Python 3 __dict__ inheritance

[QuLogic]

For whatever reason, the __dict__ property does not get inherited, so it needs to be re-implemented in a few classes.

[rhattersley]

It might be better to fix this by defining __slots__ = () for these classes (instead of defining the __dict__ property.)

[QuLogic]

I'm not sure that would do anything as I think that's already true:

In [1]: from collections import namedtuple
In [2]: Foo = namedtuple('Foo', 'foo bar')
In [3]: class Bar(Foo): __eq__ = lambda self, other: True

In [4]: f = Foo(1, 2)
In [5]: f.__slots__
Out[5]: ()

In [6]: b = Bar(1, 2)
In [7]: b.__slots__
Out[7]: ()

[rhattersley]

I'm not sure that would do anything as I think that's already true:

Without getting in to the subtleties of it all (I'd have to spend a while researching them!):

>>> class Wibble(Foo): pass
>>> Wibble(1, 2).__dict__
{}

>>> class Wobble(Foo): __slots__ = ()
>>> Wobble(1, 2).__dict__
OrderedDict([('foo', 1), ('bar', 2)])

[QuLogic]

Oh, of course, I tested it by setting __slots__ after class creation and not during.

[QuLogic]

OK, switched to __slots__, but only where necessary. Does it make sense to add it on all the namedtuple derivatives?

[rhattersley]

Does it make sense to add it on all the namedtuple derivatives?

Not sure... probably yes. It appeals from a consistency point of view, and the derivatives were all intended to be immutable.

Out of interest, how did you spot the need for this PR? (I'm trying to recreate the relevant test failures so I can better understand the possible impact.)

[rhattersley]

Out of interest, how did you spot the need for this PR? (I'm trying to recreate the relevant test failures so I can better understand the possible impact.)

I think I've tracked this down to things like _CoordMetaData.__eq__ failing because its calls to self._asdict() and other._asdict() return {} on Python 3 without this PR.

[rhattersley]

Does it make sense to add it on all the namedtuple derivatives?

Not sure... probably yes

Enough vagueness! If I have to choose an option, then 👍 for slots. 😉

[QuLogic]

I added it to everything except ArrayStructure, where it cannot be added because the __new__ sets the self.size attribute.

[QuLogic]

Tests currently broken by ContinuumIO/anaconda-issues#445.

[rhattersley]

I added it to everything except ArrayStructure, where it cannot be added because the new sets the self.size attribute.

We could define size as a property instead?

[QuLogic]

Yep, seems to work alright.

[rhattersley]

Looks good. 👍