About the use of **args

[Zunbeltz Izaola]

Hi

I'm starting a new proyect and i'm in doubt about diferent interfaces
for my clases. My clases will have a lot of attributes and i'm want to
know what aproach could be the best

1) Define one SetAttribute/GetAttribute pair of method for each
attribute.
2) Define one SetAttribute/GetAttribute which argument is a key=value
format.

Any advaice?

Thanks in advance

Zunbeltz Izaola

 

[Graham Ashton]

I'm starting a new proyect and i'm in doubt about diferent interfaces
for my clases. My clases will have a lot of attributes and i'm want to
know what aproach could be the best

1) Define one SetAttribute/GetAttribute pair of method for each
attribute.
2) Define one SetAttribute/GetAttribute which argument is a key=value
format.

I asked a similar question myself when I started out with Python, and I
got some very good answers that served me well.

http://groups.google.com/groups?hl=...CZ6.1492%24h45.9746%40news.uk.colt.net&rnum=2

My preferred technique is to just set attributes directly on the object,
like this:

shoe = Shoe()
shoe.laced_up = True
shoe.is_on_foot = True
shoe.resoled = False

You may think that it's a bit nasty because it breaks encapsulation, etc.,
but Python is designed to let you do stuff quickly and cleanly. You're
not always in a situation where you need to enforce access so
strictly. Using attributes directly is more of a Python idiom (see the
last two posts in the thread I posted earlier).

Since that earlier thread Python has been given a nice way of
customising attribute access; properties. Basically, they let you
define get/set methods that get run when you access/assign to an
attribute. You benefit by getting a cleaner interface to your class, and
you only have to define get/set methods when you actually need them (i.e.
when they have side effects other than getting/setting the attribute).

For example:

class Shoe:

def __init__(self):
self._laced_up = False
self.is_on_foot = False
self.resoled = False

def _set_laced_up(self, boolean):
self._laced_up = boolean
if boolean:
self.is_on_foot = True

def _get_laced_up(self):
return self._laced_up

laced_up = property(_get_laced_up, _set_laced_up)

I've not run that so it may have syntax errors. It should illustrate the
principle though; you can use attributes directly until you want to take
actions when you set them. Then you can make the attribute "private" and
replace it with a property that access the real attribute, and does
whatever else you want to do when it's accessed.

Hope that makes sense.

-- Graham

 

[Mark McEahern]

[snip]

I've not run that so it may have syntax errors. It should illustrate the
principle though; you can use attributes directly until you want to take
actions when you set them. Then you can make the attribute "private" and
replace it with a property that access the real attribute, and does
whatever else you want to do when it's accessed.

A couple of points on Graham's example:

1. You have to subclass from object in order to use property.

2. When Graham nominalizes private, he means that you can still access
the attribute variable directly--and that's by design.

See example below.

Cheers,

// m

#!/usr/bin/env python

class Shoe:

def __init__(self):
self._size = None
def getSize(self):
print "In getSize..."
return self._size
def setSize(self, size):
print "In setSize..."
self._size = size
size = property(getSize, setSize)

class Shoe2(object):

def __init__(self):
self._size = None
def getSize(self):
print "In getSize..."
return self._size
def setSize(self, size):
print "In setSize..."
self._size = size
size = property(getSize, setSize)

s = Shoe()
# Since we haven't subclassed from object, the property descriptor
# doesn't work; notice that setSize() isn't getting called...
s.size = 1
print s.size
# We can still access the "private" member variable. In Python,
# private is merely a convention.
print s._size

# Now that we've subclassed from object, our get/set methods are
# called.
s2 = Shoe2()
s2.size = 1
print s2.size
# And we can still access the "private" member variable.
print s2._size

 

[Mark McEahern]

[snip]

I've not run that so it may have syntax errors. It should illustrate the
principle though; you can use attributes directly until you want to take
actions when you set them. Then you can make the attribute "private" and
replace it with a property that access the real attribute, and does
whatever else you want to do when it's accessed.

A couple of points on Graham's example:

1. You have to subclass from object in order to use property.

2. When Graham nominalizes private, he means that you can still access
the attribute variable directly--and that's by design.

See example below.

