I still think I prefer Zefram's approach, remove optional_features entirely.
Adam K On Mon, Nov 2, 2009 at 10:27 PM, David Golden <xda...@gmail.com> wrote: > On Mon, Nov 2, 2009 at 4:25 AM, Zbigniew Lukasiak <zzb...@gmail.com> wrote: >> I don't know the history of this, but - risking that I'll add >> something already covered - I would propose that if optional features >> are to be formalized - then they should be allowed to appear in >> 'require_*'. Otherwise optional features lead to much chaos with >> packages relying on them but not being able to specify that. > > Ricardo and I iterated through the implications and semantics last night on > IRC. > > In short, optional_features serve two purposes: (a) they define a set > of requirements such that if the "requires" portion is satisfied, then > a feature is considered "available" (whatever that means); (b) they > define a set of requirements that are merged to the main requirements > if an install tool chooses to do so (by prompting the user or other > means). > > What I think you're asking about is beyond the scope of this META spec > round. You'd like a way to have Distribution X to say that it wants > distribution Y + distribution Y's optional feature Z. As others have > said, a better way for features to be packaged is via a separate > distribution. For example, Foo-Bar-1.23 has optional feature "wibble" > which requires Foo::Bar::Wibble, which comes in the > Foo-Bar-Wibble-1.23 distribution. In Foo-Bar-Wibble, all sorts of > other modules could be required. > > Then, if your module Really::Awesome needs the "wibble" feature, you > just say it requires Foo::Bar and Foo::Bar::Wibble and it all just > works. > > On reflection, a note about that should probably be added to the > omnibus prereq branch that describes optional_features. > > http://github.com/dagolden/cpan-meta-spec/tree/prereq-omnibus > > -- David >
