Jan 17 2012

I love both Sinatra and Rails, for different reasons. I’ve heard a few different heuristics for which framework would be better for your application, but I’m not sure the answer is all that simple, really. Regardless of which is correct for your application, I haven’t seen a lot of people discussing how to move between the two frameworks.

Step One: Evaluate your test coverage

In Working Effectively with Legacy Code (this is an affiliate link. Does this bother you?), Michael Feathers introduces a fantastic technique called “Lean on the Compiler.” It’s on page 315, for those of you playing along at home.

The primary purpose of a compiler is to translate source code into some other form, but in statically typed languages, you can do much more with a compiler. You can take advantage of its type checking and use it to identify which changes you need to make. I call this practice Leaning on the Compiler.Lean on the Compiler is a powerful technique, but you have to know what its limits are; if you don’t, you can end up making some serious mistakes.

Now, in a language like Ruby, we don’t have type checking. We do have something that double checks our code correctness: tests. Just like static type checks, tests cannot ensure that you’ve done your transformations, but they sure as hell can help. This becomes step one in the effort to move from one framework to another: evaluate your test coverage.

Acceptance Tests are most important

When looking at your tests, first check out your acceptance tests. These are the most important, for two reasons: they’re abstracted away from the framework itself, and their purpose in life is to make sure that major functionality is working. Do you have your happy paths covered? Is there any major functionality that’s not covered by acceptance tests?

While the happy path is a big deal, when moving to a new framework, we’re going to introduce a high chance of people encountering issues, so the sad path is also important. I personally tend to not write very many sad path acceptance tests and leave that for tests at the lower level. This is a good time to take stock of the worst of your worst paths: is there certain functionality that’s absolutely important to be handled 100% correctly? Then write some new acceptance tests. As with any kind of engineering project, there’s a tradeoff here: you can’t get coverage of every possible situation. Tests are supposed to give you confidence in your code, so liberally add coverage for any situation that makes you nervous.

Integration tests are important too

These tests are a bit more tricky, as they are tied into the framework that you’re using. I’m not sure that anyone writes many of these kinds of tests with Sinatra, outside of maybe model to model integration tests.

Model and View tests: not a big deal

These kinds of tests are much more of a sanity check than anything else for the purposes of this kind of move. It’s good that they pass, but really, they shouldn’t be too reliant on the framework you’re using. They’re better for making sure you’ve put things in the correct directories and haven’t forgotten anything in the move.

I’m not even sure what a ‘view test’ would be in Sinatra, and they tend to be not in favor with Rails projects anyway, so that’s more of a general ‘framework to framework’ bit of advice than anything else.

Controllers don’t exist in Sinatra…

… so you don’t really write those kinds of tests. Controller tests don’t seem to be that popular in Rails-land these days either.

Step Two: Git Can help

I’ll show you how I managed transitioning files over. However, I’m really interested in the idea of using a subtree merge to keep being able to update the original Sinatra project while working on the new Rails project.

First step is, of course, make a new branch. This transition will probably take some time, and you may want to push hotfixes into production during that period. Using master is not advised.

Next, make a copy of everything and shove it in a temp directory. Then delete everything in the git repo (except the .git directory, of course) and commit that blank slate. Back up one directory above your project directory, run rails new myproject with your project’s name, and you’ll get a blank rails app. cd myproject and mkdir sinatra, then copy the backup you made from the temp directory into the sinatra directory. Finally, commit this.

Now you’ve got a new blank Rails app with all your old code in a directory, and you can start moving things over.

Step Three: Set up your test harness

Since we’re going to allow our tests to guide our move, it pays to get tests up and running first! Depending on what testing framework you use, get it going with Rails. In our case, we were using minitest to test our code. This took a little bit of effort to get working with Rails, but there weren’t a ton of problems.

As always: red, green, refactor. I’d make a simple test that doesn’t test anything, assert true. Make sure that your rake test or rspec . or whatever you’ll use to run the tests works and then remove the dummy test.

There are two strategies for moving tests over: you can move a chunk at a time or move one at a time. Chunks are easier, but then you get commits where the build is broken. It’s really up to you and your team’s tastes: a huge test suite with a number of failures can show you how close you are to being done with the first 90% of the work. And since you’re not planning on releasing in this half-finished state, having a broken build is not that terrible… As always, you’re the professional: make the call.

Step Four: Move your models

Since models are the simplest to move, I like to do them first. You should just be able to copy over each test file and model file, though because you were using Sinatra, your models may be all in one file. This step should be largely painless, though: you’re not really relying on any framework-specific things.

