Never mind, I've figured it out. The build_class function looks at the '__class__' attribute of the first base class if there's no explicit __metaclass__ attribute. By calling type directly, the __class__ attribute as returned by MyMeta is, in fact, type.
Should have just looked at the source to begin with. On 2/21/08, Jeff McNeil <[EMAIL PROTECTED]> wrote: > > Hi list, > > Hopefully a quick metaclass question. In the following example, MyMeta is > a metaclass that does not inherit directly from type: > > #!/usr/bin/python > > class MyMeta(object): > def __new__(cls, name, bases, vars): > print "MyMeta.__new__ called for %s" % name > return type(name, bases, vars) > > class MetaWrapper(object): > __metaclass__ = MyMeta > > class M(MetaWrapper): > pass > > [EMAIL PROTECTED] ~]$ python t.py > MyMeta.__new__ called for MetaWrapper > [EMAIL PROTECTED] ~]$ > > When I run that script, it's apparent that although M inherits from > MetaWrapper, it does not use MyMeta as it's metaclass. However, if I change > MyMeta to be a subclass of builtin type, it works as I would expect: > > [EMAIL PROTECTED] ~]$ cat t.py > #!/usr/bin/python > > class MyMeta(type): > def __new__(cls, name, bases, vars): > print "MyMeta.__new__ called for %s" % name > return super(MyMeta, cls).__new__(cls, name, bases, vars) > > class MetaWrapper(object): > __metaclass__ = MyMeta > > class M(MetaWrapper): > pass > > [EMAIL PROTECTED] ~]$ python t.py > MyMeta.__new__ called for MetaWrapper > MyMeta.__new__ called for M > [EMAIL PROTECTED] ~]$ > > How exactly does Python choose which MC it will use when building a > class? It doesn't seem to me that the parent class of MyMeta should matter > in this case? > > Thanks! > > Jeff > >
-- http://mail.python.org/mailman/listinfo/python-list