Cheers,

// m

#!/usr/bin/env python

class Shoe:

def __init__(self):
self._size = None
def getSize(self):
print "In getSize..."
return self._size
def setSize(self, size):
print "In setSize..."
self._size = size
size = property(getSize, setSize)

class Shoe2(object):

def __init__(self):
self._size = None
def getSize(self):
print "In getSize..."
return self._size
def setSize(self, size):
print "In setSize..."
self._size = size
size = property(getSize, setSize)

s = Shoe()
# Since we haven't subclassed from object, the property descriptor
# doesn't work; notice that setSize() isn't getting called...
s.size = 1
print s.size
# We can still access the "private" member variable. In Python,
# private is merely a convention.
print s._size

# Now that we've subclassed from object, our get/set methods are
# called.
s2 = Shoe2()
s2.size = 1
print s2.size
# And we can still access the "private" member variable.
print s2._size

 

[John Roth]

Hi

I'm starting a new proyect and i'm in doubt about diferent interfaces
for my clases. My clases will have a lot of attributes and i'm want to
know what aproach could be the best

1) Define one SetAttribute/GetAttribute pair of method for each
attribute.
2) Define one SetAttribute/GetAttribute which argument is a key=value
format.

Any advaice?

Thanks in advance

Zunbeltz Izaola

I presume you're talking about constructing the object,
not about state changes after it's been constructed and
released into the wild?

In that case, the "best" pattern is to never let the object
appear half constructed. There are a wide variety of ways
of doing this, from passing all the needed parameters into
the constructor, to numerous variations on the factory
pattern. And I wouldn't be adverse to the other responder's
practice of simply plugging values into the instance while
it's still in the factory: most real world objects don't contain
the logic for their construction, so why should our programming
objects?

John Roth

 

[Jacek Generowicz]

Hi

I'm starting a new proyect and i'm in doubt about diferent interfaces
for my clases. My clases will have a lot of attributes and i'm want to
know what aproach could be the best

1) Define one SetAttribute/GetAttribute pair of method for each
attribute.
2) Define one SetAttribute/GetAttribute which argument is a key=value
format.

Any advaice?

3) Not have any setters or getters at all.

Ask yourself: "what is the point of getters and setters?".

The most common answers are:

a) Data hiding, privacy, etc.

b) To protect myself against breaking the interface when I change the
implementation of my object.

c) Give you and your users typing exercise.

In Python there is no enforced data hiding or privacy: you use a
leading underscore in the identifier name to indicate to your users
that something is _not_ part of the interface ... so (a) is irrelevant.

As for (b), Python provides properties, which allow you to replace an
attribute with a setter and/or a getter method, while maintaining the
illusion that there is only an attribute. Properties also allow you to
implement read-only (or write-only! (or delete-only!!)) attributes.

(I won't comment on (c)

If you are a fan of Bertrand Meyer, you might also answer

d) Attribute access and method calls should look the same.

I'm not aware of a Python mechanism to make that irrelevant.

 

[Duncan Booth]

# We can still access the "private" member variable. In Python,
# private is merely a convention.
print s._size

Except that in Python the convention for private variables is to begin them
with two underscore characters, and then it is more than just a convention:

def __init__(self):
self.__size = None
def getSize(self):
print "In getSize..."
return self.__size
def setSize(self, size):
print "In setSize..."
self.__size = size
size = property(getSize, setSize)

In getSize...
1
Traceback (most recent call last):
File "<pyshell#11>", line 1, in -toplevel-
print s2.__size
AttributeError: 'Shoe2' object has no attribute '__size'

Of course, it doesn't absolutely prevent you accessing the private variable
if you know what you are doing, but then neither do C++ or Java:
1

 

[Jacek Generowicz]

Except that in Python the convention for private variables is to begin them
with two underscore characters, and then it is more than just a convention:

Now, I always believed that "private" in Python is spelt "_", while
"mangle this name for me because I want to avoid name clashes in
multiple inheritance scenarios" is spelt "__".

However, I've heard _so_ many people make the (in my opinion
incorrect) claim that "private" is spelt "__", that I'm beginning to
question my sanity.

Where is the Python spelling of "private" set in stone ?

I find the suggestion that a mangled name is somehow "more private"
than one with a single leading underscore, mildly insulting to the
clients of your code. The implcation is that they can neither work out
the (very complicated and cryptic[*]) mangling scheme, not use dir().

If a Python programmer wants to acces a private attribute, he can do
so, regardless of the number of leading underscores.


[*] Sarcasm alert. (This alert was brought to you by the International
Campaign for Promotion of Understanding between Cultures.)

