Sunday 3 November 2013

No, You Can't Always Get What You Want...

…but as Mick Jagger and Keith Richards remind us, if you try "sometimes you get what you need". Any software developer lives with that on a daily basis; the implications of that are one of the major things that separate our craft from the profession of engineering. I had several of those "implications" blow up in my face this week as I was working on (what else?) a Rails app.

Or, more properly, an app that uses Rails. Because what the majority of my time has been spent with for the last few weeks has not been Rails, or even Ruby, but trying to get the front-end features that have to be implemented in JavaScript (or CoffeeScript or another language that compiles to JavaScript) (generically here, Script) because they're dependent on user-agent and DOM events. The last time I went down this road, I wound up with an app that had five lines of CoffeeScript for every two of Ruby; this one is well along that path. That's relevant; I suspect that there are far more non-trivial apps that use Rails in production than there are Rails apps; for anything beyond a simple pseudo-blog or relatively straightforward database front end, you sometimes wonder what the early Rails core team was thinking when they established some of these "conventions" that trump "configuration". Granted, there are a lot of people using that "sweet spot" for Rails app development, and more power to them, but I'd bet heavily that there are at least as many who started out trying to do an app that uses Rails and fairly quickly find themselves working around or against things as much as they work with them.

A prime example of this is front-end work. Rails has support for *Script and the asset pipeline as ballyhooed features. Once you start writing more than a few dozen lines of Script code in three or four classes or modules, you quickly come to the conclusion that the people who designed and held court over how Rails works really didn't have much more Script experience than using snippets of jQuery or Prototype.js to "jazz up" a form. Heck, the "unobtrusive JavaScript" feature is essentially useless outside forms. What I'm working with at the moment is an app that intelligently reacts to a user selecting content with the mouse, based on the position of the content, the permissions and authorisation of the user, and other factors. Having a few Script classes (and associated specs) to deal with that introduces issues that may not be readily apparent in the beginning, but will bite your team over time.

The JavaScript community have developed several module loaders to work around the lack of such a feature in the core language (for now, at least); there seems to be something of a transition underway from CommonJS to RequireJS, with several alternatives providing variations on the theme. This is important for several obvious reasons, including the ability that programmers have enjoyed in various languages since at least the 1960s to develop, test and maintain cooperating code modules (objects, classes, etc) relatively independently. By having each such module load, and know about, only the other modules it uses, those modules (should) remain unaffected by other modules in the system, even indirect dependencies that continue to work as their dependents expect. Programming 101 stuff; nothing new for anyone reading this.

But, and it's a very large but, the Web prizes a site's ability to perform as few network accesses as practical to get its resources loaded and start doing useful work. To this end, Rails since version 3.1 has included the asset pipeline, which has the effect (broadly) of concatenating and compressing Script files, CSS stylesheets, and other resources so that they can all be loaded in a single request. This is done using an add-in Gem (component) called Sprockets. This works quite well for its intended purpose and truly is one of the nicest things about working in Rails — until you start writing modular Script code.

If you know that the asset pipeline, used by default, means that everything is loaded by the time any of your Script front-end code starts running, but you want to break your code into modules, the "obvious" choice is to create a lot of "global" objects for your classes and shared data and such. If you're an experienced developer, this feels wrong, almost dirty. You might then try only having a single global that functions as a top-level "namespace" for your application, with artefacts spreading out in a sensibly-organised tree below that. This path leads to names like Maintaining such, or even keeping it straight in your head (and your teammates') quickly becomes an "interesting challenge". You quickly realise that there must be a better way, simply because there have been other developers before you who have remained (apparently) sane.

But to use loaders such as RequireJS from Rails, you need components like requirejs-rails, which is wildly, justifiably "popular". Sane Script dependency management should be in Rails, you start thinking. And I'd agree with you. However…

The Might-As-Well-Be-Official Rails "Solution" to the problem is: don't do it that way. Write your front-end code in a traditional Script framework like Dojo or Node or Ember or JoeBobBriggsAmazingFramework or any of no doubt numerous other excellent choices. Use Rails "merely" as the back-end API server for persistence and auth and so on. And yes, with planning and preparation and an adequately-staffed and -trained team starting well before your deadline, that's a splendid choice. You may even realise that Rails is overkill for your needs and move to something like Sinatra or Padrino or, again, numerous others.

But…what if you don't have that time or that team, but you do have a mass of rigidly-organised, fragile Script code that you just promised would be part of a working app in two weeks' time? Assuming that you agree that resignation and suicide are suboptimal strategies, you could try what I'm in the middle of retrofitting our code to:

  1. Create a single, gateway object that you use to get all your other objects. Let's call it a Gateway object. give it a global location at @ourapp_gateway. Give it two methods:
    1. a register method, that associates a string name with an arbitrary object/class/other collection of bits;
    2. a use method, that takes a string parameter and returns the item associated with that name.
  2. Remove the Sprockets #= require modulename lines from each of your existing Script files, and replace them with immediate calls to @ourapp_gateway.use corresponding to each of the objects you need from Your Outside App, as well as calls to @ourapp_gateway.register to register the object(s) declared in that source file with the Gateway;
  3. Add Sprockets #= require directives to your application.js, so that Sprockets actually loads the suckers;
  4. Assign the return values from @ourapp_gateway.use to variables as would be done in Node or whatever.

Why bother, I hear you thinking. Because, even without a proper module loader, this still delivers at least one major benefit: it abstracts away the relationship between your artefacts, the DOM, and the asset pipeline. It gives you reasonable flexibility in moving things around within the DOM and with respect to source files by having the Gateway instance as an intermediary. And, if you're thoughtful, diligent and very, very lucky, it gives you a leg up when you do get around to using a proper module loader, by letting you slip that in behind your Gateway abstraction.

And if you think I'm getting somewhat jaded/disillusioned with my continuing Rails experience and integrating a Script-using front end that's not a complete app in itself…congratulations; you nailed it.

Anybody have any better ideas that they've actually made work?

No comments: