I currently need to partially create a Python object and be able to update it for some time. Although, I must not be able to update it once I used the object as a dictionary key.
Of course there is the solution of marking the fields as private, which is mostly a warning for the programmer, and I will actually go for that solution.
But I stumbled on another solution and I want to know if this could be a good idea, or if it could simply go horribly wrong. Here it is:
class Foo():
def __init__(self, bar):
self._bar = bar
self._has_been_hashed = False
def __hash__(self):
self._has_been_hashed = True
return self._bar.__hash__()
def __eq__(self, other):
return self._bar == other._bar
def __copy__(self):
return Foo(self._bar)
def set_bar(self, bar):
if self.has_been_hashed:
raise FooIsNowImmutable
else:
self._bar = bar
Some testing proved it to work as desired, I can no longer use set_bar once I, say, used my object as a dictionary key.
What do you think? Is it a good idea? Will it turn against me? Is there an easier way? And is it somehow a bad practice?
Doing it that way is a bit fragile, since you never know when something might be used as a dictionary key, or when its hash might be called for some other reason. An object isn't supposed to "know" whether it's being used as a dictionary key. It will be confusing to have code that may raise an exception just because some other code somewhere else put the object in a dictionary.
Following the Python philosophy of "explicit is better than implicit", it would be safer to just give your object a method called .finalize() or .lock() or something, which would set a flag indicating the object is immutable. You could also reverse the exception-raising logic, so that __hash__ raises an exception if the object is not yet locked (rather than mutation raising an exception if the object has been hashed).
You would then call .lock() when you're ready to make the object immutable. It makes more sense to explicitly set it immutable when you're done with whatever mutating you need to do, rather than implicitly assuming that as soon as you use it in a dictionary, you're done mutating it.
You can do that, but I'm not sure I'd recommend it. Why do you need it in a dictionary?
It requires a lot more awareness of the state of the object... think a file object. Would you put one in a dictionary? It has to be opened for a lot of the functions to work, and once it's closed, you can't do them anymore. The user has to be aware in the surrounding code which state the object is in.
For files, that makes sense - after all, you don't normally hold files open across large parts of your program, or if you do, they have very defined init and close codes; something similar has to make sense for your object. Especially if you have some APIs that take the object, but expect an immutable version, and others that take the same object, but expect to change it...
I have used the lock method before, and it works well for complex, read-only objects that you want to initialize once and then make sure no one is messing with. E.G. you load a copy of a (say, English) dictionary from disk... it has to be mutable while you are populating it, but you don't want anyone to accidentally modify it, so locking it is a great idea. I would only use it if it was a one-time lock though - something you are locking and unlocking seems like a recipe for disaster.
There are two solutions IMHO if you just want to create a version you can use in hashable places. First is to explicitly create an immutable copy when you put it in a dictionary - tuple and frozenset are examples of this sort of behaviour... if you want to put a list in a dict, you can't, but you can create a tuple from it first, and that can be hashed. Create a frozen version of your object, then it's very clear by looking at the object type whether it's expected to be mutable or immutable, and so cases where it was used incorrectly are easily seen.
Second, if you really want it to be hashable, but need it to be mutable... that's actually legal, but implemented a little different. It goes back to the idea of hashing... hashing is used both for optimized lookups, and equality.
The first is to ensure you can get objects back... you put something in a dictionary, and it hashes to a value of 4 - goes in slot 4. Then you modify it. Then you go to look it up again, and now it hashes to 9 - there's nothing in slot 9, or worse, a different object, and you're broken.
Second is equality - for things like sets, I need to know if my object is already in there. I can hash, but if you know anything about hashing, you still need to check equality to check for hash collisions.
That doesn't preclude supporting __hash__ and being mutable, but it's unusual. You need to decide for your item what makes it the same, even though it's mutable. What you need to do then is give each object a unique id. Technically, you may be able to get away with id(self), but something like the uuid module is probably a better possibility. The UUID4 (or technically, the hash of the UUID4) is what determines both the hash and equality; two objects that contain the same UUID4 should be the exact same object; two objects that have the exact same data but a different UUID4 would be different object.
Related
Why is everything in Python, an object? According to what I read, everything including functions is an object. It's not the same in other languages. So what prompted this shift of approach, to treat everything including, even functions, as objects.
