[Python-3000] __special__ method lookup [was Re: Confused about getattr() and special methods]

[Guido van Rossum]

I only care about getting this right when there is a reasonable chance
that a class is being used as an object. For example, at the sprint we
ran into this with the __format__ special method, when someone
discovered that format(object, "") raised a weird error rather than
returning str(object), which was due to the default __format__ method
defined on the object class. It's important that you can format
*anything*, so we fixed this right away.

OTOH for the with-statement, the object passed to it is always
specially constructed to work in this context, and passing something
random like a type object just isn't a reasonable use case. As long as
you get *some* kind of error (and you do, usually complaining about
the arg count) I'm okay.

--Guido

On 9/4/07, Nick Coghlan <ncoghlan at gmail.com> wrote:
> Georg Brandl wrote:
> > Thomas Heller schrieb:
> >
> >> IIUC, in py3k, classic classes do not exist any longer, so the __metaclass__ line
> >> has no effect anyway.  Is this behaviour intended?
> >
> > It is another incarnation of special methods being looked up on the class,
> > not the instance. This was always the behavior with new-style classes, see
> > the thread at
> >
> > http://mail.python.org/pipermail/python-3000/2007-March/006261.html
> >
> > for a previous discussion.
> >
> > I think we should tackle this issue now and make sure the decided resolution
> > is consistently applied throughout Python.
>
> This issue came up when implementing PEP 343 as well - because the with
> statement is just syntactic sugar without any dedicated opcodes,
> __enter__/__exit__ are accessed via a conventional attribute lookup
> opcode. So unlike the special methods that use a C-level slot in the
> type object, these two operations *can* be affected by instance
> attributes and __getattr__.
>
> However, Guido did say at the time that he was OK with the effect of
> instance attributes on special method lookups being formally undefined
> and implementation dependent. I wasn't too worried either way - mucking
> with special methods outside the scope of 'provide this on your class to
> support operation X' has long been a pretty dubious exercise.
>
> Cheers,
> Nick.
>
> --
> Nick Coghlan   |   ncoghlan at gmail.com   |   Brisbane, Australia
> ---------------------------------------------------------------
>              http://www.boredomandlaziness.org
> _______________________________________________
> Python-3000 mailing list
> Python-3000 at python.org
> http://mail.python.org/mailman/listinfo/python-3000
> Unsubscribe: http://mail.python.org/mailman/options/python-3000/guido%40python.org
>


-- 
--Guido van Rossum (home page: http://www.python.org/~guido/)