Model tests can still suss out problems with your environment, though, like incorrect database settings, certain environment variables…

The idea to begin with the easiest thing comes from Dave Ramsey, oddly enough. My aunt works at a bank, and when I went to college, she bought me a few of his books to help me learn about personal finance. Dave is a big fan of the ‘get rid of all debt’ school of thought, and so a large portion of his work is strategies for getting out of debt. Dave contends that paying off loans with the highest interest, while mathematically the fastest way to pay off your obligations, is not actually the best strategy. People like to see progress and are heartened by seeing it. By paying off the smallest loan first, one is much more likely to feel more positive about the headway one is making, and so it’s a much better strategy in the end.

The same logic applies with this move: why start with the hard stuff? Knock out some quick wins to keep your spirits high.

Step Five: Convert your controllers

Next up: get your controllers created. This is a pretty manual process:

get "foos" do
  @foos = Foo.all
  render :"foos/index"
end

becomes

class FooController < ApplicationController
  def index
    @foos = Foo.all
  end
end

The biggest issue with moving all of this stuff is that Sinatra and Rails use redirect and redirect_to, respectively. So you’ll have to convert that stuff. However, I wouldn’t recommend changing things like redirect "/foos/#{id}" to redirect_to foo_path(foo) just yet. When dealing with legacy code, you want to change as little as possible with each step so that you know when you have introduced an error. If you try to convert things to a Rails style as well, you run the risk of introducing errors. Therefore, in all of these moves, leave the hell enough alone as much as possible. Once your code is up and running, you can gladly refactor. Just don’t do it now.

Don’t forget to generate your routes, too. Sinatra’s DSL is like a combination of Rails’ routing and controllers. Set those up in this step as well.

Since we don’t have tests, this part is very error-prone. Luckily, our acceptance tests will catch these issues in step seven. So give it your best shot, but don’t worry about being 100% perfect. Focus on getting the basic structural changes in place.

Having tests is so nice. :/

Step Six: Move your views

This should be as simple as the models: put your views in the correct directory, and things should be golden. If you were using inline views with Sinatra, well, you shouldn’t have that many of them, so breaking them out should be pretty easy.

Step Seven: Listen to your tests

Okay! Those last moves were really awkward since we didn’t have tests to check our work. This is where the acceptance tests come in. You can move these over in batches or one at a time, but the acceptance tests will tell you where you forgot a route, if your view is missing, or if you left some little bit of Sinatra-ness in your Rails somewhere.

You’re almost home! Once you get all your acceptance tests working, you should pat yourself on the back! You’re not done yet, but you’ve made a lot of progress.

Step Eight: Track your exceptions!

You should have been doing this anyway, but if you weren’t, set up some sort of mechanism to catch and handle exceptions. Airbrake is one such service, but even just recording them yourself would be fine. You need to have this set up in some form, as you’re likely to generate errors, and examining your exceptions is the best way to track down actual problems.

Step Nine: Plan out deployment strategies

Of course, now that you’re done developing, it’s time to get the app out to users. This’ll require a comprehensive plan for actually making the jump. A little bit of foresight can go a long way here.

My strategy was to roll out the new Rails app under a new domain: http://beta.myapp.com/. I then pinged my users with a message along the lines of “We’ve made some infrastructure updates, if you’d like to help us test them out, visit the beta site.”

This approach does have some sticky points, however. The first one is the database. In my case, I had a high level of confidence in my code, and this application wasn’t for anything that was considered mission-critical. We also had a decent backup strategy in place. Therefore, I connected the beta site up to the production data store. This meant there were no migration issues later. However, this may make you uncomfortable, but there are other options, too. You can treat the beta as a sandbox, and tell people that the data will not be persisted after the beta, or you can migrate people’s data back to the production data store afterwards.

Another approach is to automatically migrate a portion of your users over to the new code base. Ideally, nobody notices: your code base shouldn’t have changed in looks or functionality.

Step Ten: Push to production!

Congrats! Execute on the plan you created in step nine, and get ready to answer tons of emails about why your app is broken! Just kidding, I hope! I really recommend the beta strategy for this reason. No plan survives first contact with the enemy. Your users will find things you didn’t, no matter how much manual testing you did along the way.

Another strategy: mounting

I haven’t actually tried this myself, but I’ve been thinking about another way to make this happen: mount your Sinatra app inside of a Rails app, and use that to move things over slowly. Here’s an example of how to make this happen technically. If you had a really big Sinatra application, I could see how this might get you deploying faster; just carve out one vertical chunk of your app, move it over, and keep the rest of your app running in Sinatra.