Skip to main content

Components for Everyone

13 Sep, 2021

TL;DR This is not about web components / custom elements, rather why component-driven architecture for traditional server-rendered apps could be a better solution than adopting a client-side framework.


Recently, I've started to form the opinion that client-side JS libraries, like Vue, React, Svelte, etc*, have grown in adoption within traditionally server-rendered applications due to component-driven view architecture more than any other feature they offer. I believe this core abstraction is key to why larger frameworks, like NextJS, Nuxt, Gatsby, and SvelteKit, are built atop those libraries with a targets of static site generation (SSG) or server-side rendering (SSR). This motivation can lead to further complexity around managing a distributed client-server codebase and constant context switching for full stack developers.

For context, I've been using React for ~6 years (starting as the view layer replacement for a Backbone app) and currently work on a front end platform team that supports product squads using Rails and/or React to develop client-side features, depending on the experience of the team and requirements of the feature. Before adopting React a couple of years ago, engineers at the company developed a Ruby component abstraction that allowed them to create a shared gem of design system components** for use across the production Rails applications. Even with React in active use, we still have this Ruby abstraction and are starting to adopt ViewComponent in its place to resolve issues around lack of documentation and surrounding tooling expected by devs who have used component-driven frameworks in the past, i.e. Storybook.

With that in mind, I wondered how many other teams chose to build features using a client-side JS framework primarily for built-in component model, consciously or not. The more I used a server-side component system, the more I recognized the things I liked about building React components, without the need to context switch between Ruby and JavaScript in a single feature.

This led to some lazy web research on Twitter to see if other folks thought about this subject; if so, where are they seeing this workflow in other frameworks?

While no one provided any other examples, outside of using SSR-only with one of the popular JS libraries I mentioned previously, there was enough engagement to continue exploring the wider web.


Before diving into the various language implementations I found, I want to break down the core advantages the component model offers application devs. When I refer to "component-driven view architecture", I am including the surrounding ecosystem of patterns and tooling as motivation for its accession:

  • isolated development playground (Storybook / Pattern Lab)
  • unit testing (quickly verify UI logic without spinning up a browser)
  • packaging (ship as a reusable resource/gem/pub)
  • sidecar assets (easily scope styles and behavior)

Each of those values can help scale the efficiency of teams building user interfaces, no matter the platform technology.

Taking a look at open source design system components from large Rails shops like Primer from GitHub and Polaris from Shopify, the alternative to using JS components is usually collection of CSS classes; this can lead to inconsistent usage when applying them to markup and a worse developer experience. The Primer ViewComponents gem is still fairly new when compared to the CSS and React implementations of that design system but solves the job of providing reusable pieces of UI in a familiar interface to Rubyists, including an interactive Storybook using @storybook/server!

Although I continue to mention the benefits in terms of pattern libraries, they can also be appreciated within a single app codebase to share common functionality across features.


Ruby / Rails

I have mentioned ViewComponent several times already, as the inspiration / motivation for this post. The docs do a great job explaining the philosophy behind the framework:

Traditional Rails templates have an implicit interface, making it hard to reason about their dependencies. This can lead to subtle bugs when rendering the same template in different contexts.

ViewComponent Data Flow

It has been growing in popularity since its introduction by GitHub; looking at the open source projects using ViewComponent:

But that's not the only solution on the block, especially when looking to build a static site. Bridgetown bills itself as "a fast, scalable, modular, and thoroughly forward-looking framework for building websites and frontend applications" and a modern successor to Jekyll. Although it also supports ViewComponent, the Bridgetown team is the first place I've seen an implementation of components for the Liquid templating language. Liquid component does not go as far as ViewComponent in providing a Ruby class as the interface for a component, but uses YAML front matter well to document and describe pieces of reusable view logic.

PHP / Laravel

It appears there has been a concept of "components" in Laravel's Blade templates system since version 5.4; however they are more comparable to partials with support for slots until version 7. This current iteration of Blade component is backed by a PHP class with an associated template (or an inline component view), which is more similar to ViewComponent. These components are rendered using a familiar kebab-case, HTML tag-like syntax: <x-alert>.

Python / Django

django-component provides "declarative and composable components inspired by javascript frameworks". They are made up of a Python class and associated HTML template, like the previously mentioned ViewComponent and Blade components.

There is also a library called slippers, that reminds me of liquid components more than ViewComponent since they don't rely on some kind of class or backing data structure defined outside the template.

.Net / Blazor

As a modern replacement for Web Forms, Blazor supports UI encapsulation through components using the Razor templating language to define markup and logic in a single file; although that logic can be broken out as a separate C# class.

Elixir / Phoenix

The Phoenix.Component API allows Elixir developers to create function components, similar to React, including ability to pass in "assigns" (props) and "blocks" (children).

Surface UI builds on top of the Phoenix component API to provide "a fast and productive solution to build modern web applications". It uses an extended templating syntax, helpful macros, and built-in components to become a more full-featured view framework for Phoenix applications.

JS / Node

Lastly, we get to the curious case of server-rendered component systems for NodeJS. The closest library in comparison to the previous examples is Marko, which allows developers to "start with simple HTML templates and add powerful components as needed". Marko templating can be integrated into popular server libraries as the core view engine, rather than forcing a typical client-side framework into that workflow. There was an official view engine for React in Express, but it no longer appears actively supported; React Server Components are also still theoretical at the time of writing.

There is active development in this space with the advent of tools like Astro and Slinkity that aim to take a framework-agnostic approach to introduce components into modern, HTML-first static sites. Astro does a great job comparing itself to existing tools and Slinkity is building on the shoulders of the awesome Eleventy ecosystem (slight bias given this site is currently built with Eleventy).

Previously, it seemed like NodeJS devs didn't invest in a server-side component framework because existing client-side libraries could be worked into the SSR pipeline; why reinvent the wheel? There is less of a context switch in terms of syntax when compared to the workflows of other web languages, so the friction lies in the uncanny valley of managing a client-side component system in a traditional server application. Even if a library like React or Vue is used as SSR-only, the APIs were designed for the front end, which can lead to confusion about how to constrain the usage of third-party libraries and patterns.

Who knows where the future is headed in this use case? Could there be some type of mustache.js extension or Handlebars helper for rendering a JavaScript class in a template? That could be pretty neat!


This post is not meant to denigrate client-side component frameworks; I love developing front end applications with JavaScript and will continue to do so. But choosing to build with that stack comes with additional complexity that not every team is ready to handle, so there should be useful alternatives for server-rendered ecosystems and take advantage of that environment. I hope we see more tools like Storybook Server being used to develop those server components in isolation and generate documentation for easier discoverability. There should be a rich collection of accessible, open source component libraries, similar to Headless UI or Radix UI, so apps can ship features that are enjoyed by everyone by default.

If I missed something, like an example implementation or assumed detail, please reach out on Twitter to help me correct it here. I'll continue to shape these thoughts, put them into practice through personal and professional projects, and follow up with more articles to share any lessons learned.

* I purposely left out more complete frameworks like Ember and Angular since they focus on shipping full client-side applications out of the box.

** This was created before my time at the company but is explained well in this Rails Conf talk from 2018: