Ben Finney wrote: > Howdy all, > > Okay, so Guido doesn't like Abstract Base Classes[0], and interfaces > are the way of the future[1]. But they're not here now, and I > understand ABCs better. This is a very interesting discussion - not all of it understandable to me.
Are interfaces really in our future? I found the contributions of "steffen" particularly appealing. Interfaces seem to add another level of complexity without significant benefit. Colin W. > > I want my modules to (sometimes) define an abstract base exception > class, that all other exceptions in that module inherit from. > > class FooException(Exception): > """ Base class for all FooModule exceptions """ > > class FooBadFilename(FooException): > """ Raised when a bad filename is used in a foo """ > > class FooUnknownBar(FooException, KeyError): > """ Raised when an unknown bar is used with a foo """ > > However, there should never be an exception raised of class > FooException, and in fact I want that to cause an appropriate error to > be raised from the module. > > Normally, I'd pick some key part of the functionality of the class, > and cause that to raise NotImplementedError. It's then the > responsibility of subclasses to override that. > > However, in the case of exceptions, I don't want to override *any* > functionality; everything should be provided by the base classes. It'd > be messy to have to override something in every subclass just to > ensure the abstraction of the module base exception. > > I've tried doing this in the __init__(): > > class FooException(Exception): > """ Base class for all FooModule exceptions """ > def __init__(self): > raise NotImplementedError, \ > "%s is an abstract class for exceptions" % self.__class__ > > When I use this, I discovered to my horror that the subclasses were > calling FooException.__init__ -- which I though wasn't supposed to > happen in Python! > > It's also rather semantically weird, to my eye. > > Can I do something tricky with checking base classes in the > FooException.__init__() ? > > > [0] Although he's apparently been quoted earlier as saying he did. > He's changed his mind[1] since then. > > [1] <URL:http://www.artima.com/weblogs/viewpost.jsp?thread=92662> > -- http://mail.python.org/mailman/listinfo/python-list
