implementedBy() is broken in Python 3.2

[Barry Warsaw]

While porting some existing code using zope.interface to Python 3.2, I noticed that Interface.implementedBy() appears to be broken in Python 3.2. This sample code prints True in Python 2.7 in a virtualenv with zope.interface-3.8.0, but False in a Python 3.2 virtualenv.

from __future__ import print_function
__metaclass__ = type

from zope.interface import Interface, implements

class MyInterface(Interface):
    pass

class MyClass:
    implements(MyInterface)

print(MyInterface.implementedBy(MyClass))

I think the reason is that MyClass doesn't end up with an __implements__ attribute, but I'm not sure why that is.

[Tres Seaver (tseaver)]

This is a know problem with the "class advice" mechanism used for
'implements', 'implementsOnly', and 'classProvides': they abuse the stack
frame to mutate the locals of their suite in a way which doesn't work in Py3k.
The (portable) workaround is to use the 'implementer' ('implementor_only' /
'provider') class decorators::

  from zope.interface import implementer

  @implementer(MyInterface)
  class MyClass(object):
      pass

"Portable" obviously only includes Py2 versions which support class
decorators (>= 2.6).

[Barry Warsaw (barry)]

On Jan 04, 2012, at 05:53 PM, Tres Seaver wrote:

>This is a know problem with the "class advice" mechanism used for
>'implements', 'implementsOnly', and 'classProvides': they abuse the stack
>frame to mutate the locals of their suite in a way which doesn't work in Py3k.
>The (portable) workaround is to use the 'implementer' ('implementor_only' /
>'provider') class decorators::
>
> from zope.interface import implementer
>
> @implementer(MyInterface)
> class MyClass(object):
> pass
>
>"Portable" obviously only includes Py2 versions which support class
>decorators (>= 2.6).

Cool, I didn't even know about @implementer. That syntax isn't too bad
either, and I think it might even be an improvement over the implements()
syntax (though it takes a little getting used to).

I only care about Python >= 2.6 so that's fine.

Thanks! I'm going to add this to my followup post about porting to Python 3.

[Jim Fulton (jim-zope)]

We should cause the old syntax to fail more clearly in py3k and maybe warn in Python 2.7. I'll create a separate issue.

[Barry Warsaw (barry)]

On Jan 04, 2012, at 09:17 PM, Jim Fulton wrote:

>We should cause the old syntax to fail more clearly in py3k and maybe
>warn in Python 2.7. I'll create a separate issue.

+1. As it stands, the failure is completely silent.

Please subscribe me to the new issue.

[Marius Gedminas (mgedmin)]

For the record: the new issue is bug 911995.