How should I expose read-only fields from Python classes?
Question:
I have many different small classes which have a few fields each, e.g. this:
class Article:
def __init__(self, name, available):
self.name = name
self.available = available
What’s the easiest and/or most idiomatic way to make the name
field read only, so that
a = Article("Pineapple", True)
a.name = "Banana" # <-- should not be possible
is not possible anymore?
Here’s what I considered so far:
-
Use a getter (ugh!).
class Article:
def __init__(self, name, available):
self._name = name
self.available = available
def name(self):
return self._name
Ugly, non-pythonic – and a lot of boilerplate code to write (especially if I have multiple fields to make read-only). However, it does the job and it’s easy to see why that is.
-
Use __setattr__
:
class Article:
def __init__(self, name, available):
self.name = name
self.available = available
def __setattr__(self, name, value):
if name == "name":
raise Exception("%s property is read-only" % name)
self.__dict__[name] = value
Looks pretty on the caller side, seems to be the idiomatic way to do the job – but unfortunately I have many classes with only a few fields to make read only each. So I’d need to add a __setattr__
implementation to all of them. Or use some sort of mixin maybe? In any case, I’d need to make up my mind how to behave in case a client attempts to assign a value to a read-only field. Yield some exception, I guess – but which?
-
Use a utility function to define properties (and optionally getters) automatically. This is basically the same idea as (1) except that I don’t write the getters explicitely but rather do something like
class Article:
def __init__(self, name, available):
# This function would somehow give a '_name' field to self
# and a 'name()' getter to the 'Article' class object (if
# necessary); the getter simply returns self._name
defineField(self, "name")
self.available = available
The downside of this is that I don’t even know if this is possible (or how to implement it) since I’m not familiar with runtime code generation in Python. 🙂
So far, (2) appears to be most promising to me except for the fact that I’ll need __setattr__
definitions to all my classes. I wish there was a way to ‘annotate’ fields so that this happens automatically. Does anybody have a better idea?
For what it’s worth, I’mu sing Python 2.6.
UPDATE:
Thanks for all the interesting responses! By now, I have this:
def ro_property(o, name, value):
setattr(o.__class__, name, property(lambda o: o.__dict__["_" + name]))
setattr(o, "_" + name, value)
class Article(object):
def __init__(self, name, available):
ro_property(self, "name", name)
self.available = available
This seems to work quite nicely. The only changes needed to the original class are
- I need to inherit
object
(which is not such a stupid thing anyway, I guess)
- I need to change
self._name = name
to ro_property(self, "name", name)
.
This looks quite neat to me – can anybody see a downside with it?
Answers:
I would stick with your option 1 but refined it to use Python property:
class Article
def get_name(self):
return self.__name
name = property(get_name)
I would use property
as a decorator to manage your getter for name
(see the example for the class Parrot
in the documentation). Use, for example, something like:
class Article(object):
def __init__(self, name, available):
self._name = name
self.available = available
@property
def name(self):
return self._name
If you do not define the setter for the name
property (using the decorator x.setter
around a function) this throws an AttributeError
when you try and reset name
.
Note: You have to use Python’s new-style classes (i.e. in Python 2.6 you have to inherit from object
) for properties to work correctly. This is not the case according to @SvenMarnach.
It should be noted that it’s always possible to modify attributes of an object in Python – there are no truly private variables in Python. It’s just that some approaches make it a bit harder. But a determined coder can always lookup and modify the value of an attribute. For example, I can always modify your __setattr__
if I want to…
For more information, see Section 9.6 of The Python Tutorial. Python uses name mangling when attributes are prefixed with __
so the actual name at runtime is different but you could still derive what that name at runtime is (and thus modify the attribute).
Based in the Chris answer, but arguably more pythonic:
def ro_property(field):
return property(lambda self : self.__dict__[field])
class Article(object):
name = ro_property('_name')
def __init__(self):
self._name = "banana"
If trying to modify the property it will raise an AttributeError
.
a = Article()
print a.name # -> 'banana'
a.name = 'apple' # -> AttributeError: can't set attribute
UPDATE: About your updated answer, the (little) problem I see is that you are modifying the definition of the property in the class every time you create an instance. And I don’t think that is such a good idea. That’s why I put the ro_property
call outside of the __init__
function
What about?:
def ro_property(name):
def ro_property_decorator(c):
setattr(c, name, property(lambda o: o.__dict__["_" + name]))
return c
return ro_property_decorator
@ro_property('name')
@ro_property('other')
class Article(object):
def __init__(self, name):
self._name = name
self._other = "foo"
a = Article("banana")
print a.name # -> 'banana'
a.name = 'apple' # -> AttributeError: can't set attribute
Class decorators are fancy!
As pointed out in other answers, using a property is the way to go for read-only attributes. The solution in Chris’ answer is the cleanest one: It uses the property()
built-in in a straight-forward, simple way. Everyone familiar with Python will recognize this pattern, and there’s no domain-specific voodoo happening.
If you don’t like that every property needs three lines to define, here’s another straight-forward way:
from operator import attrgetter
class Article(object):
def __init__(self, name, available):
self._name = name
self.available = available
name = property(attrgetter("_name"))
Generally, I don’t like defining domain-specific functions to do something that can be done easily enough with standard tools. Reading code is so much easier if you don’t have to get used to all the project-specific stuff first.
