I am confident that adopting a client-side framework through progressive decoupling will give us the best of both worlds. Of course, this does not mean I oppose fully decoupling through any other framework; in fact, I believe we should redouble our efforts toward a first-class API-first Drupal. But progressive decoupling means that we will be able to work toward a next-generation user experience without abandoning much of the work we've done so far.
Special thanks to all of the following experts who provided review and input: Miško Hevery (creator of Angular; Google) and Igor Minar (technical lead for Angular; Google); Ed Faulkner (core maintainer for Ember); Amitai Burstein (Drupal and Elm contributor; Gizra); Sebastian Siemssen (Drupal contributor, Elm and React developer; Zensations); and John Albin Wilkins (Drupal 8 mobile initiative lead), Alex Bronstein (Drupal core maintainer; Acquia), Wim Leers (Drupal core contributor; Acquia), and Preston So (Drupal contributor, Acquia).
First, have we decided on the right criteria regardless of the frameworks themselves? This is probably the most important at this stage. While many organizations choose to adopt client-side frameworks for fully decoupled implementations, Drupal is the first to consider layering a framework on top to allow both richly dynamic and more traditional modules to coexist gracefully through progressive decoupling. What issues around templates, rendering, and client-side caching should we incorporate into these criteria? Is there anything missing or overemphasized?
Finally, have we drawn the right conclusions against these criteria? In other words, did we fill out the cells correctly? While they have been reviewed by some of the frameworks' experts, there might be unexpected gotchas or caveats.
At the moment, the most promising candidates in the comparison matrix appear to be Angular 2, Ember, and React, given their technical robustness, relative suitability for progressively decoupled Drupal, and their strong levels of community support and broader adoption. Given that Backbone is already in core and several modules already rely on it, we have included it too.
What we've learned from talking to the different projects is that they are often converging on similar techniques and best practices; they are by and large adding support for Virtual DOM implementations or rehydration (seamless state transfer), and they are all obsessing over small payload size and performance, better testability, etc. Therefore it is important to focus on the fundamental, often philosophical, differences between the projects that will likely be unchanged in time; key architectural differences, their release cadence and stance on backward compatibility, their license, their governance model, their flexibility and learning curve, etc.
From a quick glance at the criteria and our needs, it seems that Ember is currently our best candidate, as it appears to have a slight technical edge overall. Ember 2.0 has an all-new rendering engine named Glimmer, and it has server-side rendering through FastBoot. On the other hand, however, Ember is quite bulky and opinionated (enforcing patterns for code structure) compared to other candidate frameworks. A more fundamental difference is that unlike Angular and React, which have corporate governance and funding, Ember is a community-driven project like Drupal.
While React is lightweight, it needs integration with a variety of other libraries in the React ecosystem to work as a full-fledged implementation, which gives it a steep learning curve from an implementation standpoint. Because React is a relatively young project, best practices are shifting quickly and making it less attractive. The Virtual DOM, among React's most compelling features, has also seen its core ideas filter into other framework projects. But more importantly, React is licensed with what I believe to be a potentially unacceptable patent clause, which states that an organization can no longer use React once it sues Facebook for any (unrelated) patent infringement. This has already generated some concerted pushback from both WordPress's Calypso and React contributors.
Whatever the result of the debate around which client-side framework to adopt, I believe that Drupal needs to move quickly to embrace a framework that will aid development of a progressively decoupled architecture and corresponding user experience improvements. By providing some baseline criteria and including our accomplished community, I have no doubt we can reach this decision quickly but also intelligently.
Special thanks to Preston So for contributions to this blog post.