 

[Fredrik Lundh]

I find the suggestion that a mangled name is somehow "more private"
than one with a single leading underscore, mildly insulting to the
clients of your code. The implcation is that they can neither work out
the (very complicated and cryptic[*]) mangling scheme, not use dir().

no, the implication is that they can use their own __variables without
having to think about what you're calling your private parts.

in contrast, a _variable is not hidden at all.

</F>

 

[Duncan Booth]

Now, I always believed that "private" in Python is spelt "_", while
"mangle this name for me because I want to avoid name clashes in
multiple inheritance scenarios" is spelt "__".

However, I've heard _so_ many people make the (in my opinion
incorrect) claim that "private" is spelt "__", that I'm beginning to
question my sanity.

Where is the Python spelling of "private" set in stone ?

Section 5.2.1 of the Python reference manual in the paragraph headed
'Private name mangling'.

I find the suggestion that a mangled name is somehow "more private"
than one with a single leading underscore, mildly insulting to the
clients of your code. The implcation is that they can neither work out
the (very complicated and cryptic[*]) mangling scheme, not use dir().

Not at all, the mangling is clearly documented. Its purpose is most
emphatically NOT to prevent access to your clients.

If a Python programmer wants to acces a private attribute, he can do
so, regardless of the number of leading underscores.

Good points. I suppose it all depends on your definition of 'private'. To
me, there are two reasons why someone might want to use 'private'
variables. One is to prevent accidental nameclashes when someone subclasses
one of your classes, the other is because you don't trust those poxy
programmers who are going to use your class, so you want to burn their
fingers if they try.

The first of these, Python provides. Ok, it doesn't get it quite right (the
name mangling should maybe include the module and package as well as the
class), but it does a reasonable job of minimising accidental nameclashes.
Python does not attempt to address the second of these issues, preferring
to believe that all Python programmers are mature & responsible.

C++ fails dismally on the first count, it doesn't hide the names so adding
a 'private member' to a base class can break the derived classes. It also
fails on the second count, since you can easily bypass private if you wish.

To my mind, data hiding is a good reason for using private variables, but
preventing knowledgeable users from bypassing the privacy is dubious at
best. You are of course free to disagree.

 

[Peter Hansen]

Now, I always believed that "private" in Python is spelt "_", while
"mangle this name for me because I want to avoid name clashes in
multiple inheritance scenarios" is spelt "__".

However, I've heard _so_ many people make the (in my opinion
incorrect) claim that "private" is spelt "__", that I'm beginning to
question my sanity.

Distinguish "private" and "protected", as in C++ and perhaps wider usage.

"private" is spelled __ in Python and is supported in a limited fashion
by the language, while *protected* is spelled _ and is by convention.

-Peter

 

[Zunbeltz Izaola]

I want to sum up some ideas in the thread:

1) setters and getters are not mandatory.

2) The usual way in Python is the direct access

[=] ... b.width # is get
b.width = # is set
del b.width # is del

3) If when setting some attribute there is necesary to have side
effects, it can be achive defining __setattr__ and doing the right
thing depend on the attribute. (It is possible to raise an
exception when the attribute is not in a list/dictionary of usable
attributes, to prevent typos and such kind of errors)

4) __getattr__ can be defined to deal with undefined attributes. For
example it can handle in different way the attributes in a list of
permited attributes (see point 3)

5) attributes that are computed in the fly, can be get like normal
attributes using "property" (property is new in 2.3 or 2.2?)

6) attributes and methods that are no part of the user interface are
named with and _ in the front. (This is a convention)

Are those ideas correct? We can continue with the discussion!!

Zunbeltz

 

[Jacek Generowicz]

[snip]

3) If when setting some attribute there is necesary to have side
effects, it can be achive defining __setattr__ and doing the right
thing depend on the attribute. (It is possible to raise an
exception when the attribute is not in a list/dictionary of usable
attributes, to prevent typos and such kind of errors)

4) __getattr__ can be defined to deal with undefined attributes. For
example it can handle in different way the attributes in a list of
permited attributes (see point 3)

5) attributes that are computed in the fly, can be get like normal
attributes using "property" (property is new in 2.3 or 2.2?)
[snip]

Are those ideas correct?

Pretty much.

Properties are available in 2.2. They only work on new style classes.

Just about anything that can be done with properties can be done with
__setattr__ &co, but I think that properties are generally preferred,
if available.