Looking for magic method to override to prevent dict(d) from grabbing subclass inst d contents directly

[Bengt Richter]

Has anyone found a way besides not deriving from dict?
Shouldn't there be a way?
TIA
(need this for what I hope is an improvement on the Larosa/Foord OrderedDict ;-)

I guess I can just document that you have to spell it dict(d.items()), but I'd
like to hide the internal shenanigans ;-)

Regards,
Bengt Richter

[Mike Meyer]

bo**@oz.net (Bengt Richter) writes:

Has anyone found a way besides not deriving from dict?
Shouldn't there be a way?
TIA
(need this for what I hope is an improvement on the Larosa/Foord OrderedDict ;-)

I guess I can just document that you have to spell it dict(d.items()), but I'd
like to hide the internal shenanigans ;-)

So why not just hide them:

class md(dict):

.... def __new__(cls, arg):
.... return dict.__new__(cls, arg.items())
I'm not sure exactly what you mean by "grabbing sublass inst d contens
directly", but if dict(d.items()) does it, the above class should do
it as well. Of course, *other* ways of initializing a dict won't work
for this class.

<mike
--
Mike Meyer <mw*@mired.org> http://www.mired.org/home/mwm/
Independent WWW/Perforce/FreeBSD/Unix consultant, email for more information.

[Bengt Richter]

On Tue, 22 Nov 2005 09:30:49 -0500, Mike Meyer <mw*@mired.org> wrote:

bo**@oz.net (Bengt Richter) writes:

Has anyone found a way besides not deriving from dict?
Shouldn't there be a way?
TIA
(need this for what I hope is an improvement on the Larosa/Foord OrderedDict ;-)

I guess I can just document that you have to spell it dict(d.items()), but I'd
like to hide the internal shenanigans ;-)

So why not just hide them:

class md(dict):... def __new__(cls, arg):
... return dict.__new__(cls, arg.items())
I'm not sure exactly what you mean by "grabbing sublass inst d contens
directly", but if dict(d.items()) does it, the above class should do
it as well. Of course, *other* ways of initializing a dict won't work
for this class.

It's not initializing the dict subclass that is a problem, it is passing
it to the plain dict builtin as a constructor argument and being able to
control what is pulled from the subclass object to do the construction
of the ordinary dict.

Here's where I am on this: (odictb is my version of odict ;-)

from odictb import OrderedDict
d = OrderedDict()
d['a'] = 1
d['b'] = 2
d {'a': 1, 'b': 2} dict(d.items()) {'a': 1, 'b': 2}

Looks normal, but I am secretly using key:(insertionseqno, value) to represent key:value
dict(d)

{'a': (0, 1), 'b': (1, 2)}

and I want to control dict(d) that so the result is like dict(d.items()) above.

The dict builtin apparently skips looking for methods in the dict subclass. If it were
a subclass of object, dict would probably be looking for __iter__ or __getitem__ or __???___
but with a dict subclass it seems to bypass such a check. I was hoping that even so
there might be a __???__ that I didn't know about.

[Oop, just discovered a bug in my implementation ;-/ Hope it doesn't ripple...]

Regards,
Bengt Richter

[Mike Meyer]

bo**@oz.net (Bengt Richter) writes:

I'm not sure exactly what you mean by "grabbing sublass inst d contens
directly", but if dict(d.items()) does it, the above class should do
it as well. Of course, *other* ways of initializing a dict won't work
for this class.

It's not initializing the dict subclass that is a problem, it is passing
it to the plain dict builtin as a constructor argument and being able to
control what is pulled from the subclass object to do the construction
of the ordinary dict.

Ah, I got it backwards.

In looking through the dict sources in 2.4.2, it looks like the
initializer creates the new dictionary, then calls the C update for
dictionary it was passed. Update checks to see if the "other" is a
real dictionary, and if so assumes that it can just use the C code for
a dict for performance reasons. So there's no way to do what you're
trying to do. You either have to not pass the dictionary, or not
subclass dictionary. You might subclass UserDict.

IIRC, there are other places where this kind of thing happens:
specifically with files before.

<mike
--
Mike Meyer <mw*@mired.org> http://www.mired.org/home/mwm/
Independent WWW/Perforce/FreeBSD/Unix consultant, email for more information.