h1. Extending Buildr

h2. Organizing Tasks

A couple of things we learned while working on Buildr.  Being able to write your own Rake tasks is a very powerful feature.  But if you find yourself doing the same thing over and over, you might also want to consider functions. They give you a lot more power and easy abstractions.

For example, we use OpenJPA in several projects.  It's a very short task, but each time I have to go back to the OpenJPA documentation to figure out how to set the Ant MappingTool task, tell Ant how to define it.  After the second time, you're recognizing a pattern and it's just easier to write a function that does all that for you.

Compare this:

{{{!ruby
file('derby.sql') do
  REQUIRES = [
    'org.apache.openjpa:openjpa-all:jar:0.9.7-incubating',
    'commons-collections:commons-collections:jar:3.1',
    . . . 
    'net.sourceforge.serp:serp:jar:1.11.0' ]
  ant('openjpa') do |ant|
    ant.taskdef :name=>'mapping',
      :classname=>'org.apache.openjpa.jdbc.ant.MappingToolTask',
      :classpath=>REQUIRES.join(File::PATH_SEPARATOR)
    ant.mapping :schemaAction=>'build', :sqlFile=>task.name,
      :ignoreErrors=>true do
        ant.config :propertiesFile=>_('src/main/sql/derby.xml')
        ant.classpath :path=>projects('store', 'utils' ).
          flatten.map(&:to_s).join(File::PATH_SEPARATOR)
    end
  end
end
}}}

To this:

{{{!ruby
file('derby.sql') do
  mapping_tool :action=>'build', :sql=>task.name,
    :properties=>_('src/main/sql/derby.xml'),
    :classpath=>projects('store', 'utils')
end
}}}

I prefer the second.  It's easier to look at the Buildfile and understand what it does.  It's easier to maintain when you only have to look at the important information.

But just using functions is not always enough.  You end up with a Buildfile containing a lot of code that clearly doesn't belong there.  For starters, I recommend putting it in the @tasks@ directory.  Write it into a file with a @.rake@ extension and place that in the @tasks@ directory next to the Buildfile.  Buildr will automatically pick it up and load it for you.

If you want to share these pre-canned definitions between projects, you have a few more options.  You can share the @tasks@ directory using SVN externals. Another mechanism with better version control is to package all these tasks, functions and modules into a "Gem":http://rubygems.org/ and require it from your Buildfile.  You can run your own internal Gem server for that.

For individual task files, you can also use "Sake":http://errtheblog.com/post/6069 for system-wide Rake tasks deployment.


h2.  Creating Extensions

The basic mechanism for extending projects in Buildr are Ruby modules.  In fact, base features like compiling and testing are all developed in the form of modules, and then added to the core Project class.
  
A module defines instance methods that are then mixed into the project and become instance methods of the project.  There are two general ways for extending projects.  You can extend all projects by including the module in Project:

{{{!ruby
class Project
  include MyExtension
end
}}}

You can also extend a given project instance and only that instance by extending it with the module:

{{{!ruby
define 'foo' do
  extend MyExtension
end
}}}

Some extensions require tighter integration with the project, specifically for setting up tasks and properties, or for configuring tasks based on the project definition.  You can do that by adding callbacks to the process.

The easiest way to add callbacks is by incorporating the Extension module in your own extension, and using the various class methods to define callback behavior.

|_. Method        |_. Usage |
| @first_time@    | This block will be called once for any particular extension.  You can use this to setup top-level and local tasks. |
| @before_define@ | This block is called once for the project with the project instance, right before running the project definition.  You can use this to add tasks and set properties that will be used in the project definition. |
| @after_define@  | This block is called once for the project with the project instance, right after running the project definition.  You can use this to do any post-processing that depends on the project definition. |

This example illustrates how to write a simple extension:

{{{!ruby
module LinesOfCode
  include Extension

  first_time do
    # Define task not specific to any projet.
    desc 'Count lines of code in current project'
    Project.local_task('loc')
  end

  before_define do |project|
    # Define the loc task for this particular project.
    define_task 'loc' do |task|
      lines = task.prerequisites.map { |path| Dir["#{path}/**/*"] }.flatten.uniq.
        inject(0) { |total, file| total + File.readlines(file).count }
      puts "Project #{project.name} has #{lines} lines of code"
    end
  end

  after_define do |project|
    # Now that we know all the source directories, add them.
    task('loc'=>compile.sources + compile.test.sources)
  end

  # To use this method in your project:
  #   loc path_1, path_2
  def loc(*paths)
    task('loc'=>paths)
  end

end

class Buildr::Project
  include LinesOfCode
end
}}}


h2.  Using Alternative Layouts

Buildr follows a common convention for project layouts: Java source files appear in @src/main/java@ and compile to @target/classes@, resources are copied over from @src/main/resources@ and so forth.  Not all projects follow this convention, so it's now possible to specify an alternative project layout.

The default layout is available in @Layout.default@, and all projects inherit it.  You can set @Layout.default@ to your own layout, or define a project with a given layout (recommended) by setting the @:layout@ property.  Projects inherit the layout from their parent projects.  For example:

{{{!ruby
define 'foo', :layout=>my_layout do
  ...
end
}}}

A layout is an object that implements the @expand@ method.  The easiest way to define a custom layout is to create a new @Layout@ object and specify mapping between names used by Buildr and actual paths within the project.  For example:

{{{!ruby
my_layout = Layout.new
my_layout[:source, :main, :java] = 'java'
my_layout[:source, :main, :resources] = 'resources'
}}}

Partial expansion also works, so you can specify the above layout using:

{{{!ruby
my_layout = Layout.new
my_layout[:source, :main] = ''
}}}

If you need anything more complex, you can always subclass @Layout@ and add special handling in the @expand@ method, you'll find one such example in the API documentation.

The built-in tasks expand lists of symbols into relative paths, using the following convention:

|_. Path                          |_. Expands to |
| @:source, :main, <lang/usage>@  |  Directory containing source files for a given language or usage, for example, @:java@, @:resources@, @:webapp@. |
| @:source, :test, <lang/usage>@  | Directory containing test files for a given language or usage, for example, @:java@, @:resources@. |
| @:target, :generated@           | Target directory for generated code (typically source code). |
| @:target, :main, <lang/usage>@  | Target directory for compiled code, for example, @:classes@, @:resources@. |
| @:target, :test, <lang/usage>@  | Target directory for compile test cases, for example, @:classes@, @:resources@. |
| @:reports, <framework/usage>@   | Target directory for generated reports, for example, @:junit@, @:coverage@. |

All tasks are encouraged to use the same convention, and whenever possible, we recommend using the project's @path_to@ method to expand a list of symbols into a path, or use the appropriate path when available.  For example:

{{{!ruby
define 'bad' do
  # This may not be the real target.
  puts 'Compiling to ' + path_to('target/classes')
  # This will break with different layouts.
  package(:jar).include 'src/main/etc/*'
end

define 'good' do
  # This is always the compiler's target.
  puts 'Compiling to ' + compile.target.to_s
  # This will work with different layouts.
  package(:jar).include path_to(:source, :main, :etc, '*')
end
}}}