On Tue, Aug 6, 2013 at 12:36 PM, BrJohan <brjo...@gmail.com> wrote: > On 06/08/2013 16:02, Chris Angelico wrote: > >>> My classhierarchy is like a multilevel tree where each non-leaf node >>> (class) >>> is given knowledge about its nearest subclasses and their 'capacities'. >>> >>> So, my idea is to let the 'upper' class recursively choose which of its >>> nearest subclasses is the 'correct' one, until approaching a 'leaf' class >>> from which the instance should be created. And, given my knowledge that a >>> solution along the lines of this idea has been designed and was working, >>> I'm >>> still hopeful ... (or I'll have to investigate all those old backup-DVDs) >> >> >> [ responding on-list - I hope it was mere oversight that had this come >> privately to me alone ] >> >> This is code smell; this recursive search for the "right" class seems >> likely to be wrong. Can you have the classes perhaps register >> themselves in some way? On what basis is a superclass to determine >> that one of its subclasses should handle this request? >> >> ChrisA >> > > > Consider a botanical classification system (somewhat analogous to my > 'problem' as it effectively is related to classification of entities): > > A Domain should know about its Kingdoms, > a Kingdom should know about its Phylums, > ... > a Genus should know about its Species. > > Of course it is possible to implement such a decision tree as a 'factory'. > However, I would rather prefer to encapsulate those decisions at the class > level where they 'belong'. > > BrJohan > --
I think it's a "has a" vs a "is a" problem. A Domain has a Kingdom, a Kingdom is not a Domain, so it shouldn't actually inherit Domain. In this case you should use composition instead of inheritance. When you say that a A Domain should know about it's Kingdons note that you're talking about a specific Domain and it's specific Kingdons, which means, a Domain instance and various Kingdom instances. JoeS -- http://mail.python.org/mailman/listinfo/python-list
