Web Development Principles: Progressive Enhancement
Posted on May 31, 2015
Progressive enhancement is quite an old topic in web development yet it seems to resurface every few years as we break things. In a nutshell, web applications should not be developed as if the end user always has a powerful machine and a fast internet connection like you have as you sit and develop. My eyes were really opened up to this when I first visited my team in Santiago, Chile and saw that Juan was rocking an iPhone 3GS on a slow cellular network. It had been easy for me to forget that not everyone lived in a high-tech town like Huntsville with the latest smart phones, LTE networks, broadband home internet connections, and fast PCs.
I was mindful of this philosophy when I wrote the code for serving this blog. One thing you may notice is the panel on the right (assuming you’re on a desktop) loads via ajax once the container element is visible. To see this, scroll to the bottom of this post and refresh the browser. Scroll/page up and near the top you will momentarily notice a spinner before the right panel pops into existence. I did this because I used a responsive layout which will push the right panel below the post on a small device. This way I delay loading that part of the page, possibly never loading it for visitors who don’t finish reading the post. I haven’t burdened that page load with the twitter feed, Google calendar, etc which will bog down the page load and eat up more data. Once a user does find the bottom of the page, the right panel will load and allow navigation to more content.
Progressive enhancement is not a free or easy approach, and in some cases doesn’t make much sense. My team’s project PartQuest is a great example of not needing progressive enhancement. The application does not have a sensible mobile use case worth investing in. We don’t actually own the content we’re serving (it is retrieved from a 3rd party SOAP API), and we have no value in allowing crawlers to index it. In fact that content is not even accessible unless the user is logged in. The application is a complement to Mentor Graphics’s desktop applications, which already require a respectable machine to operate. We expect such workstations to always exist in an office with a good internet connection. In my mind the only thing we could gain from progressive enhancement is the resilience against failure. While this is important, plenty of other factors weighed in where we felt that Angular would buy us enough benefit to forgo this principle.
I believe that most applications would benefit from progressive enhancement as the default philosophy. Web application libraries should have a happy path which walks the code base down a road of progressive enhancement yet without preventing deviation when desirable (in general, you’ll find I always insist that friction to deviate should be minimized). Many front-end application frameworks have taken us away from this philosophy entirely, and I don’t think that is a good thing. I completely understand the benefit of using them (I did pick one for my day job), but I feel we’re too eagerly throwing something away without thinking enough about it. I know I’m not the only one thinking this is a problem for our client-side rendering frameworks. I doubt that server-side prerendering would be such a big topic right now otherwise.
The implication here is that the page should be rendered on the server fully for any data which is readily available. Means of loading data later with ajax should likewise be provided, but should not be the default approach. The idea of always serving a skeleton, letting a client-side framework initialize, and calling back home across the entire internet for the data is silly. What we need to do is render that page to the client ready to rock whether it be a powerful modern browser, a limited mobile device, or a web crawler. Then we can use our JS/ajax magic to update the DOM on the fly for an enhanced user experience.
While I think Angular has served my PartQuest team well, the approach of client-side rendering is not something I want to do again. In my dream web stack, we still could have succeeded because its preference of server-side rendering would not have prevented us from serving everything but the data which loads later. My ideal web stack will definitely be serving the pages ready to render from the server.