I have a little design conundrum. I'm not sure the best way to model this. Basically I have two model "parts" and I want to support two ways of using them. Eg. class Part1 attr_accessor :a, :b, :x end class Part2 attr_accessor :x, :y, :z end Now in the first case I want effectively: class Model1 < Part1 def subpart @subpart ||= Part2.new end end but in the second I want the equivalent of: class Model2 < Part1 < Part2 but obviously I can't do multiple inheritance. So I've been trying to figure out the best way to do it. Do I use modules? Do I use Forwardable? What's the cleanest, fastest, least troubling way to go about it? Thanks, T.
on 2007-07-25 23:11
on 2007-07-26 19:56
On 25.07.2007 23:11, Trans wrote: > end > > class Model2 < Part1 < Part2 > > but obviously I can't do multiple inheritance. > > So I've been trying to figure out the best way to do it. Do I use > modules? Do I use Forwardable? What's the cleanest, fastest, least > troubling way to go about it? It's difficult to comment without knowing the domain or your use case. IMHO the fact that you want different inheritance hierarchies indicates that there is something wrong, namely inheritance might not be the best approach here. It seems like composition would be better but since I do not know what you are doing please take this with a large grain of salt. Can you disclose more detail? Kind regards robert
on 2007-07-26 21:20
On Jul 26, 10:55 am, Robert Klemme <shortcut...@googlemail.com> wrote: > Can you disclose more detail? Sure. It's for configuration information. On the one hand, I want to be able to read package settings from a file that is strictly dedicated to packaging. So, for instance my project might have a .package file that has basically: project: foo version: 1.1 dependencies: [ facets ] formats: [zip, tgz, gem] However, I also want to support an uber-project configuration file, that would look like: project: foo version: 1.1 package: dependencies: [ facets ] formats: [zip, tgz, gem] So I have two slightly different formats that I want to support with strict classes. Eg. in the first case: pkg = Package.load(file) pkg.project #=> 'foo' pkg.formats #=> ['zip','tgz','gem'] and in the other: pkg = Project.load(file) pkg.project #=> 'foo' pkg.package.formats #=> ['zip','tgz','gem'] So, the same info, but organized in the two different manners, as I described in the original post. Note, by "strict class" I mean I'm not just keying off a YAML-loaded hash. I need a class b/c various attributes have default values, validation and formating applied, plus supporting methods. T.
on 2007-07-26 21:39
On 7/25/07, Trans <email@example.com> wrote: You just recently accepted a patch for snapshot, right? A nice one indeed I think. Does this ring the same bells to you than to me? I am thinking of manipulating Hashes instead of classes, did I miss something or might it work? Well from the design point of view that would recall Module Inclusion. Cheers Robert
on 2007-07-26 22:36
On 26.07.2007 21:18, Trans wrote: > project: foo > dependencies: [ facets ] > > pkg = Project.load(file) > pkg.project #=> 'foo' > pkg.package.formats #=> ['zip','tgz','gem'] > > So, the same info, but organized in the two different manners, as I > described in the original post. > > Note, by "strict class" I mean I'm not just keying off a YAML-loaded > hash. I need a class b/c various attributes have default values, > validation and formating applied, plus supporting methods. So as far as I can see you have these requirements: 1. reuse of config parameters and their defaults in multiple places 2. default values for config parameters 3. supporting functionality (whatever that is) Some random thoughts (as I'm pretty tired already): Personally I'd limit 3 to a bare minimum. I am sure there are multiple ways to handle default values (for example, using a template of nested Hashes and merging that with something parsed from the config file). OpenStruct or a similar concept might come in handy. Is validation of config values better done in (generic?) config classes or in application classes (I tend to believe the latter, because ensuring proper arguments is a task of the model). Hm... Probably not too useful. Maybe I have more ideas after sleeping. :-) Kind regards robert