I have many different small classes which have a few fields each, e.g. this:
class Article:
def __init__(self, name, available):
self.name = name
self.available = available
What’s the easiest and/or most idiomatic way to make the name
field read only, so that
a = Article("Pineapple", True)
a.name = "Banana" # <-- should not be possible
is not possible anymore?
Here’s what I considered so far:
-
Use a getter (ugh!).
class Article: def __init__(self, name, available): self._name = name self.available = available def name(self): return self._name
Ugly, non-pythonic – and a lot of boilerplate code to write (especially if I have multiple fields to make read-only). However, it does the job and it’s easy to see why that is.
-
Use
__setattr__
:class Article: def __init__(self, name, available): self.name = name self.available = available def __setattr__(self, name, value): if name == "name": raise Exception("%s property is read-only" % name) self.__dict__[name] = value
Looks pretty on the caller side, seems to be the idiomatic way to do the job – but unfortunately I have many classes with only a few fields to make read only each. So I’d need to add a
__setattr__
implementation to all of them. Or use some sort of mixin maybe? In any case, I’d need to make up my mind how to behave in case a client attempts to assign a value to a read-only field. Yield some exception, I guess – but which? -
Use a utility function to define properties (and optionally getters) automatically. This is basically the same idea as (1) except that I don’t write the getters explicitely but rather do something like
class Article: def __init__(self, name, available): # This function would somehow give a '_name' field to self # and a 'name()' getter to the 'Article' class object (if # necessary); the getter simply returns self._name defineField(self, "name") self.available = available
The downside of this is that I don’t even know if this is possible (or how to implement it) since I’m not familiar with runtime code generation in Python. 🙂
So far, (2) appears to be most promising to me except for the fact that I’ll need __setattr__
definitions to all my classes. I wish there was a way to ‘annotate’ fields so that this happens automatically. Does anybody have a better idea?
For what it’s worth, I’mu sing Python 2.6.
UPDATE:
Thanks for all the interesting responses! By now, I have this:
def ro_property(o, name, value):
setattr(o.__class__, name, property(lambda o: o.__dict__["_" + name]))
setattr(o, "_" + name, value)
class Article(object):
def __init__(self, name, available):
ro_property(self, "name", name)
self.available = available
This seems to work quite nicely. The only changes needed to the original class are
- I need to inherit
object
(which is not such a stupid thing anyway, I guess) - I need to change
self._name = name
toro_property(self, "name", name)
.
This looks quite neat to me – can anybody see a downside with it?
I would stick with your option 1 but refined it to use Python property:
class Article
def get_name(self):
return self.__name
name = property(get_name)
I would use property
as a decorator to manage your getter for name
(see the example for the class Parrot
in the documentation). Use, for example, something like:
class Article(object):
def __init__(self, name, available):
self._name = name
self.available = available
@property
def name(self):
return self._name
If you do not define the setter for the name
property (using the decorator x.setter
around a function) this throws an AttributeError
when you try and reset name
.
Note: You have to use Python’s new-style classes (i.e. in Python 2.6 you have to inherit from This is not the case according to @SvenMarnach.object
) for properties to work correctly.
It should be noted that it’s always possible to modify attributes of an object in Python – there are no truly private variables in Python. It’s just that some approaches make it a bit harder. But a determined coder can always lookup and modify the value of an attribute. For example, I can always modify your __setattr__
if I want to…
For more information, see Section 9.6 of The Python Tutorial. Python uses name mangling when attributes are prefixed with __
so the actual name at runtime is different but you could still derive what that name at runtime is (and thus modify the attribute).
Based in the Chris answer, but arguably more pythonic:
def ro_property(field):
return property(lambda self : self.__dict__[field])
class Article(object):
name = ro_property('_name')
def __init__(self):
self._name = "banana"
If trying to modify the property it will raise an AttributeError
.
a = Article()
print a.name # -> 'banana'
a.name = 'apple' # -> AttributeError: can't set attribute
UPDATE: About your updated answer, the (little) problem I see is that you are modifying the definition of the property in the class every time you create an instance. And I don’t think that is such a good idea. That’s why I put the ro_property
call outside of the __init__
function
What about?:
def ro_property(name):
def ro_property_decorator(c):
setattr(c, name, property(lambda o: o.__dict__["_" + name]))
return c
return ro_property_decorator
@ro_property('name')
@ro_property('other')
class Article(object):
def __init__(self, name):
self._name = name
self._other = "foo"
a = Article("banana")
print a.name # -> 'banana'
a.name = 'apple' # -> AttributeError: can't set attribute
Class decorators are fancy!
As pointed out in other answers, using a property is the way to go for read-only attributes. The solution in Chris’ answer is the cleanest one: It uses the property()
built-in in a straight-forward, simple way. Everyone familiar with Python will recognize this pattern, and there’s no domain-specific voodoo happening.
If you don’t like that every property needs three lines to define, here’s another straight-forward way:
from operator import attrgetter
class Article(object):
def __init__(self, name, available):
self._name = name
self.available = available
name = property(attrgetter("_name"))
Generally, I don’t like defining domain-specific functions to do something that can be done easily enough with standard tools. Reading code is so much easier if you don’t have to get used to all the project-specific stuff first.