[Python-ideas] namedtuple is not as good as it should be

[Christian Tismer]

Hi Raymond,

I have now written a preliminary and rough, but almost complete
implementation of namedtuple/namelesstuple.

The result is:

For simple tuples, the behavior is like I intended it to be:

- definitions are only created once, and on demand
- pickling works by value without class reference
- tuple-like behavior therefore seems to be complete

When you derive a class, the generated objects are aware of it
and fall back to their original behavior.
I.E. for derived classes, you need to write the pickling, yourself.

No idea if this can be extended, but it is sufficient to close this thread.

cheers - chris

https://bitbucket.org/ctismer/namelesstuple/


On 10.06.13 17:27, Christian Tismer wrote:
> Hi Raymond,
>
> I have written a simple implementation of something like a
> namelesstuple, as a POC.
>
> I did not make generic tuple classes, yet, there is no class generator.
> It shows that the result of pickling is quite compact and has only
> a single helper function to restore the pickled objects.
>
> What needs to be implemented is
>
> - the unique dict that holds all fieldname tuples
> - the class store that hold all generated classes.
>
> And of course the features that make namedtuple nice. ;-)
>
> What I wanted to see is how efficiently this stuff can be pickled.
>
> ------------------------------------------------------------
>
> # namelesstuple
> # proof of concept for nametuple implementation using
> # the identity of the tuple of field names.
>
> from operator import itemgetter as _itemgetter
>
> _tuple = tuple
> _repr_template = '{name}=%r'
>
> class _nlt1(tuple):
>     __slots__ = ()
>
>     _fields = ('eins', 'zwei', 'drei')
>
>     def __new__(_cls, args):
>         return _tuple.__new__(_cls, args)
>
>     eins = property(_itemgetter(0), doc='Alias for field number 0')
>     zwei = property(_itemgetter(1), doc='Alias for field number 1')
>     drei = property(_itemgetter(2), doc='Alias for field number 2')
>
>     def __repr__(self):
>         'Return a nicely formatted representation string'
>         fmt = '{}=%r'
>         return self._repr_fmt % self
>
>     _repr_fmt =  ', '.join(_repr_template.format(name=name)
>                                  for name in _fields)
>     _repr_fmt = '(%s)' % _repr_fmt
>
>     def __reduce__(self):
>         return (rebuild_namelesstuple, (tuple(self), self._fields))
>
>
> class _nlt2(tuple):
>     __slots__ = ()
>
>     _fields = ('alpha', 'beta', 'gamma')
>
>     def __new__(_cls, args):
>         return _tuple.__new__(_cls, args)
>
>     alpha = property(_itemgetter(0), doc='Alias for field number 0')
>     beta = property(_itemgetter(1), doc='Alias for field number 1')
>     gamma = property(_itemgetter(2), doc='Alias for field number 2')
>
>     def __repr__(self):
>         'Return a nicely formatted representation string'
>         fmt = '{}=%r'
>         return self._repr_fmt % self
>
>     _repr_fmt =  ', '.join(_repr_template.format(name=name)
>                                  for name in _fields)
>     _repr_fmt = '(%s)' % _repr_fmt
>
>     def __reduce__(self):
>         return (rebuild_namelesstuple, (tuple(self), self._fields))
>
>
> def rebuild_namelesstuple(args, fields):
>     # just a dummy for testing pickle
>     if fields[0] == 'eins':
>         return _nlt1(args)
>     else:
>         return _nlt2(args)
>
> # quick check
> insts = [
>     _nlt1((2, 3, 5)),
>     _nlt1((7, 8, 9)),
>     _nlt2((2, 3, 5)),
>     _nlt2((101, 102, 103)),
>     ]
>
>
> import pickletools, pickle
> x=pickle.dumps(insts)
> pickletools.dis(x)
> pickle.loads(x)
>


-- 
Christian Tismer             :^)   <mailto:tismer at stackless.com>
Software Consulting          :     Have a break! Take a ride on Python's
Karl-Liebknecht-Str. 121     :    *Starship* http://starship.python.net/
14482 Potsdam                :     PGP key -> http://pgp.uni-mainz.de
phone +49 173 24 18 776  fax +49 (30) 700143-0023
PGP 0x57F3BF04       9064 F4E1 D754 C2FF 1619  305B C09C 5A3B 57F3 BF04
       whom do you want to sponsor today?   http://www.stackless.com/