derived / base class name conflicts

[christopherlmarshall@yahoo.com]

Suppose you want to write a subclass of some existing class you are
importing from a module you didn't write and that you don't want to
study the internals of, and you want to define a data member i in your
constructor.

As in the following:

from module1 import A

class B(A):
def __init__(self):
A.__init__(self )
self.i= 0
b= B()

Now, 'i' might have already been defined by A or by the call to
A.__init__() so if you define it without knowing that, you could be
changing the behavior of A's methods in unknown ways, which is
obviously a bad thing.

One way to avoid this is to run the following program to clear the name
'i' first:

from module1 import A
a= A()
print a.i

If you get an AttributeError, you know the name 'i' is safe to use. If
you actually get some sort of report from the print statement, then you
will know that 'i' is not safe to use.

This strikes me as a rather odd procedure to go through, but I don't
see any way around it.

It there some other way to handle this issue?

Do I actually need to study the sources for / implementation of
Tkinter.Canvas before I can safely subclass it (and assign attributes
to an instance of the subclass) or clear attribute names as I outlined
above?

Chris Marshall

[Steve Juranich]

Re: derived / base class name conflicts

This should prove most enlightening:

import Tkinter
dir(Tkinter.Can vas)



On 10 Nov 2005 14:53:04 -0800, christopherlmar shall@yahoo.com
<christopherlma rshall@yahoo.co m> wrote:[color=blue]
> Suppose you want to write a subclass of some existing class you are
> importing from a module you didn't write and that you don't want to
> study the internals of, and you want to define a data member i in your
> constructor.
>
> As in the following:
>
> from module1 import A
>
> class B(A):
> def __init__(self):
> A.__init__(self )
> self.i= 0
> b= B()
>
> Now, 'i' might have already been defined by A or by the call to
> A.__init__() so if you define it without knowing that, you could be
> changing the behavior of A's methods in unknown ways, which is
> obviously a bad thing.
>
> One way to avoid this is to run the following program to clear the name
> 'i' first:
>
> from module1 import A
> a= A()
> print a.i
>
> If you get an AttributeError, you know the name 'i' is safe to use. If
> you actually get some sort of report from the print statement, then you
> will know that 'i' is not safe to use.
>
> This strikes me as a rather odd procedure to go through, but I don't
> see any way around it.
>
> It there some other way to handle this issue?
>
> Do I actually need to study the sources for / implementation of
> Tkinter.Canvas before I can safely subclass it (and assign attributes
> to an instance of the subclass) or clear attribute names as I outlined
> above?
>
> Chris Marshall
>
> --
> http://mail.python.org/mailman/listinfo/python-list
>[/color]


--
Steve Juranich
Tucson, AZ
USA

[Fredrik Lundh]

Re: derived / base class name conflicts

christopherlmar shall@yahoo.com wrote:
[color=blue]
> Now, 'i' might have already been defined by A or by the call to
> A.__init__() so if you define it without knowing that, you could be
> changing the behavior of A's methods in unknown ways, which is
> obviously a bad thing.[/color]

http://docs.python.org/tut/node11.html#SECTION0011600000000000000000

</F>

[christopherlmarshall@yahoo.com]

Re: derived / base class name conflicts

Steve Juranich wrote:[color=blue]
> This should prove most enlightening:
>
> import Tkinter
> dir(Tkinter.Can vas)
>
>[/color]

Huh?

Chris Marshall

[christopherlmarshall@yahoo.com]

Re: derived / base class name conflicts


Fredrik Lundh wrote:[color=blue]
> christopherlmar shall@yahoo.com wrote:
>[color=green]
> > Now, 'i' might have already been defined by A or by the call to
> > A.__init__() so if you define it without knowing that, you could be
> > changing the behavior of A's methods in unknown ways, which is
> > obviously a bad thing.[/color]
>
> http://docs.python.org/tut/node11.ht...00000000000000
>
> </F>[/color]

I see. Thanks for the link.

So putting two underscores in front of an instance variable (or any
identifier used inside the scope of a class statement) invokes a name
mangling mechanism ( __x becomes __classname_x internally)

so the following would not result in any conflicts

class A:
def __init__(self):
self.__i= 0

class B(A):
def __init__(self):
A.__init__(self )
self.__i= 1

Is it commonplace to use underscores when defining derived class
instance variables, or is this considered against the grain of python?

Chris Marshall

[Ben Finney]

Re: derived / base class name conflicts

christopherlmar shall@yahoo.com wrote:[color=blue]
> Suppose you want to write a subclass of some existing class you are
> importing from a module you didn't write and that you don't want to
> study the internals of[/color]

No need to study its internals. Fire up a Python interpreter and
inspect its outside:
[color=blue][color=green][color=darkred]
>>> import foomodule
>>> dir(foomodule.F ooClass)[/color][/color][/color]

Any attributes named there should be avoided in your subclasses if you
don't want to clobber existing behaviour.

