Can you create an instance of a subclass with an existing instance of the base class?

[Sandra-24]

Can you create an instance of a subclass using an existing instance of
the base class?

Such things would be impossible in some languages or very difficult in
others. I wonder if this can be done in python, without copying the
base class instance, which in my case is a very expensive object.

Any ideas?

Thanks,
-Sandra

 

[Lawrence D'Oliveiro]

Can you create an instance of a subclass using an existing instance of
the base class?

I think you're taking Python's OO-ness too seriously. One of the
strengths of Python is that it can _look_ like an OO language without
actually being OO.

 

[Cavingdeep]

With new-style classes you can find out a class' subclasses and then
you can instantiate the subclass you want. Suppose you have two classes
A and B, B is a subclass of A, A is a new-style class. Now you have an
A's instance called "a", to instance B you can do the following:

b = a.__class__.__subclasses__()[0]()

 

[Peter Otten]

Can you create an instance of a subclass using an existing instance of
the base class?

Such things would be impossible in some languages or very difficult in
others. I wonder if this can be done in python, without copying the
base class instance, which in my case is a very expensive object.

You can change the class of an instance by assigning to the __class__
attribute. The new class doesn't even need to be a subclass of the old:
.... def __init__(self, name):
.... self.name = name
.... def show(self): print self.name
........ def show(self): print self.name.upper()
....ALPHA

Peter

 

[Sandra-24]

Now that is a clever little trick. I never would have guessed you can
assign to __class__, Python always surprises me in it's sheer
flexibility.

In this case it doesn't work.

TypeError: __class__ assignment: only for heap types

I suspect that's because this object begins its life in C code.

The technique of using the __class__.__subclasses__ also fails:

TypeError: cannot create 'B' instances

This seems more complex than I thought. Can one do this for an object
that beings it's life in C?

Thanks,
-Sandra

 

[Lawrence D'Oliveiro]

Now that is a clever little trick. I never would have guessed you can
assign to __class__, Python always surprises me in it's sheer
flexibility.

That's because you're still thinking in OO terms.

 

[Peter Otten]

Now that is a clever little trick. I never would have guessed you can
assign to __class__, Python always surprises me in it's sheer
flexibility.

In this case it doesn't work.

TypeError: __class__ assignment: only for heap types

I suspect that's because this object begins its life in C code.

The technique of using the __class__.__subclasses__ also fails:

TypeError: cannot create 'B' instances

This seems more complex than I thought. Can one do this for an object
that beings it's life in C?

The restriction originates in the metaclass. Perhaps you can use a
customized metaclass that allows subclass assignment, but I don't know
enough about Python's internals to tell you how/whether that is possible.

An alternative might be that you always start out with a subclass coded in
Python:
Traceback (most recent call last):
File "<stdin>", line 1, in ?
TypeError: __class__ assignment: only for heap types

.... first = property(lambda self: self[0])
....'a'

In the example a Tuple should be able to do everything a tuple can do;
because Tuple is coded in Python you can also change the __class__.

Peter

 

[Sandra-24]

That's because you're still thinking in OO terms.

It's not quite as simple as all that. I agree that people, escpecially
people with a Java (ew) background overuse OO, when there's often
simpler ways of doing things.

However in this case I'm simply getting an object (an mp_request object
from mod_python) passed into my function, and before I pass it on to
the functions that make up and individual web page it is modified by
adding members and methods to add functionality. It's not that I'm
thinking in OO, but that the object is a convienient place to put
things, especially functions that take an mp_request object as their
first argument.

Sadly I'm unable to create it as a python object first, because it's
created by the time my code comes into play. So I have to resort to
using the new module to add methods.

It works, but it has to be redone for every request, I thought moving
the extra functionality to another object would simplify the task. A
better way might be to contain the mp_request within another object and
use __getattr__ to lazily copy the inner object. I'd probably have to
first copy those few fields that are not read-only or use __setattr__
as well.

Thanks,
-Sandra

 

[Lawrence D'Oliveiro]

However in this case I'm simply getting an object (an mp_request object
from mod_python) passed into my function, and before I pass it on to
the functions that make up and individual web page it is modified by
adding members and methods to add functionality.

All you want is a dictionary, then. That's basically what Python objects
are.

 

[bruno at modulix]

Lawrence D'Oliveiro wrote:
(snip)

I think you're taking Python's OO-ness too seriously. One of the
strengths of Python is that it can _look_ like an OO language without
actually being OO.

According to which definition of OO ?

 

[Sandra-24]

All you want is a dictionary, then. That's basically what Python objects
are.

Yes, that's it exactly. I made a lazy wrapper for it, and I was really
happy with what I was able to accomplish, it turned out to be very
easy.

Thanks,
-Sandra

 

[Bruno Desthuilliers]

Sandra-24 a écrit :

It's not quite as simple as all that. I agree that people, escpecially
people with a Java (ew) background overuse OO, when there's often
simpler ways of doing things.

