella.org/content/2010/06/25/the-media-object-saves-hundreds-of-lines-of-code/), etc. can all exist as non-cosmetic styles that handle the skeletal aspect of a lot of UI components, without ever actually looking like designed ‘things’. This leads to much DRYer and drastically smaller stylesheets, but does bring with it one problem: how do we know which classes might be purely structural, and therefore possibly being used in an open-ended number of instances? This poses problems on projects quite frequently. Picture the following example. Imagine you’re a developer new to a project, and you have no intimate knowledge of the CSS or what its classes mean or do. You’re asked by a Product Owner to add some padding around the testimonials that appear on the site. You right click, Inspect Element, and you see this:
Now, it should be fairly clear here that what you should do is go and find the `.testimonial {}` ruleset in your CSS and add the padding there. However, using DevTools, you find that adding the padding to the `.media {}` ruleset has exactly the outcome you expected. Perfect! Let’s go and add that into the source CSS file. The issue here is that `.media` is an abstraction (it’s actually the poster child of Nicole Sullivan’s OOCSS) which, by definition, is a reusable and non-cosmetic design pattern that can underpin any number of different UI components. Sure, altering the padding of it in this instance gave us the desired results, but it also may have just unintentionally broken 20 other pieces of UI elsewhere. Because objects don’t belong to any one specific component, and can underpin several vastly different components, it is incredibly risky to ever modify one. This is why we should introduce a namespace, to let other developers know that this class forms an abstraction and that any changes here will be reflected in every object sitewide. The object itself does not necessarily have anything to do with the implementation-specific bit of the UI that you are trying to change. By adding a leading `o-` to the classes for our objects, we can tell other developers about their universal nature, and hopefully avoid ever having people binding onto them and breaking things. If you ever see a class that begins with `o-`, alarm bells should ring and you should know to stay well away from it.
* Objects are abstract. * They can be used in any number of places across the project—places you might not have even seen. * Avoid modifying their styles. * Be careful around anything with a leading `o-`. ## Component Namespaces: `c-` Format: .c-component-name[|] {} Example: .c-modal {} .c-modal__title {} .c-modal--gallery {} Components are some of the safest types of selectors we will encounter. Components are finite, discrete, implementation-specific parts of our UI that most people (users, designers, developers, the business) would be able to identify: This is a button; This is the date picker; etc. Usually when we make changes to a Component’s ruleset, we will immediately see those changes happening every- (and only) where we’d expect. Unlike with Objects, changing the padding on the `.c-modal__content` should not affect anything else in the site other than the content area of our modal. Where Objects are implementation-agnostic, Components are implementation-specific. If we revisit the previous example, and introduce the Object and Components’ namespaces, we’d be left with this:
Now I can tell _purely_ from this HTML that any changes I make to the `.o-media` class may be felt throughout the entire site, but any changes I make to the `.c-testimonial` ruleset will only modify testimonials, and nothing else. * Components are implementation-specific bits of UI. * They are quite safe to modify. * Anything with a leading `c-` is a specific _thing_. ## Utility Namespaces: `u-` Format: .u-utility-name {} Example: .u-clearfix {} You will most likely be familiar with the Utility notation because of [SUIT](https://github.com/suitcss/utils). Utilities are complete [single responsibility](/2012/04/the-single-responsibility-principle-applied-to-css/) rules which have a very specific and targeted task. It is also quite common for these rules’ declarations to carry `!important` so as to guarantee they beat other less specific ones. They do one thing in a very heavy-handed and inelegant way. They are to be used as a last resort when no other CSS hooks are available, or to tackle completely unique circumstances, e.g. using `.u-text-center` to centrally align one piece of text once and once only. They are only one step away from inline styles, so should be used sparingly. Because of their heavy-handed approach, their global reusability, and their exceptional use-case, it is incredibly important that we signal Utilities to other