Issue 2366: Fixer for new metaclass syntax is needed

* new metaclass syntax (removing __metaclass__?)
   - __metaclass__ = type at global level disappear
   - __metaclass__ = <anything else> should generate warning
   - __metaclass__ = <something> within a class should use new syntax
   - class __metaclass__ should be warned about
   - any other use of __metaclass__ should be warned about

Does the old way still work in 3.0?  If so, I don't think we should have
a fixer.

A quick test indicates that the old way doesn't work anymore.

Here is a partial implementation.  It doesn't warn about __metaclass__
at the module level and doesn't handle multiple __metaclass__
assignements in one class.  tests pending.

New patch that does more.  Collin, could you take a look at the fixer? 
I listed some stumbling blocks at the top (and at least one bug in
2to3).  The fixer seems to work fine on actual files but the unit tests
that use strings do nothing.  Thanks.

Ping

Thanks for the ping.  I just rewrote the patch from scratch and it
handles corner cases (of which there are many in the parse tree) better.
 I'll upload/checkin sometime today.

The new patch works and handles all the corner cases I could think of. 
I tried to comment the heck out of it because it does a lot of manual
walking and manipulation of the syntax tree.

This only handles __metaclass__ inside classes.  I wasn't sure what to
do with __metaclass__ at the module level (which is now a NOP, I think).

Good work on the patch! It looks pretty good. I'm attaching a patch with
a few minor changes. I don't really understand the need for
fixup_parse_tree, since I see in the Grammar that a suite is required
for every classdef.

Benjamin, the 2to3 parse tree straddles the 2.x Grammar and 3.x Grammar
(it's its own thing) which is why fixup_parse_tree is there.  From the
docstring:
  one-line classes don't get a suite in the parse tree so we add one to
normalize the tree
The 2to3 parser is very line oriented - only the try/except logic uses
blocks (I think).

There haven't been any objection so I've assigned this to myself and
I'll check it in and close it soonish.

On Thu, Aug 21, 2008 at 9:38 PM, Jack Diederich <report@bugs.python.org> wrote:
>
> Jack Diederich <jackdied@gmail.com> added the comment:
>
> Benjamin, the 2to3 parse tree straddles the 2.x Grammar and 3.x Grammar
> (it's its own thing) which is why fixup_parse_tree is there.  From the
> docstring:
>  one-line classes don't get a suite in the parse tree so we add one to
> normalize the tree
> The 2to3 parser is very line oriented - only the try/except logic uses
> blocks (I think).

Yes. Sorry that was a brain seg fault. :)

>
> There haven't been any objection so I've assigned this to myself and
> I'll check it in and close it soonish.

Add an example example.py while you're at it.
>
> ----------
> assignee: collinwinter -> jackdied
>
> _______________________________________
> Python tracker <report@bugs.python.org>
> <http://bugs.python.org/issue2366>
> _______________________________________
>

Added in r66438.

One open issue is that the comment is lost in the following example:

class X:
    __metaclass__ = Meta # Spam