Nope. I mean : they don't overuse OO, they overuse *classes*. AFAIK, OO
means *object* oriented - not class oriented. There are OO languages
that don't even have a notion of class.

However in this case I'm simply getting an object (an mp_request object
from mod_python) passed into my function, and before I pass it on to
the functions that make up and individual web page it is modified by
adding members and methods to add functionality.

Which is a well-known design pattern called "decorator".

(snip)

Sadly I'm unable to create it as a python object first, because it's
created by the time my code comes into play. So I have to resort to
using the new module to add methods.

This is OK IMHO.

It works, but it has to be redone for every request,

Is this really a problem ?

I thought moving
the extra functionality to another object would simplify the task.
>
A
better way might be to contain the mp_request within another object and
use __getattr__ to lazily copy the inner object. I'd probably have to
first copy those few fields that are not read-only or use __setattr__
as well.

Why copy ? You could as well just use composition/delegation (also using
__getattr__ - and BTW, this is another possible implementation of the
decorator pattern).

 

[Lawrence D'Oliveiro]

Lawrence D'Oliveiro wrote:
(snip)

According to which definition of OO ?

Isn't there one?

 

[bruno at modulix]

Isn't there one?

Your claim that Python "_look_ like an OO language without actually
being OO" implicitely relies on a definition of OO - or is just
meaningless. So I ask you: what definition of OO do you use to support
your claim ?

 

[Lawrence D'Oliveiro]

Sandra-24 a écrit :

Nope. I mean : they don't overuse OO, they overuse *classes*. AFAIK, OO
means *object* oriented - not class oriented.

Oh great. Now we have someone redefining the concept of OO to evade the
point I was making.

There are OO languages that don't even have a notion of class.

Sounds like stuff I was doing in C (a non-OO language) years ago. Unless
you want to count C as an OO language, I think you're going to have to
retract this claim.

Which is a well-known design pattern called "decorator".

If you have to think of it as a design pattern, that means you haven't
figured out the code reuse angle yet.

 

[Lawrence D'Oliveiro]

Your claim that Python "_look_ like an OO language without actually
being OO" implicitely relies on a definition of OO - or is just
meaningless.

Which nicely evades answering the question.

 

[bruno at modulix]

Oh great. Now we have someone redefining the concept of OO to evade the
point I was making.

"redefining" ? lol...

Sounds like stuff I was doing in C (a non-OO language) years ago. Unless
you want to count C as an OO language, I think you're going to have to
retract this claim.

I think I'm not going to retract anything. And I think you should learn
a bit more about prototype-based languages.

If you have to think of it as a design pattern, that means you haven't
figured out the code reuse angle yet.

Please stop saying non-sense and learn the difference between design and
implementation.

 

[Marc 'BlackJack' Rintsch]

Oh great. Now we have someone redefining the concept of OO to evade the
point I was making.


Sounds like stuff I was doing in C (a non-OO language) years ago. Unless
you want to count C as an OO language, I think you're going to have to
retract this claim.

That sounds like stuff you do in a language that has objects but no
classes. As C has no objects I would not count it as an OO language. But
I count Io as an OO language::

#!/usr/bin/env io
Foo := Object clone
Foo value := 42
Foo setValue := method(newValue, self value = newValue; self)
Foo beep := method("beep" linePrint)
Foo asString := method("I'm a Foo. My value is " .. self value)

Bar := Foo clone
Bar asString := method("I'm a Bar and " .. super asString)

foo := Foo clone
foo beep
foo asString linePrint

bar := Bar clone setValue(23)
bar beep
bar asString linePrint

Output is:

beep
I'm a Foo. My value is 42
beep
I'm a Bar and I'm a Foo. My value is 23

That's OO IMHO. Clonable objects that know their "ancestors" so you can
build an object hierarchy. Missing attributes are looked up in the
"ancestors" and one can explicitly look up the inheritance tree with
``super``. There are no classes, just four objects. Convention in naming
and usage makes two of them something like templates or "classes" for new
objects but they are in no way special.

Ciao,
Marc 'BlackJack' Rintsch

 

[Steve Holden]

Which nicely evades answering the question.

Well I have to say you are also nicely evading answering the question,
which is enough to make me suspect your are trolling (deliberately
asking contentious questions for the purposes of creating futile
argument and discussion).

If you *aren't* trolling then what's your objection to saying what led
you to make the assertion that Python could look like an OO language
without being one?

But sine you say later that "Python objects are basically dictionaries"
it's clear your understanding of Python isn't terribly complete, which
might cast doubt on your understanding of object orientation.

For the record, Python *is* an object-oriented language, but it happens
to offer convenient features for procedural programming as well. Since
these features are orthogonal to its OO features, the fact that they
exist doesn't stop Python from being an OO language.

So why do you assert that it "merely looks like" one?

regards
Steve