developers. We do not want anyone trying to bind onto these in future selectors. Take the following example, which actually happened on a project I worked on. A number of months into a project, a developer wrote this bit of CSS: .footer .text-center { font-size: 75%; } Here we can see a problem: the `.text-center` class now has two responsibilities when it appears anywhere inside `.footer`. It now has side effects, which are something that Utilities should never, ever have. By using a namespace, we can introduce a simple and unbreakable rule: if it begins with `u-`, never reassign to it. Utilities should be defined once, and never need changing. Another problem that the Utility namespace solves is that it actually lets people know that there is a heavyweight rule being applied to the section of the DOM. It will help explain why certain things might be happening and hard to override. Take this example:
...
...
A developer inheriting this might be confused as to why the `blockquote`’s font size is different to what they expected. This is because it’s inheriting the font size from a `.font-size-large` class used a little further up the DOM tree. By adding a little more clarity to our classes, we can more quickly identify any potential offenders: Ah, here’s a Utility, that must be what’s causing it. (This is actually a fairly good example of why we should use Utilities sparingly.)
...
...
Please see this post’s sister article [Immutable CSS](/2015/03/immutable-css/) for more detail on these kinds of rule. * Utilities are style heavyweights. * Alert people as to their existence. * Never reassign to anything that carries a leading `u-`. ## Theme Namespaces: `t-` Format: .t-theme-name {} Example: .t-light {} When we work with [Stateful Themes](https://speakerdeck.com/csswizardry/4half-methods-for-theming-in-s-css?slide=29) (that is to say, themes that we toggle on and off) we normally do so by adding a class to the `body` element. Examples of this approach to theming include style-switchers (a user can toggle between different themes) and sub-sections of a site (all blog posts have one theme colour, all news pages have another theme colour, etc.). We simply add a class high up the DOM which then invokes that theme for that particular page. A simple way to denote any theme-related classes is to simply prepend them with `t-`. Seeing a `t-` class in your HTML should tell you that Ah, right, the view probably looks the way it currently does because we have a theme invoked. Now, all of the namespaces we’ve looked at so far are mainly of use to us in our markup, but Theme namespaces are helpful in both our HTML and our CSS. Seeing, for example, `.t-light` in our markup tells us that the entire DOM has a current state applied to it, which is important to know whilst debugging. Seeing that class in our CSS also tells us a lot: it helps to sandbox and isolate any chunks of theme-related CSS inside namespaced rulesets: .c-btn { display: inline-block; padding: 1em; background-color: #333; color: #e4e4e4; .t-light & { background-color: #e4e4e4; color: #333; } } Here we can see that our buttons have a light grey text colour on top of a dark grey background, but when we invoke the `.t-light` theme, those colours invert. Here we are encapsulating the style information, which means that finding, debugging, and modifying Theme rules becomes much simpler. * Theme namespaces are very high-level. * They provide a context or scope for many other rules. * It’s useful to signal the current condition of the UI. ## Scope Namespaces: `s-` Format: .s-scope-name {} Example: .s-cms-content {} Scoped contexts in CSS solve a very specific and particular problem: please be entirely certain that you actually have this problem before employing Scopes, because they can be misused and end up leading to actively bad CSS. Oftentimes it can be useful to set up a brand new styling context for a particular section of your UI. A perfect example of this is areas of user-generated content, where some long-form/prose HTML has come from a CMS. The styling of this kind of content usually differs from the more app-like UI around it. You may have a class-heavy UI architecture to provide complex pieces of design like navigations, buttons, modals, etc., and inside all of this you may have a simple blog post which is populated via a CMS where the user writes plain text and cannot add any classes or complexity. For a really terse but effective example of Scoping styles, see [David Bushell](https://twitter.com/dbushell)’s [Scoping Typography CSS](http://dbushell.com/2012/04/18/scoping-typography-css/). You **might** want to style this free-form text differently from the rest of the surrounding UI, so you _might_ employ a scoping context. For example:

...

...

...

    ...

...

Next article... Everything inside the `.s-cms-content` is inaccessible: we can’t get at the DOM to add any classes to the nodes inside of it, so we _might_ begin styling via a Scope. That _might_ look something like this: /** * Create a new styling context for any free-text CMS content (blog posts, * news pages, etc.). * * 1. Use a larger and more readable typeface for continuous prose. * 2. Force all headings to have the same appearance, regardless of their * hierarchy. * 3. Make links inside long text more apparent. */ .s-cms-content { font: 16px/1.5 serif; /* [1] */ h1, h2, h3, h4, h5, h6 { font: bold 100%/1.5 sans-serif; /* [2] */ } a { text-decoration: underline; /* [3] */ } } I cannot stress the word _might_ enough here. [Nesting selectors is bad](http://cssguidelin.es/#nesting): it leads to location-based styling, meaning that styles are now tightly coupled to DOM structure; it prevents people from being able to opt into styles, because nested selectors are very dictatorial (i.e. this **will** happen if you put that in there); having a type selector as a Key Selector creates very greedy selectors, which can match more of the DOM than you intend; and our specificity gets increased, meaning our Scope will override previously defined styles, and in turn the Scope itself becomes harder to override. There’s a really good example we can grab from the Sass above. When compiled, that code will give us this selector: `.s-cms-content a {}`. This selector is in charge of adding underlines to links, and is also of a higher specificity than a selector like `.c-btn {}`. This means that if we were to put a button inside of this Scope, it would get an underline—this is something we probably don’t want. This simple example outlines the potential for problems when working with Scopes, so tread carefully. Please make triple sure that that you need to employ a Scope before you start writing lots of nested selectors. If you are unsure, it may be best to err on the side of caution and leave Scopes out entirely. Warnings aside, the actual `s-` namespace becomes incredibly useful for signalling to developers that an entire area of the DOM is subject to one big caveat. Anything we see styled in here might have an extra layer of styling applied to it in a pretty opinionated and greedy manner. * Scopes are pretty rare: make triple sure you need them. * They rely entirely on nesting, so make sure people are aware of this. ## Stateful Namespaces: `is-`/`has-` Format: .[is|has]-state {} Example: .is-open {} .has-dropdown {} Stateful namespaces are lovely. They come from [SMACSS](https://smacss.com/book/type-state), and they tell us about short-lived or temporary states of the UI that need styling accordingly. When looking at a piece of interactive UI (e.g. a modal overlay) through developer tools, we’ll probably spend some time toggling things on and off. Being able to see classes like `.is-open` appear and disappear in the DOM is a highly readable and very obvious way of learning about state:
...
It’s also incredibly handy in our CSS to tell people possible states that a piece of UI can exist in, for example: .c-modal { ... &.is-open { ... } } .c-modal__content { ... &.is-loading { ... } } These classes work by chaining other classes, for example `.c-modal.is-open`. This heightened specificity ensures that the State always takes prominence over the default styling. It also means that we would never see a bare Stateful class on its own in a stylesheet: it must always be chained to something. The way in which States are different to BEM’s Modifiers is that States are temporary. States (can) change from one moment to the next, perhaps based on user action (e.g. `.is-expanded`) or from changes that are being pushed from a server (e.g. `.is-updating`). * States are very temporary. * Ensure that States are easily noticed and understood in our HTML. * Never write a bare State class. ## Hack Namespaces: `_` Format: ._hack-name {} Example: ._c-footer-mobile {} In certain and usually quite rare circumstances, we might need to add a class to our markup purely in order to help us hack or override something. If we ever do that, we need to signal that this class is hacky, it’s (hopefully) quite temporary, we want to get rid of it at some point, therefore do not bind onto, reuse or otherwise interface with it. The reason for the leading underscore is simply to mirror the paradigm of private variables in programming languages. Variables that are private to the program should not be relied upon or reused by other developers, and that’s the same with our Hack classes. These types of class are pretty easy to spot in our codebase, so any hacks will become very apparent, which is a [good thing](/2013/04/shame-css/). @media screen and (max-width: 30em) { /** * We need to force the footer to be a fixed height on smaller screens. */ ._c-footer-mobile { height: 80px; } } * Hacks are ugly—give them ugly classes. * Hacks should be temporary, do not reuse or bind onto their classes. * Keep an eye on the number of Hacks classes in your codebase. ## JavaScript Namespaces: `js-` Format: .js-component-name {} Example: .js-modal {} JavaScript namespaces are pretty common now, and most people tend to use them. The idea is that—in order to properly separate our concerns—we should never have styling and behaviour bound to the same hooks. To bind both technologies onto the same hook means we can’t have one without the other: our UI becomes all-or-nothing, which makes it very opinionated and inflexible. When I worked at [Sky](/case-studies/bskyb/), we had an incident where a developer had built a text-callout UI component that had a distinct appearance, and some behaviour to fade text in and out of it. A Product Owner asked that we reuse the same piece of UI elsewhere, but we didn’t need to fade multiple pieces of text in and out; it was just going to say the same thing all the time. Because the component had been built with JS and CSS binding onto the same hook, it meant that I couldn’t have a configuration of the component with its look and feel but without its behaviour. It took a chunk of refactoring to fix, and it could have been avoided simply by binding onto separate hooks. It also means that we can work a lot more safely. It means that CSS developers can work and refactor freely without the worry that they will break some JS, and vice versa. It separates our concerns and leaves each team with its own hooks for its own purposes. It’s probably also worth noting that because the JS namespace has nothing at all to do with CSS, its format should be determined by your JS engineers. If your JS team’s naming convention for variables etc. is camel case, then they should be allowed to choose JS hooks like `.jsModal` if they so desire. * JavaScript and CSS are separate concerns—use separate hooks for them. * Giving different teams/roles different hooks makes for safer collaboration. ## QA Namespaces: `qa-` Format: .qa-node-name {} Example: .qa-error-login {} An unusual, but potentially very useful namespace is this one, for your QA team. When running automated UI tests with something like [Selenium](http://www.seleniumhq.org/), or a headless browser, it is quite common to do something like: 1. Visit `site.dev/login` 2. Enter an incorrect username. 3. Enter an incorrect password. 4. Expect to see an error appear in the DOM. I’ve had problems before where the authors of these automated UI tests were binding onto CSS classes: e.g. Does `.message--error` appear in the DOM? The problem with these tests looking out for style hooks is that simply refactoring your CSS to use a different name can cause a test to fail, even if the functionality is exactly the same. In a similar vein to our JS hooks, automated UI tests should not be reliant on CSS classes. To do so breaks our separation of concerns. What we need to do is have the QA team bind onto a suite of their own classes that we leave well alone. This means that if we start out with this: