Once outside, I realised I’d left my friends in another city as I stubbornly wanted to see this one, even though it was pretty run down. It was a small Indian town, it was raining, and there was a fish stall allowing you to sponsor parts of the fish that were labelled with different prices. Some people had apparently already sponsored parts of it for as much as £200.

Realising I was carrying a bag of cement in a large woman’s handbag slung over my shoulder, I had to consolidate that with my backpack to avoid looking feminine.

I found my way to a hardware store, a chain of which I’d previously visited in my own country. It was mostly second hand junk. I was also carrying a camera on a piece of shower flex, and it had a light on the other end.

Browsing around the store, I found a similar item being demonstrated by a woman, and there was a crowd gathering around the demo, watching a screen displaying what the camera saw. I pushed to the front of the crowd and put my eye against the lens, then put my camera’s lens against hers. To try to entertain the crowd, I turned the light on and shone it into the demo lens. I thanked her, because I’d finally found a use for the junk I’d bought.

Answers on a postcard, I suppose.

I only wanted the JSON format for the create action, so I thought this would be pretty harmless. It isn’t. In Internet Explorer 7, the first declaration causes refreshes of the DesignsController#index action to return the JSON representation, whereas initial page loads show the HTML version. Swapping the first and second declarations fixed this issue.

I presume that the cause is the same as the linked article?

The royal paths

There are a few semi-official options for running YUI3 over SSL:

• Using the PHP Loader and hosting your own combo service
• Using a custom combo service like Combo Handler for Node
• Using the dependency configurator

None of the above worked for me.

Firstly, the PHP Loader is buggy and requires some code changing to make it work anyway. After a day or two of playing with that I gave up on it.

Secondly, I got distracted by some efforts to do combo services in node.js. This was silly, introducing an enormous new set of dependencies to my Rails app would have been torture.

Thirdly, the dependency configurator is great, but you need to know what you depend on. The chances are, though, you don’t. YUI’s clever loader only loads the bits it needs for the browser it’s running on. If you customise a rollup that works with Chrome, the chances are that it won’t work on Internet Explorer 7 (for example) because you won’t have included the compatibility modules.

The peasant’s path

So how did I solve this horrible problem? I wanted to know exactly which modules were being loaded by the lowest-common-denominator browser to be supported by the project, which was Internet Explorer 7. On a more modern browser, you’d have a nice developer console. Getting such a thing for IE7 is like self harm.

As it happens, I’d installed Fiddler on my IE7 VM to track down some other rogue SSL requests (it later turned out to be a Modernizr bug). Using Fiddler whilst browsing around the site I could find the modules being loaded by YUI. Happily, the compatibility modules don’t negatively impact modern browsers (apart from being an extra download).

Once I’d found the modules being loaded, I curl’d the official YUI3 combo URL into a file (without minification, as this is done in the project in a deployment step). This file could then be served on my project’s own CDN. Sorted.

I should note that the process above sounds really easy. Conversely, it was one of the most frustrating development problems I’ve come across in my career. If you’re listening, Yahoo!, please host your combo service for SSL. Please.

I should see “blah” within the cart summary
I should see the 1st design within the cart summary
I should see the cart summary

“the cart summary” here is defined in selectors.rb as, for example:

However, when adding a simple implementation of a “I should see” step that used selectors, I ran into an ambiguity. Here are the ambiguous steps:

And here’s the type of error you’ll get:

I resolved the ambiguity with some use of negative look-ahead:

This says that the step should match “I should see …” steps, so long as they don’t contain the string “within” or have quotation marks. The “(?:” designates a non-capturing group, whereas the “(?!” designates a negative look-ahead.

While I’m on the subject, I think that the antipathy towards Pickle is misplaced. Pickle is great for keeping state about models in stories. However, using only the built-in steps is quite foolish, and will definitely end up in some unreadable stories.

I spent a while trying to implement this fix for our project, to no avail.

The solution: for a reason I’m yet to fathom to do with the way / frequency that Solr indexes stuff, you must do an explicit Sunspot.commit to ensure any database changes have been stored in the index. I added this to the end of a step definition that set up some data before a search step, and hey presto!

UPDATE: commits normally occur when a controller action finishes, hence why you need to do them manually when creating records outside of controller actions. sunspot_rails inserts an after_filter to do this.

To get to this point, I added some debug lines to net/http in the get and post methods, just to dump out the bodies of the requests and responses. They were initially showing POSTs like:

But after the change, the all-important commit element is added to the end as a separate request:

After I’d discovered this little bastard, VCR worked fine, and now I’ve recorded the responses from Solr I’ll no longer need to faff with a separate search server.

UPDATE 2: adding an AfterStep hook with Sunspot.commit_if_dirty helped DRY up my step definitions.

The solution was to explicitly define the controller action’s method in the controller. So the question to the world is: what changed, ActionController or something in the Cucumber chain?

RSpec 2 doesn’t have have_tag method. Now using webrat’s have_selector.

Plugin to help with upgrade only relevant to Rails 2 project, which seems silly.

Should use ‘extend ActiveModel::Naming’ in non-ActiveRecord classes used with view helpers that do record introspection e.g. form_for.

right_aws gem clobbers ActiveSupport 3, must use mikel fork

cucumber wasn’t following redirects. used this fork of webrat.

Builder needed requiring in application_helper, as I was using it to build markup.

Machinist 2 is a big change, but probably worth it. Cucumber needs database
cleaner.

Strange problem with matcher. I’d written matcher to allow ‘blah.should save’,
which clobbered ActiveRecord’s actual save method! Very specific, but worth
checking if you’re doing anything funky like that.

Routes were a breeze.

Array#sample rather than Array#rand

spec.opts now .rspec in Rails.root or ~/

Liking the enforced declarative style for validations.

email_spec had changed, features needed upgrading

#fullpath instead of #request_uri

Write more blog posts in shorthand!

So here it is, HTML5 Canvas Pong.

Pong is effectively a Hello, World! application for game developers. It’s not particularly fun, and my version is full of amusing bugs. It’s currently single-player only, and mouse-driven. There’s also no scoring or any game over: the ball just keeps travelling off-screen.

By the time you read this, however, all of these bugs could be fixed, and it could have amazing features you never thought possible in a computer game, let alone in a browser. But I doubt it. You can follow its progress over on Github.

My intention is to write a well-tested, JSLinted 2D game library for JavaScript that I can understand. Yes, I know someone else has probably done it. Humour me.

I was inspired by Chris Lowis’ version.