Dr Nic

I know HTML. Its on my CV. Expert level. I also know CSS and a whole bunch of JavaScript. I can even do TDD with JavaScript.

And on the iPhone there is this nifty object called UIWebView. Otherwise known as WebKit. Otherwise known as an embedded browser in your iPhone app.

And if you want a sexily awesome looking UI view, like the Today view of the Surf Report app (see right or free on AppStore) that was released on the AppStore recently, then the WebKit is just the best thing since the electric bread slicer for speed of development.

Holy grail of iPhone development?

Well, that’s what we thought. When I chatted with Dan Grigsby last week I mentioned there were good and bad things about the WebKit within an iPhone app.

This article is about good and bad things. The pros and cons. How we managed the integration of the two code-bases. And the answer to the big question: Would we do it again?

Its probably wonderfully useful stuff to know.

Negatives

So first the downsides. It would be remiss of me to fill you full of unbounded promises of easy non-Objective-C victories in your iPhone dev, and not tone them down with a full bucket of wet reality.

Its slow. When the WebKit is first loaded into memory, and we try to do this behind the scenes as soon as we get control of our app from the OS, it can take a good few seconds for your WebKit object to be available. You get notified of its readiness for action via the delegate:

- (void)webViewDidFinishLoad:(UIWebView *)webView

Its slow. You’re running an interpreter (JavaScript runtime) on top of a device with a small CPU and small memory. Go figure.

This reminds me of the fabulous “two minor drawbacks” scene from the 80s British sci-fi comedy Red Dwarf:

CAT: Why don't we drop the defensive shields?
KRYTEN: A superlative suggestion, sir, with just two minor flaws.
  One, we don't have any defensive shields, and
  two, we don't have any defensive shields. Now I realise that,
  technically speaking, that's only one flaw but I thought
  it was such a big one it was worth mentioning twice.

The JavaScript bridge does not appear to block the main thread. This is a good/bad thing. You can invoke JavaScript code within the WebKit via your native Objective-C code.

[webview stringByEvaluatingJavaScriptFromString:@"loadData({some: 'data'})"];

It seems to return control back to your Objective-C code before it has finished executing, so you may need to poll the JavaScript runtime with `isFinished();` calls if you need to know when its complete.

Annoyingly, the Apple documentation for this method suggests otherwise:

JavaScript execution time is limited to 5 seconds for each top-level entry point. If your script executes for more than 5 seconds, Safari stops executing the script. This is likely to occur at a random place in your code, so unintended consequences may result. This limit is imposed because JavaScript execution may cause the main thread to block, so when scripts are running, the user is not able to interact with the webpage.

Quirky.

The JavaScript bridge is one directional. From Objective-C/UIKit you can invoke JavaScript upon the WebKit and henceforth do wonderful things (as per example above).

From JavaScript you have no native nor nice way to invoke methods on Objective-C objects, like you can in the Cocoa implementation of embedded WebKits. What you can do is use custom HTTP protocols, such as surfreport://, to give the OS a way for the webkit to communicate with your app.

We used this for ‘Related Photos’ buttons on some news and athlete’s pages. When you click a button in the webkit, it attempts to redirect to a surfreport:// url. The Objective-C code (your `UIApplicationDelegate`) is notified of this, hides the webkit, and displays the related photos using native UI elements. But we could do anything. Your `UIApplicationDelegate` needs to implement the follow method to receive these requests:

- (BOOL)application:(UIApplication *)application handleOpenURL:(NSURL *)url

Alternately, you could use JavaScript to invoke this url, possible, via an Ajax call. Haven’t tried it. Might work. Either way, its a dirty hack, and a very annoying situation given that in Cocoa development there is a beautiful two-way bridge. I want that.

Multiple languages in one project. Whilst we mainly just wanted to take a static HTML file, and dynamically update various elements with application data (e.g. the surf and weather conditions in the example above), we needed to do that via JavaScript.