The power of everything being an object is that you can define behavior for each object. For example a function being an object gives you an easy way to access the docs of the function for introspection.
print( function.__doc__ )
The alternative would be to provide a library of function that took
a function and returned its interesting properties.
import function_lib
print( function_lib.get_doc( function )
Making int, str etc classes means that you can extend those provide types
in interesting ways for your problem domain.
In my opinion, the 'Everything is object' is great in Python. In this language, you don't react to what are the objects you have to handle, but how they can interact. A function is just an object that you can __call__, a list is just an object that you can __iter__. But why should we divide data in non overlapping groups. An object can behave like a function when we call it, but also like an array when we access it.
This means that you don't think your "function" like, "i want an array of integers and i return the sum of it" but more "i will try to iterate over the thing that someone gave me and try to add them together, if something goes wrong, i will tell it to the caller by raiseing error and he will hate to modify his behavior".
The most interesting exemple is __add__. When you try something like Object1 + Object2, Python will ask (nicely ^^) to Object1 to try to add himself with object2 (Object1.__add__(Object2)). There is 2 scenarios here: either Oject1 knows how to add himself to Object2 and everything is fine, either he raises a NotImplemented error and Python will ask to Object2 to radd himself to Object1. Just with this mechanism, you can teach to your object to add themselves with any other object, you can manage commutativity,...
why is everything in Python, an object?
Python (unlike other languages) is a truly Object Orient language (aka OOP)
when everything is an object, it becomes easier to search, manipulate or access things. (But everything comes at the cost of speed)
what prompted this shift of approach, to treat everything including, even functions, as objects?
"Necessity is the mother of invention"
USAGE CONTEXT ADDED AT END
I often want to operate on an abstract object like a list. e.g.
def list_ish(thing):
for i in xrange(0,len(thing)):
print thing[i]
Now this appropriate if thing is a list, but will fail if thing is a dict for example. what is the pythonic why to ask "do you behave like a list?"
NOTE:
hasattr('__getitem__') and not hasattr('keys')
this will work for all cases I can think of, but I don't like defining a duck type negatively, as I expect there could be cases that it does not catch.
really what I want is to ask.
"hey do you operate on integer indicies in the way I expect a list to do?" e.g.
thing[i], thing[4:7] = [...], etc.
NOTE: I do not want to simply execute my operations inside of a large try/except, since they are destructive. it is not cool to try and fail here....
USAGE CONTEXT
-- A "point-lists" is a list-like-thing that contains dict-like-things as its elements.
-- A "matrix" is a list-like-thing that contains list-like-things
-- I have a library of functions that operate on point-lists and also in an analogous way on matrix like things.
-- for example, From the users point of view destructive operations like the "spreadsheet-like" operations "column-slice" can operate on both matrix objects and also on point-list objects in an analogous way -- the resulting thing is like the original one, but only has the specified columns.
-- since this particular operation is destructive it would not be cool to proceed as if an object were a matrix, only to find out part way thru the operation, it was really a point-list or none-of-the-above.
-- I want my 'is_matrix' and 'is_point_list' tests to be performant, since they sometimes occur inside inner loops. So I would be satisfied with a test which only investigated element zero for example.
-- I would prefer tests that do not involve construction of temporary objects, just to determine an object's type, but maybe that is not the python way.
in general I find the whole duck typing thing to be kinda messy, and fraught with bugs and slowness, but maybe I dont yet think like a true Pythonista
happy to drink more kool-aid...
One thing you can do, that should work quickly on a normal list and fail on a normal dict, is taking a zero-length slice from the front:
try:
thing[:0]
except TypeError:
# probably not list-like
else:
# probably list-like
The slice fails on dicts because slices are not hashable.
However, str and unicode also pass this test, and you mention that you are doing destructive edits. That means you probably also want to check for __delitem__ and __setitem__:
def supports_slices_and_editing(thing):
if hasattr(thing, '__setitem__') and hasattr(thing, '__delitem__'):
try:
thing[:0]
return True
except TypeError:
pass
return False
I suggest you organize the requirements you have for your input, and the range of possible inputs you want your function to handle, more explicitly than you have so far in your question. If you really just wanted to handle lists and dicts, you'd be using isinstance, right? Maybe what your method does could only ever delete items, or only ever replace items, so you don't need to check for the other capability. Document these requirements for future reference.
When dealing with built-in types, you can use the Abstract Base Classes. In your case, you may want to test against collections.Sequence or collections.MutableSequence:
if isinstance(your_thing, collections.Sequence):
# access your_thing as a list
This is supported in all Python versions after (and including) 2.6.
If you are using your own classes to build your_thing, I'd recommend that you inherit from these abstract base classes as well (directly or indirectly). This way, you can ensure that the sequence interface is implemented correctly, and avoid all the typing mess.
And for third-party libraries, there's no simple way to check for a sequence interface, if the third-party classes didn't inherit from the built-in types or abstract classes. In this case you'll have to check for every interface that you're going to use, and only those you use. For example, your list_ish function used __len__ and __getitem__, so only check whether these two methods exist. A wrong behavior of __getitem__ (e.g. a dict) should raise an exception.
Perhaps their is no ideal pythonic answer here, so I am proposing a 'hack' solution, but don't know enough about the class structure of python to know if I am getting this right:
def is_list_like(thing):
return hasattr(thing, '__setslice__')
def is_dict_like(thing):
return hasattr(thing, 'keys')
My reduce goals here are to simply have performant tests that will:
(1) never call a dict-thing, nor a string-like-thing a list List item
(2) returns the right answer for python types
(3) will return the right answer if someone implement a "full" set of core method for a list/dict
(4) is fast (ideally does not allocate objects during the test)
EDIT: Incorporated ideas from #DanGetz
I have a bunch of File objects, and a bunch of Folder objects. Each folder has a list of files. Now, sometimes I'd like to lookup which folder a certain file is in. I don't want to traverse over all folders and files, so I create a lookup dict file -> folder.
folder = Folder()
myfile = File()
folder_lookup = {}
# This is pseudocode, I don't actually reach into the Folder
# object, but have an appropriate method
folder.files.append(myfile)
folder_lookup[myfile] = folder
Now, the problem is, the files are mutable objects. My application is built around the fact. I change properites on them, and the GUI is notified and updated accordingly. Of course you can't put mutable objects in dicts. So what I tried first is to generate a hash based on the current content, basically:
def __hash__(self):
return hash((self.title, ...))
This didn't work of course, because when the object's contents changed its hash (and thus its identity) changed, and everything got messed up. What I need is an object that keeps its identity, although its contents change. I tried various things, like making __hash__ return id(self), overriding __eq__, and so on, but never found a satisfying solution. One complication is that the whole construction should be pickelable, so that means I'd have to store id on creation, since it could change when pickling, I guess.
So I basically want to use the identity of an object (not its state) to quickly look up data related to the object. I've actually found a really nice pythonic workaround for my problem, which I might post shortly, but I'd like to see if someone else comes up with a solution.
I felt dirty writing this. Just put folder as an attribute on the file.
class dodgy(list):
def __init__(self, title):
self.title = title
super(list, self).__init__()
self.store = type("store", (object,), {"blanket" : self})
def __hash__(self):
return hash(self.store)
innocent_d = {}
dodge_1 = dodgy("dodge_1")
dodge_2 = dodgy("dodge_2")
innocent_d[dodge_1] = dodge_1.title
innocent_d[dodge_2] = dodge_2.title
print innocent_d[dodge_1]
dodge_1.extend(range(5))
dodge_1.title = "oh no"
print innocent_d[dodge_1]
OK, everybody noticed the extremely obvious workaround (that took my some days to come up with), just put an attribute on File that tells you which folder it is in. (Don't worry, that is also what I did.)
But, it turns out that I was working under wrong assumptions. You are not supposed to use mutable objects as keys, but that doesn't mean you can't (diabolic laughter)! The default implementation of __hash__ returns a unique value, probably derived from the object's address, that remains constant in time. And the default __eq__ follows the same notion of object identity.
So you can put mutable objects in a dict, and they work as expected (if you expect equality based on instance, not on value).
See also: I'm able to use a mutable object as a dictionary key in python. Is this not disallowed?
