I had an idea this morning that I could make a simple desktop app by combining the lightning-fast website generation capabilities of jekyll with the awesome Node-Webkit that lets us native wrappers for HTML5 apps. I checked out this nice intro to Node-Webkit, and unsurprisingly ran into a few gotchas, documented below for other adventurer and my future self:

Simple Website with Jekyll

gem install jekyll
jekyll new experiment
cd experiment
jekyll serve

go to http://localhost:4000
you should see the default sample jekyll site

Make a Native OSX App

Download and install Node-Webkit pre-built binary

At the root of your jekyll site, create a file named “package.json”

  "name" : "nwk-experiment",
  "window" : {
    "width" : 800,
    "height" : 600,
    "toolbar" : true
  "main" : "app://whatever/index.html"

The app root url is a nice feature of node-webkit which makes it pretty easy to transport any website into this system of building a native app.

jekyll build  # creates the site
cd _site
zip -r ../app.nw * 
cd ..

Most tutorials tell you to zip the directory. The first time I tried, I got an Invalid package error “There is no ‘package.json’ in the package, please make sure the ‘package.json’ is in the root of the package.” On OSX, we need to zip the files from the root of the directory that had the ‘package.json file in it. (via crashtheuniverse)

Run the App

open -n -a node-webkit "./app.nw"

When I double-click on the app.nw file, I see the directory, not my index file. I haven’t figured out that part yet. Still a work in progress!

I have a relatively short list of cities which I want to plot on a world map. The list is a little too long for a manual lookup, but I don’t know exactly how I’ll use it in an app, so I figured out how to do it with some simple ruby in irb using the lovely geocoder gem.

I had my cities in a spreadsheet. Selected a few cells, and was able to simply paste into irb and split the string on newlines to get an array of the cities.

gem install geocoder
>> require 'geocoder
>> a = "Honolulu, HI
>> Boston, MA
>> New York, NY".split("\n")
 => ["Honolulu, HI", "Boston, MA", "New York, NY"] 

>>a.map do |city| 
    d = Geocoder.search(city)
    ll = d[0].data["geometry"]["location"]
    puts "#{city}\t#{ll['lat']}\t#{ll['lng']}" 

Honolulu, HI    21.3069444  -157.8583333
Boston, MA  42.3584308  -71.0597732
New York, NY    40.7143528  -74.00597309999999

Then I could copy/paste the irb output back into my spreadsheet. Ta Da!

It appears that the free Google API has some kind of throttling, so this is only good for short lists of <20 cities.

I’m using capybara-webkit for integration testing with rspec, which is awesome, because it is faster than other full-browser testing solutions, like Selenium, but it is slower than RackTest (default for RSpec testing). RSpec provides a nice way to specify an alternate driver when running Javascript tests, but configuration can be a little tricky. I got this working via this excellent blog post via Sarah Mei, who was pairing with me for the day.

I think it is important to actually understand the code that I copy/paste, so I took a little time to read up on the details which I’ve summarized below.

Favorite Testing Gems

I won’t elaborate on RSpec, the concepts in this post likely applies to test-unit as well. I written before about why RSpec is my favorite.


When using RSpec in Rails, we use the rspec-rails gem, which configures a bunch of stuff that makes it easy to get started. By default, integration tests will use rack-test, a lovely little gem that supports methods like get, post, put and delete and handles the rack request and response objects. (via platformatec) It maintains cookies and follows re-directs, but is far from a full-browser, most notably, pages won’t execute Javascript. Rack::Test is quite lightweight, with tests running in the same process as your Rails code. Its speed is a huge advantage and worth retaining for your tests that don’t need more.


Capybara is the world’s largest rodent, and the Ruby community is also the name of a favorite gem for testing the content of web pages. (A successor to Bryan Helnkamp‘s WebRAT, so named for Ruby Acceptance Testing, which instigated the rodent naming theme) Capybara is wonderful with its support for many “drivers” which allow for a consistent API across different solutions that offer different levels of browser support, with different performance characteristics.


Thoughtbot kindly created the capybara-webkit gem a few years ago, which I’ve found to be more reliable and performant than Selenium, and is my favorite choice for testing pages that need Javascript.

One of its creators, Joe Ferris, explains how it works (via stackoverflow)

  1. Capybara boots up your rack application using webrick or thin in a background thread.
  2. The main thread sets up the driver, providing the port the rack application is running on.
  3. Your tests ask the driver to interact with the application, which causes the fake web browser to perform requests against your application.

Database Cleaner

The DatabaseCleaner gem is super helpful for our typical Rails app that relies on a database. We always want a “clean slate” when we start our tests and this nifty gem gives us a bunch of options with a consistent interface for various database choices.


To configure these solutions correctly, it is critical to understand that with Capybara::Webkit our target app code is running in a separate process from our tests. This means that when we set up our test data RSpec is running in one process and needs to actually write to the database, then our app code reads from the database from another process. Wheras with Rack::Test, the tests and the target code runs in the same process. That’s why we can’t use a “transaction” strategy to reset our test environment with Capybara::Webkit. Instead we use the “truncation” strategy, which simple blows away all of the data after each test run.

Why bother with transactions?

Truncation will work just as well with Rack::Test as transcations, so why introduce the complexity of two different configurations? The Database Cleaner README explains: “For the SQL libraries the fastest option will be to use :transaction as transactions are simply rolled back.” Sarah Mei elaborated on this by reminding me that the commit to the database is what takes the most time, and the transaction is never committed, it is simply rolled back at the end of your test. Transactions are pretty speedy, so we want to only use the truncation method when absolutely necessary.

Just Show me the Code

Here’s the configuration that was documented by Eric Saxby from Wanelo, which worked for me as well:

config.use_transactional_fixtures = true

config.before(:each, js: true) do
  self.use_transactional_fixtures = false
  DatabaseCleaner.strategy = :truncation

config.after(:each, js: true) do
  self.use_transactional_fixtures = true

How does this work exactly?

We are set up to use transactions by default, which is built into rspec-rails and does not rely on DatabaseCleaner. Then, for our JS tests, we tell RSpec not to use transactions and instead instruct DatabaseCleaner to set up before each test runs with DatabaseCleaner.start and then clean up after with DatabaseCleaner.clean.

I have no idea why ActiveRecord::Base.establish_connection is needed, but if we don’t do that, then rake spec hangs after my first JS test with this ominous warning:

WARNING: there is already a transaction in progress

Perhaps someone reading this can explain this detail, but happy to have a configuration that works and hope this helps other folks who want fast tests that run reliably.