We came up with a consistent API loadData(json) for all pages, as discussed below; yet when something displayed incorrectly we now had an extra possible point of failure: JavaScript, Objective-C, and the original web service data source (e.g. the Surfline data for live updates).

To help isolate issues, we used JavaScript unit testing and a layer of “fixtures” or samples of JSON that might be sent from Objective-C to JavaScript. The tests passed against the fixtures; so if a QA error appeared in the app, we first checked the JSON being sent against the fixtures schema to isolate whether it was a JavaScript error or a data format error. It happened sufficiently often that its worth raising here.

WebKit for rapid prototyping

Nonetheless, the WebKit exists and it is awesome at rendering HTML and CSS, with access to the powerful CSS3 transformations and webkit-specific bonus features.

It is highly likely that your designer can make something beautiful looking in Photoshop and cut it up into HTML + CSS. Comparatively, its highly unlikely they can cut it up into native Objective-C code.

So in the initial phases of application development/prototyping, the WebKit is a sweet option to give the designer on your team direct, immediate access to building the app. If performance is an issue, you later replace it with native UI elements.

Code layout and samples

In our Xcode project, the structure we use for including HTML and associated assets into our apps is:

Classes/           - normal Objective-C .m/.h files
Html/src           - HTML, CSS, and JavaScript files including libraries like jQuery
Html/test          - JavaScript tests
Html/test/fixtures - expected JSON formats to be sent from Objective-C to loadData methods

These aren’t Xcode Groups, rather normal filesystem folders. In the Xcode project, the Html/test files are not included nor bundled with the app. The files in Html/src are included in the “Copy Bundle Resources” of the Target (see image) and also linked into the project via a sub-group of “Resources”, so they can be easily accessed within the project.

But when I did the HTML/JavaScript development I worked from the command-line and TextMate, and lived entirely inside the Html/ folder.

For each static HTML page, such as the ‘Today’ view above, there will probably be the following files:

Html/src/today.html
Html/src/today.css
Html/src/today.js
Html/src/jquery.min.js
Html/test/unit/today_test.html

The WebKit will display today.html, which uses normal <link>/<script> to pull in the .css and .js assets. You can quickly see what the page will look like, even before integration into your app, by loading it into the iPhone Simulator (launch the simulator and drag the file in), or even Safari 3.

We settled on a standard API for the primary call from Objective-C into JavaScript: loadData(json). Bare-bones, this method starts off looking like:

// This function is called from Objective-C-land to apply
// the surf information to the HTML template
// +data+ is a hash of key -> value
function loadData(data) {
    $('.data').html('N/A');
    $('img.image_data').removeAttr('src');

    // Assume that any data key can be copied into any HTML element with the same ID
    for (var key in data) {
        var value = data[key];
        var value_exists = fieldExists(value);
        var data_value = value_exists ? value : 'N/A';
        var img_src_value = value_exists ? value : 'unknown';
        $('#' + key).
            filter('.data').html(data_value).end().
            filter('.image_data').attr('src', key + '_' + img_src_value + '.png');
    };
}

What it does is take a data hash like { surf_height_amount: '2-3', surf_height_unit: 'ft' } and HTML like:

<div id="surf_size">
  <span class="data" id="surf_size_amount"></span>
  <span class="data" id="surf_size_unit"></span>
</div>

And the loadData method updates it to:

<div id="surf_size">
  <span class="data" id="surf_size_amount">2-3</span>
  <span class="data" id="surf_size_unit">ft</span>
</div>

So we can build arbitrarily complex HTML templates. By allocating meaningful element IDs (surf_height_amount) and tagging the template elements with class="data" we have a simple mechanism for Objective-C to update the HTML. As the template gets more complex, then so too does the loadData method.

Would we do it again?

Yes.

The WebKit isn’t the holy grail for non-Objective-C developers, but if your grand-poobar level skills are in JavaScript and HTML, and your Objective-C/iPhone skills are still catching up, then its a wonderful prototyping platform. Especially for static, complicated displays of data. Especially if that data includes HTML content from an external feed which needs to be rendered.

For Oakley’s Surf Report app, Anthony is toying with replacing some of the WebKit usage with native UI elements (normal UITableView with custom UITableViewCells) so that we can get back those precious seconds and give them to the user as a Christmas present.

The New Gem Generator (newgem) was exciting, moderately revolutionary, and definitely helpful two years ago when I created it. Of late it seems to attract a chunk of criticism:

• making a new gem, but newgem seems broken… hoe hoe
• NewGem has the “hoe” virus. Much prefer Mr. Bones.
• the newgem site is begging for someone to put the word fuck on its front page
• Almost two days fighting with newgem, but today i won! The secret is hoe 1.7.0 and rubygem 1.2.0.
• unfortunately for the one project I went with newgem, will give Mr. Bones a try on the next gem I throw out there.
• sow or newgem, neither, Mr. Bones
• egads, the website newgem sets up for you looks awful
• newgem seems to be failed to generate package if AUTHOR is an array of authors.

On the positive side of the spectrum is the following list:

• newgem? really? you rock drnic!

Comparatively, the two lists are awfully different in length. And not in a good way. No, not at all.

I know about these criticisms and platitude(s) because they appear publicly on newgem’s website within a live Twitter search of ‘newgem’. So that the twitter messages on the project’s own homepage are more positive, it was time for a new direction. A new beginning. It was time for a change.

So I fixed it. All of it. As of release 1.0.3 it is perfect [1].

To help you realise how wonderful newgem now is, I shall use the time-proven medium for proving awesomeness: a list. The bullet points are for free.

• newgem now finally gone 1.0.0. It made it all the way to 0.29.0, but I think 1.0.0 was needed to transfer the message of a new beginning. This is the Obama of Gem Generators.
• Generated gems are 50% smaller. No more config/hoe.rb. No more website folder (by default). No more tasks folder. No more license file. No more version.rb file.
• Very little config required before releasing your gem. Just a few fields in the Rakefile.
• You can use rspec or test/unit for unit testing (option: -T rspec)
• You can use cucumber for functional testing (option: -i cucumber or run script/generate install_cucumber)
• Generated gems are future-proofed. They will automatically benefit from future newgem releases.
• Executable apps within gems now have a lib/appname/cli.rb file for the code base, and a lightweight bin/appname (option: -b appname or run script/generate executable appname)
• Your README file can be called README.rdoc so it appears nicely formatted on github. No more hoe warning messages about “README.txt is missing” (see feature)
• GitHub RubyGem support. rake gemspec generates a clean my_project.gemspec file that will work with GitHub
• RubyForge support. As before, rake release VERSION=X.Y.Z releases your project to RubyForge (see ‘preparing for releases to rubyforge’ help page)
• newgem’s website is a different colour. Its a nice peppermint green colour. The default website template is now this theme too.
• You can raise bugs or suggest improvements via Lighthouse tracker

Installation

sudo gem install newgem

Usage

To create your RubyGem scaffold:

newgem mygem
newgem mygem -b myapp              # create a CLI executable
newgem mygem -T rspec -i cucumber  # use rspec and cucumber for gem tests
newgem mygem -w                    # create a simple website
newgem -h                          # get help

Now your code goes in lib folder, and your tests go in test, spec, and/or features as appropriate.

There are a bunch of rails-esque generators (like model or migration) that you can use to help your development:

script/generate executable myapp                       # create your own command-line interface (CLI)
script/generate extconf mylib                          # starting point for C-extensions, plus TDD framework
script/generate component_generator mygenerator scope  # create your own generators for Rails, Merb, RubyGems
script/generate application_generator myapp            # create a CLI that is a generator for something
script/generate -h                                     # get help

Bugs and suggestions

You can raise bugs or suggest improvements via Lighthouse tracker

Thanks goes to… Cucumber

Aside from several days of my time refactoring it, reducing it, and doubling the amount of awesomeness within it, all its wondefulness is thanks to Cucumber.

Cucumber is the successor to Rspec Story Runner. I never found time to play with Story Runner, but Cucumber is blowing my mind with awesomeness. My attention-span is short so I may be forgetting something but I think Cucumber could be the most important piece of software released in 2008 for Ruby-based developers.

Cucumber gave me a framework to specify newgem’s expected behaviour; its features. First I wrote feature descriptions for known, expected behaviour. Then I refactored the crap out of newgem until it was in tip-top shape.

There are over 90 feature steps defining newgem’s current features. To run them:

gem unpack newgem
cd newgem-*
sudo gem install cucumber
cucumber features

And watch the awesomeness of Cucumber unfold before your eyes. What you are seeing isn’t just newgem’s generators being executed, but also the generated code is being executed, rake tasks executed, and generated test files tested.

I can now setup continuous integration for newgem. I have a framework to know that newgem, or any other RubyGem, is doing what it should do.

UPDATE: I want to thank David Chelimsky for our time hanging out in Brazil during RailsSummit. I saw him using Cucumber, and talking about it on stage and then help helped me whilst I integrated it into newgem as a generator and then using it internally itself. For a day and a half we hung out in the hotel foyer. He’s so wonderful.

Cucumber makes me so happy.

Summary

Use newgem. Write gems.

Use mrbones. Write gems.

Use sow. Write gems.

Use gemhub. Write gems.

And write cucumber feature descriptions first. Then unit tests. Then code. Then release. Then profit.

[1] All claims of perfection are for the express purpose of making you try the product enough to use it, share it with your friends, and wrap it up and give it as a gift to family on Xmas day. Gift boxes are available upon request.

I’ve used sake intermittently in my workflow. It competes against me writing helper/admin scripts in my ~/ruby/bin folder. Normally, executable Ruby scripts have won. But I think I have a new solution that could make sake a permanent winner for me.

Ruby scripts are easy to create and execute. You just open new file, change the TextMate grammar to ‘Ruby’, type ‘rb’ and press TAB and you’re off and running (the ‘rb’ snippet generates #!/usr/bin/env ruby or a variation of that). You then make the file executable and BAM! you can run the script from any folder in your environment.

Sake tasks are more annoying to write. After creating a new file, you need to create the namespace and task wrappers for your functionality, such as:

namespace 'foo' do
  namespace 'bar' do
    desc "This task ..."
    task :baz do

    end
  end
end

Your task isn’t instantly executable either. After each change, you need to uninstall the task (sake -u foo:bar:baz) and then reinstall the sake file (sake -i foo/bar/baz.sake) and then run it (sake foo:bar:baz). Perhaps there’s a way to inline edit a sake task, but I can’t see it from the help options.

But once you’ve got your script installed in sake, you get all the wonders that sake provides: a named list (with summary) of tasks (sake -T) and the ability to run those tasks anywhere. Ok, that’s really only one advantage over standard Ruby scripts. But I like it. Oh, namespacing. The baz task exists in a namespace foo:bar. That’s nice too.

So to make me happy, I need a solution to the dubious “create-install-execute” process above. I also want the raw source for all my sake tasks in one place so I can fix/add/change them, reinstall them and move on with my life. I want simple.

So I’ve forked Chris Wanstrath’s empty sake-tasks repo (mine) and added some infrastructure for managing sake tasks. Of course the repo itself is the repository for my sake tasks (which includes a lot from Luke Melia), but most importantly it has a single rake task to reinstall all the tasks without any manual fuss.

The rest of this article assumes you want to have your own repository for your own sake tasks hosted on github. This paragraph is probably unnecessary, but I don’t want to be accused of not being mildly thorough.

Fork the sake-tasks repo

For thoroughness and a chance to demonstrate some gold-medal git-fu, I’ll show two ways: fork my repo and forking the original repo from Chris and pulling my stuff into yours. It’s git, it’s distributed, you can do anything.

If you want to fork my repo and skip a nifty git lesson, go to my sake-tasks repo and click “fork”. Then follow the clone instructions as you normally do when you are blatantly, systematically duplicating someone else’s hard work, using a command that will look something like:

git clone git@github.com:your-github-username/sake-tasks.git

Now, lazy man, you can skip to the next step.

If you want to flex your git-fu, then go and fork Chris’ repo instead. Again, follow the clone instructions.

Now take a moment to reflect on just how empty your repository is. A fine moment in open-source where you’ve essentially cloned an empty repository. Hardly worth the effort, but since Chris is a creator of github then if he creates an empty repository then who am I to disagree. Empty it shall start.

Now let’s pull in the code and tasks from my repo. My repo could be any git repo anywhere on the tubes.

One way you could pull my code into your local repository is to add my repo as a remote and then pull in the goodness:

git remote add drnic git://github.com/drnic/sake-tasks.git
git pull drnic master

This is useful if you ever plan on re-pulling from a target repo again in the future.

If you just want to pull from someone’s repo one time only, then you can merge these two lines together:

git pull git://github.com/drnic/sake-tasks.git master

If you get occasional pull requests for your projects, then the latter option is handy to know.

Your local repo is now different to your remote repo (your fork on github) so push it back to your remote:

git push origin master

Installing the sake tasks

I originally created my sake-tasks fork so I could store a git:manpages:install task. I’ve just upgraded to git 1.6 (note to self: I want an ‘upgrade to latest git version via src’ task; UPDATE the repository now includes a git:src:install task to do this) and found some instructions for installing the pre-built manpages. Then I got over excited and refactored all of Luke Melia’s git+mysql+ssh tasks in to my repo so it looked like I’d done a lot of work.

To install all the tasks, first install sake:

sudo gem install sake

Then run the install task (check below for the list of tasks to be installed):

WARNING: This will uninstall any tasks you already have by the same name.

rake install

Now, check that your sake tasks are installed:

sake -T

Gives you:

sake git:analyze:commits:flog_frequent   # Flog the most commonly revised files in the git history
sake git:close                           # Delete the current branch and switch back to master
sake git:manpages:install                # Install man pages for current git version
sake git:open                            # Create a new branch off master
sake git:pull                            # Pull new commits from the repository
sake git:push                            # Push all changes to the repository
sake git:status                          # Show the current status of the checkout
sake git:topic                           # Create a new topic branch
sake git:update                          # Pull new commits from the repository
sake mysql:dump                          # Dump the database to FILE (depends on mysql:params)
sake mysql:load                          # Load the database from FILE (depends on mysql:params)
sake ssh:install_public_key              # Install your public key on a remote server.

Sexy.

Adding new recipes/tasks

The installer rake task rake install works by assuming that each .sake file contains one sake task. This allows the rake task to uninstall the tasks from sake first, and then re-install it (sake barfs if you attempt to reinstall an existing task). Without the one-task-per-file rule, the solution would be to load all the sake tasks as rake tasks into memory. But I like one-task-per-file; it seems clean.

So, to create a task foo:bar:baz, you’ll need to add a folder foo/bar and create a file baz.sake inside it. Within that file you would then specify your task using namespace and task method calls:

namespace 'foo' do
  namespace 'bar' do
    desc "This task ..."
    task :baz do

    end
  end
end

To install new tasks or reinstall modified tasks, just run the rake task (rake install or rake).

TextMate users

The latest Ruby.tmbundle on github includes a task command that generates the above namespace/task snippet based on the path + file name. That is, inside the foo/bar/baz.sake file, make sure your grammar is ‘Ruby’ or ‘Ruby on Rails’ and then type “task” and press TAB. The above snippet will be generated ready for you to specify your task.

Summary

So now I have a single place for all my original sake source and a simple rake task to re-install the tasks if I add or modify them. And because its all in one git repo, if other people fork it and add their own tasks then I can steal them.

Everything to love about Ruby: the concise, powerful language; the sexy testing frameworks; and finally, the people.

Everything to love about Objective-C: hmmm; well…; and finally, its the only high-level language you can use to write iPhone apps.

On iPhone 2.0, to arrive on the 11th of July, you cannot run RubyCocoa. But you can run it on your Mac, so let’s use it to unit test your Objective-C classes. This tutorial shows you how to get started using a new project rbiphonetest [GitHub | Lighthouse | Google Group]

If you followed some of my recent tweets, this project was previously called “iphoneruby”. And alas, the screencast also calls it “iphoneruby” but that was a crap name. People thought it was a way to run Ruby on the iphone. I can’t do that yet. So, a far better name is ‘rbiphonetest’. [track on summize]

Even if you’ve never touched Objective-C, Cocoa, the iPhone SDK, nor RubyCocoa I recommend watching the video anyway. It should give you hope that if you make the transition to iPhone development you don’t have to go alone without Ruby: your trusty swiss army knife of language/libraries/tools.

The screencast is also available in high-def video (55Mb QuickTime)

Unit Testing iPhone apps using Ruby from Dr Nic on Vimeo.

Installation and Usage

To summarize the video, but change ‘iphoneruby’ to ‘rbiphonetest’, you install the framework via RubyGems:

sudo gem install rbiphonetest

Then change to your project’s folder and install the test framework:

rbiphonetest .

Finally, for each generic, non-UIKit-framework-using class you want to test:

script/generate model WidgetModel

Then write your tests in test/test_widget_model.rb

Supported Cocoa & iPhone frameworks

The mysterious, magical premise upon which rbiphonetest depends is possibly erroneous: that your Objective-C class can be compiled and tested against your OS X/Intel frameworks, and if your tests pass you assume you can then compile and include your class with the the iPhone/ARM frameworks.

I’m willing to go with this assumption until its proven dangerously flawed by some angry 20-year veteran of NextStep/Cocoa/iPhone. But really, how different could NSString be on the iPhone versus your Mac?

Fortunately there is one way to check for significant differences between your available Mac-based frameworks, such as Cocoa, and the iPhone-based frameworks, such as UIKit. We need to compare the framework names, header files and method signatures.

So for example, you cannot currently unit test any class that depends on/includes the UIKit framework. Why? It doesn’t exist on your Mac, so the Mac/Intel compiler cannot link it in. We’re compiling and running our tests with RubyCocoa, which itself is built against the Mac/Intel frameworks, not the iPhone frameworks. Hell, Laurent doesn’t even own an iPhone [Laurent is the Apple-employee maintainer of RubyCocoa and the newer MacRuby]

Similarly, its no use including/linking the Cocoa framework into your Objective-C class. Why? It doesn’t exist on the iPhone. It has its own UI frameworks, collectively called ‘UIKit’.

So for the moment we cannot test UI-related, iPhone-API-specific code. But we can test generic Objective-C. That’s better than a kick in the teeth. Surely. I mean, in the teeth… that’d friggin’ hurt.

“Fair enough Dr Nic, so which frameworks can my code use and yet still unit test it with your oh-so-special test library thingy?” Keep your pants on, I’m getting there. [ref]

To the best of my ability, I’ve compared the two sets of frameworks and listed the available Frameworks that are available on both the iPhone and your Mac. There are about a dozen. The most important is called ‘Foundation’. It holds gold nuggets like ‘NSString’.

The list of platform differences is on the wiki as a reference.