I was having problems because I was pickling/unpickling the objects, which of course changed the hashes. One could generate a unique ID in the constructor, and use that for equality and deriving a hash to overcome this.
(For the curious, as to why such a "lookup based on instance identity" dict might be neccessary: I've been experimenting with a kind of "object database". You have pure python objects, put them in lists/containers, and can define indexes on attributes for faster lookup, complex queries and so on. For foreign keys (1:n relationships) I can just use containers, but for the backlink I have to come up with something clever if I don't want to modify the objects on the n side.)
I have a class holding a table (list of lists). This class should return a rowpointer similar to sql. For this row pointer I would like to week ref the table row (a list) with a weakref.proxy. However, I would like to add additional capabilities to a row pointer, e.g. overwrite the __getitem__ method to allow access via, say the column names.
Is there an easy way to get the same behaviour (translating access to my object to the object beeing referenced), or do I have to reimplement all the special methods?
As an easy way I could think of inheritance (but since I found no doc on weakref.ProxyType I wont even try to inherit from that, (how to init?). The other option could be to define some special method even to always redirect "special" (__xxx__) function calls to the referred object, even though this makes that seem impossible.
Iresearched some more and found out this:
http://code.activestate.com/recipes/496741-object-proxying/
http://pypi.python.org/pypi/ProxyTypes
So in short, one can forward all calls (I think the recipi on active state is better), but I have not found a way to implement:
$a = proxy([1,2,3])
$b = a
$print type(b)
>>list
I will settle for just working with an object wich pretty much behaves like the list.
Is it common in Python to keep testing for type values when working in a OOP fashion?
class Foo():
def __init__(self,barObject):
self.bar = setBarObject(barObject)
def setBarObject(barObject);
if (isInstance(barObject,Bar):
self.bar = barObject
else:
# throw exception, log, etc.
class Bar():
pass
Or I can use a more loose approach, like:
class Foo():
def __init__(self,barObject):
self.bar = barObject
class Bar():
pass
Nope, in fact it's overwhelmingly common not to test for type values, as in your second approach. The idea is that a client of your code (i.e. some other programmer who uses your class) should be able to pass any kind of object that has all the appropriate methods or properties. If it doesn't happen to be an instance of some particular class, that's fine; your code never needs to know the difference. This is called duck typing, because of the adage "If it quacks like a duck and flies like a duck, it might as well be a duck" (well, that's not the actual adage but I got the gist of it I think)
One place you'll see this a lot is in the standard library, with any functions that handle file input or output. Instead of requiring an actual file object, they'll take anything that implements the read() or readline() method (depending on the function), or write() for writing. In fact you'll often see this in the documentation, e.g. with tokenize.generate_tokens, which I just happened to be looking at earlier today:
The generate_tokens() generator requires one argument, readline, which must be a callable object which provides the same interface as the readline() method of built-in file objects (see section File Objects). Each call to the function should return one line of input as a string.
This allows you to use a StringIO object (like an in-memory file), or something wackier like a dialog box, in place of a real file.
In your own code, just access whatever properties of an object you need, and if it's the wrong kind of object, one of the properties you need won't be there and it'll throw an exception.
I think that it's good practice to check input for type. It's reasonable to assume that if you asked a user to give one data type they might give you another, so you should code to defend against this.
However, it seems like a waste of time (both writing and running the program) to check the type of input that the program generates independent of input. As in a strongly-typed language, checking type isn't important to defend against programmer error.
So basically, check input but nothing else so that code can run smoothly and users don't have to wonder why they got an exception rather than a result.
If your alternative to the type check is an else containing exception handling, then you should really consider duck typing one tier up, supporting as many objects with the methods you require from the input, and working inside a try.
You can then except (and except as specifically as possible) that.
The final result wouldn't be unlike what you have there, but a lot more versatile and Pythonic.
Everything else that needed to be said about the actual question, whether it's common/good practice or not, I think has been answered excellently by David's.
I agree with some of the above answers, in that I generally never check for type from one function to another.
However, as someone else mentioned, anything accepted from a user should be checked, and for things like this I use regular expressions. The nice thing about using regular expressions to validate user input is that not only can you verify that the data is in the correct format, but you can parse the input into a more convenient form, like a string into a dictionary.