--
\ "Those who will not reason, are bigots, those who cannot, are |
`\ fools, and those who dare not, are slaves." -- "Lord" George |
_o__) Gordon Noel Byron |
Ben Finney

[bruno at modulix]

Re: derived / base class name conflicts

christopherlmar shall@yahoo.com wrote:
(snip)[color=blue]
> So putting two underscores in front of an instance variable (or any
> identifier used inside the scope of a class statement) invokes a name
> mangling mechanism[/color]

(snip)
[color=blue]
> Is it commonplace to use underscores[/color]

I assume you mean double underscore...
[color=blue]
> when defining derived class
> instance variables,
> or is this considered against the grain of python?[/color]

I don't know if it's 'commonplace', but that's something I only do when
I absolutely want to protect a given attribute of a base class from
being accidentally overriden by a derived class. Else I use the usual
convention ('_name' => implementation, 'name' => API), and assume users
of my code will read the documentation (or the source FWIW) before
subclassing. Note that since 'users of my code' are mostly my coworkers
and I, this is a quite sensible assumption !-)

--
bruno desthuilliers
python -c "print '@'.join(['.'.join([w[::-1] for w in p.split('.')]) for
p in 'onurb@xiludom. gro'.split('@')])"

[christopherlmarshall@yahoo.com]

Re: derived / base class name conflicts

I see what you mean now.

It would indeed be enlightening if I wanted to study the internals of
Tkinter, and perhaps one day I will.

[christopherlmarshall@yahoo.com]

Re: derived / base class name conflicts

Your suggestion ('_name' -> implementation, 'name' -> API) makes sense
as a convention between programmers that know a fair amount about each
other's classes before using them.

I don't think it is reasonable in general to only subclass from base
classes you have studied the full API of, however. The double
underscore is a decent solution to my problem.

I imagine it must be used a lot in domains where people are making
heavy use of third party python code.

Chris Marshall

[Alex Martelli]

Re: derived / base class name conflicts

<christopherlma rshall@yahoo.co m> wrote:
...[color=blue]
> I don't think it is reasonable in general to only subclass from base
> classes you have studied the full API of, however. The double[/color]

I think you underestimate the level of coupling that inevitably occurs
between base and derived classes. In general, such coupling being
strong, knowing the full API of the base class is inevitable (although
in some special corner-case, when you do only need to add private data,
you may, exceptionally, get away with less knowledge).


Alex

[Bruno Desthuilliers]

Re: derived / base class name conflicts

christopherlmar shall@yahoo.com a écrit :[color=blue]
> Your suggestion ('_name' -> implementation, 'name' -> API)[/color]

This is not "my" convention, it's *the* (mostly agreed upon) Python
convention. Like 'self', or CONSTANT, or a whole lot of things in Python.
[color=blue]
> makes sense
> as a convention between programmers that know a fair amount about each
> other's classes before using them.[/color]

No need to know much. dir(modulename. ClassName) is enough.
[color=blue]
> I don't think it is reasonable in general to only subclass from base
> classes you have studied the full API of, however.[/color]

Depends on your definition of "studying the full API" !-)

Now if your fear is to accidentally override something in the base
class, it's just a matter of:
print "API:"
print [name for name in dir(modulename. ClassName) \
if not name.startswith ('_')]

print "IMPLEMENTATION :"
print [name for name in dir(modulename. ClassName) \
if not name.startswith ('_')]
[color=blue]
> The double
> underscore is a decent solution to my problem.[/color]

Possibly. It can also become a PITA. But it's you who know what your
code need !-)
[color=blue]
> I imagine it must be used a lot in domains where people are making
> heavy use of third party python code.[/color]

I almost never used it (perhaps half-a-dozen times, and only in
frameworks abstract base classes), and rarely saw it in 3rd part source
code.

[Steven Bethard]

Re: derived / base class name conflicts

christopherlmar shall@yahoo.com wrote:[color=blue]
> so the following would not result in any conflicts
>
> class A:
> def __init__(self):
> self.__i= 0
>
> class B(A):
> def __init__(self):
> A.__init__(self )
> self.__i= 1
>[/color]

Be careful here. The above won't result in any conflicts, but related
cases, where you have two classes with the same name in different
modules, may still result in conflicts. See my previous posts on this:

http://groups.google.com/group/comp.lang.python/msg/503984abaee1c2b5

http://groups.google.com/group/comp.lang.python/msg/f03183a2c01c8ecf

However, I tend not to use double-underscore name mangling, and I don't
think I've ever had a shadowing problem, so clearly I wouldn't have
shadowing problems with double-underscore name mangling either...

STeVe

[Steven Bethard]

Re: derived / base class name conflicts

christopherlmar shall@yahoo.com wrote:[color=blue]
> so the following would not result in any conflicts
>
> class A:
> def __init__(self):
> self.__i= 0
>
> class B(A):
> def __init__(self):
> A.__init__(self )
> self.__i= 1
>[/color]

Be careful here. The above won't result in any conflicts, but related
cases, where you have two classes with the same name in different
modules, may still result in conflicts. See my previous posts on this:

http://groups.google.com/group/comp.lang.python/msg/503984abaee1c2b5

http://groups.google.com/group/comp.lang.python/msg/f03183a2c01c8ecf

However, I tend not to use double-underscore name mangling, and I don't
think I've ever had a shadowing problem, so clearly I wouldn't have
shadowing problems with double-underscore name mangling either...

STeVe