Note, this list doesn’t guarantee that any two framework classes - the iPhone and matching Mac framework - will behave the same. This list is compiled by finding the set of Frameworks with the same name on both platforms, e.g. Foundation.

Then it compares the set of public header files (Foundation.framework/Headers/*.h files) This comparison is by method signature. It pulls all lines from each header that start with + or - (+ is a class method and - is an instance method in Objective-C) and compares the two lists. If there is a single difference in the method signatures of the header files in the two platforms it is marked on the wiki page. You’ll need to look at the two header files yourself to see the differences. Some header files are ugly. C-based anything starts ugly and goes down from there, I think.

Python testing of iPhone Objective-C?

In the Python world there is PyObjC, a bridge-based twin to RubyCocoa. If you are a Python developer you could easily port this project to use PyObjC I would think. Ping me if you are attempting this and need any help.

Summary

I think this project can give Ruby developers a happy place to work from as they write their Objective-C/iPhone code. You still need to wire up your UI views and controller classes manually, but if you push all the “oooh that code really needs some tests” classes away from the UI-dependent frameworks then you can hook it up to rbiphonetest and write your tests in Ruby.

Currently the generator creates test/unit test stubs. I personally then add the Shoulda gem into my test_helper.rb for my apps. If an rspec and/or test/spec developer can help with adding support to the generators I’m certain the large rspec user-base would be happy campers.

Similarly, someone might like to investigate using MacRuby to run the tests instead of RubyCocoa. Fast tests vs slow tests. You choose.

What the?

Sometimes I re-read what I’ve written and notice things that don’t seem to make sense, but are in my vocabulary nonetheless. Yep, the things you learn living in Australia.

“Keep your pants on” - this seems to imply that until I mentioned otherwise you were about to take your pants off. Hardly relevant at any stage during this article, we’d both agree. Most code-based blog articles are “pants on”. This phrase means “don’t get upset”. You can try to figure out how you go from “don’t get upset” to “keep your pants on”. I have no idea.

Two years ago Dave Thomas did a keynote at the first RailsConf in 2006 and outlined a few things missing in Rails. One was the seeming unnecessary un-DRYness of duplicating associations and validations in Active Record models since the same information is in the DB schema. Another was support for Composite Primary Keys on Active Record models.

A few weeks later I created my first RubyGems as an attempt to solve these problems: Dr Nic’s Magic Models and Composite Primary Keys. The former was funny and an entertaining way to use const_missing? and Class.new. The latter was not funny. It started at the heights of “mildly entertaining,” dropped down from there and then over the subsequent months it never again rose above the humour-scale heights of “please shoot me in the foot.”

There were entire months that I hated ever having created the Composite Primary Keys project

Why? It sits precariously atop of ActiveRecord::Base, and a half dozen other ActiveRecord classes. It overrides methods, recreates entire blocks of SQL, and must work on all the different database adapters. That is, it is very sensitive to many changes in edge Rails. And refactoring ActiveRecord is a favourite pastime of the Rails Core team.

Unlike the Magic Models project which is perfectly useful for party tricks but probably partially pointless elsewhere, the CPK project just seemed to be so damned useful to some people.

Many people kept using it, several people contributed patches for adapter support and new features, and finally the Prophet arrived. The man that would lead the faithful forward to the Promised Land.

His name is Darrin “Champion” Holst.

Today, CPK was given the golden release number 1.0.0 and officially supports Rails 2.1 (but no longer supports 2.0.2 afaik).

CPK users might remember that we went 0.9.90 back in January, and its now June. This led Darrin to comment in his release email to the mailing list:

I’d like to name this release the “we’re running out of 0.9.x numbers, so it has to go to 1.0 sooner or later” release.

Maintaining CPK is like being on the team that paints the Sydney Harbour Bridge - it takes then 12 mths to paint it from one end to the other, which is just in time to restart painting the bloody thing again.

So in every way that I (tried to) abandon the project, I am proud of Darrin for looking after it and all its users.