views:

550

answers:

5

I currently toying with Python 3.1 and I need a deep update function for dictionaries (a function that willl recursively update child dictionaries that are inside a parent dictionary).

But I think, in the future, my function could have to deal with objects that behave like dictionaries but aren't. And furthermore isinstance and type are bad, as I can read on almost every Pythonista's blog.

But duck typing is part of Python's philosophy right? So how could I check if an object is dictionary-like? This way my function will be (I hope) Pythonic and quite powerful and will fit exactly my needs :o).

Note that even if I can find already coded deep update functions with Google, I believe coding my own is a part of my learning (and my first attempt to write true Python code).

Thanks!

PS - Feel free to tell me how I'm wrong on any point, I believe that's too a part of my learning!

Edit : Thank all for the answers. Just in case, the function I coded can be found at this place : http://blog.cafeaumiel.com/public/python/dict_deep_update.py

A: 

use isinstance, there is nothing wrong with it and it's routinely used in code requiring recursion.

If by dictionary-like you mean the object's class inherit from the dict, isinstance will also return True.

>>> class A(dict):
    pass

>>> a = A()
>>> isinstance(a, dict)
True
SilentGhost
The fac is I don't want to code against the *dict* class but against its interface :)
thomas
+1  A: 

Duck typing is where you do what you want to do, and deal with the fallout if your objects don't behave the way you expected them to.

You want to check if something is dict-like, and update it if it is, right? Just call the object's update method, and handle the Exception you get if there is no such method.

Of course, this will fall flat if you deal with custom class objects which have update methods that do something else entirely -- I'm not quite sure how to deal with that.

sykora
I don't use the update method from the *dict* class. instead I manually walk the dictionary and when a value is a dictionary like object, I try to call recursively my function in order to update it.
thomas
+5  A: 

Check out the (new in 3.x) abstract base classes (ABC's) in the collections module:

http://docs.python.org/3.1/library/collections.html

I would consider checking with isinstance against Mapping like in the following:

>>> import collections
>>> isinstance({},collections.Mapping)
True

Then, if you make your own dict-like class, make collections.Mapping one of its bases.

The other route is trying and catching whatever exceptions for the dictionary operations, - but with the recursion you're talking about, I'd rather check against the base type first than handle figuring out what dict or subdict or other dict member was or was not there to throw an exception.

Editing to add: The benefit of checking against the Mapping ABC instead of against dict is that the same test will work for dict-like classes regardless of whether or not they subclass dict, so it's more flexible, just in case.

Anon
Absolutely right: ABCs are there to let you "program against an interface, not an implementation", just as the Gof4 urges you to!-)
Alex Martelli
your way doing that sound good. But, what are Gof4 ??
thomas
Anon
Yep, and btw I intensely recommend that book even though my colleague Joe Gregorio disagrees just as intensely (he claims there are no DPs in Python -- heh -- hope we can get a debate on that issue at some future meeting of our local Python user group!-).
Alex Martelli
A: 

If you also have to handle custom classes that behave like dictionaries but aren't subclassed from dict, you can use getattr to get the function you require.

# loop
# ... more code
someObject = getObject(..)
try:
    getattr(someObject, "update")("someValue")
except AttributeError:
    # Wasn't a type we can handle
Josh Smeaton
+1  A: 

Well, to update recursively a dictionary, you must call the 'items' method to iterate on it.

So I suggest, you just do :

try :
    for key, value in data.items() :
         # recursive call
except AttributeError :
    # handling the trouble
e-satis