RFC: Added DefaultDict

[kmsquire]

This is a DefaultDict implementation which mimics the Python implementation of the same.

Need to add docs and tests, but the basics are here.

Now includes docs and tests.

Some examples (from my updates to the README):

## Constructors
DefaultDict(default, d::Associative=Dict())  # create a DefaultDict with a default value or function,
                                             # optionally wrapping an existing dictionary
DefaultDict(KeyType, ValueType, default)     # create a DefaultDict with Dict type (KeyType,ValueType)

# Examples
dd = DefaultDict(1)               # create an (Any=>Any) DefaultDict with a default value of 1
dd = DefaultDict(String, Int, 0)  # create a (String=>Int) DefaultDict with a default value of 0

d = ['a'=>1, 'b'=>2]
dd = DefaultDict(0, d)            # provide a default value to an existing dictionary
dd['c'] == 0                      # true
d['c']  == 0                      # true

dd = DefaultDict(time)            # call time() to provide the default value
dd = DefaultDict(Dict)            # Create a dictionary of dictionaries
                                  # Dict() is called to provide the default value
dd = DefaultDict(()->myfunc())    # call function myfunc to provide the default value

# create a Dictionary of String=>DefaultDict{String, Int}, where the default of the
# inner set of DefaultDicts is zero
dd = DefaultDict(String, DefaultDict, ()->DefaultDict(String,Int,0))

I could make a separate package if that is preferable, but this does seem to at least go with the Accumulator/Counter type (and in fact, could provide an alternate implementation of those). The coding/writing style might be slightly different, so I'm open to any thought/suggestions.

[kmsquire]

@lindahua, I've updated this request, and I think it's complete. Let me know what you think.

[stevengj]

Why not simply use getindex(d::DefaultDict, key) = get(d, key, d.default)? Isn't this the whole point of get?

[kmsquire]

As of right now, get doesn't modify the dictionary if a key doesn't exist, but getindex does.

I'm mimicking Python here. I think that choosing whether or not get modifies the dictionary is arbitrary, and I couldn't think of a good argument against the python implementation. Well, other than the fact that getindex isn't as efficient... Hmm.

[kmsquire]

Actually, unless we don't modify the dictionary at all, simply calling get wouldn't change much, as the logic in this function would then move to get. The main thing getting in the way of simplifying all of this is that we only want to call d.default() when the value is not in the dictionary.

Abstractly, what would be useful is a function which, e.g, returns a token that indicates whether or not a key is in the dictionary, as well as indicating (to the dictionary) the location of the key if it exists, or where it would go if it doesn't exist. Making such a function public would allow hash values to be returned to the caller.

[lindahua]

@kmsquire This looks great to me.

I granted the commit privilege to you. Please merge it when you think it is ready.

[kmsquire]

Thanks, Dahua. I think it's good to go for now. Cheers!