Puppet » Puppet Forge

((( Everything below is copied from a recent email thread )))

When a module is created, a “Modulefile” is generated. It’s a Ruby DSL for describing the module and looks like this:

name 'MyUsername-MyModule'
version '0.4.2'
dependency 'SomeUser-SomeModule', '>= 0.2'
dependency 'AntherUser-AnotherModule', '0.7'

When a user builds a module, the tool does an “eval” on the above code and uses it to generate a “metadata.json” file that canonically describes the contents of a release. The resulting JSON file isn’t meant for human use. The tool compiles a bunch of other data into it during the build, ranging from checksums to the details about the Puppet types and dependencies, and clobbers it each time a module’s built.

Maintaining this JSON file by hand would be possible, but would be tedious unless we make the tool powerful enough to update it’s contents — e.g., updating the file checksums, Puppet types/params/docs, etc.

Do we really want to have two different file? A Ruby-based “Modulefile” and the the JSON-based “metadata.json”? I don’t know. I’ve got it working well enough and definitely think the JSON file must be kept because we need a format that the server can safely parse. However, the Ruby “Modulefile” DSL makes logical sense to keep because it eliminates some tedious coding and enforces a complete split what the user enters and what’s generated for them during a build.

At this point, I’d argue for keeping the current arrangement because it works and it’d take at least a day to replace it. However, if you and/or Luke are annoyed enough by this approach, we should address it soon.