Levelling up our client side developer experience
Developers spend a lot of time improving the performance, features and overall experience of our websites for the benefit of our users. And rightly so. However this can be to the detriment of the other people who use our applications – namely, well, ourselves.
Developers developers developers
A few of us, however, specialise in client side development. We often found ourselves asked quite reasonable questions. Questions such as:
- How do I install dependencies? Do I need to run npm install, or are the dependencies checked in somewhere? Or are they installed as a WebJar? (again, all three –and don’t forget we use yarn now)
- How do we write components? Are we using React yet, or Preact, or do we have an undocumented bespoke solution of our own? (take a guess)
These questions, and their harrowing answers, seem to derive from two approaches: keep existing features stable; new features make good test beds for new technologies.
It meant there was no one way to do things: it was the wild west. Something needed to be done.
Defeat in detail
There were too many problems to tackle all in one go, so we did what any reasonable group of web professionals would do and borrowed a strategy from Napoleon. We brought our meagre client side developer force to bear against each problem in isolation, with the hope of eventually destroying our enemy as a whole while exposing ourselves to the smallest possible risk. We wrote a Client Side Developer Experience Defeat in Detail Document (a title we catchily shortened to CSDXDIDD) that captures each of our tech debt enemies, and explores what needed to be done in order to defeat each one.
Now when a developer clones our frontend repo, all they have to run is `make compile` and sit back as all the dependencies are installed and the application is built.
Our application uses AMD modules bundled by the requirejs tool r.js. We wanted to move to a bundler that supported a more modern workflow, integrating tools like Babel and hot reloading. Ideally we’d be able to take advantage of the latest language features in ES6 and ES7, and incorporate static type checking using Flow.
We identified Webpack and Rich Harris’s Rollup as our potential bundlers of choice.
After running some tests, we noticed that the execution time of bundles built with Rollup was significantly shorter than those built with Webpack. Bundle sizes were smaller too. Many eyebrows were raised.
However after some analysis, we ended up choosing Webpack for two main reasons. Firstly Webpack has a larger community, is more mature and has more support for larger applications such as ours. Rollup’s target audience is currently smaller applications and libraries, lacking support for essential features such as code splitting and commons bundles.
The second reason is that Webpack maintainers are looking to utilise the Rollup engine, or at least Rollup’s most successful strategies of live code inclusion (AKA tree-shaking) as opposed to dead code elimination, and hoisting all bundled modules into a single scope. This will hopefully bring Webpack bundles sizes and execution times into line with Rollup’s.
Something to watch out for when migrating from RequireJS/r.js to Webpack is that you will need to find a new way to load external scripts. This proved slightly problematic for us during migration as some of the external libraries have state, which needs to be shared between the newly growing Webpack application and the remaining RequireJS/r.js application. We solved this problem by exposing the libraries and functions shared between the methods on the window object, allowing them to all call the same singleton.
In the near future, we will turn our attention to our neglected component architecture.
Since our HTML is server-rendered using the Scala Play Framework, it seems like overkill to serve a heavyweight framework such as Angular or React to handle presentation logic to the client.
We have experimented with Preact, a skinny clone of React, but we are becoming more excited by “frameworkless” approach exemplified by Svelte. This fits in well with our modular, microlibrary strategy, borrowing from the Unix playbook by ensuring each of our dependencies are focussed on achieving one goal, and can be switched out for an alternative at a later stage without too much refactoring.
We’d love to hear your thoughts and stories relating to client side developer experience.