Question

I have a problem where I need a .NET dictionary that supports multiple items per key. In the past I’ve used the STL multimap in my C++ programs. How does the design of a multimap differ from a dictionary of lists i.e. performance, size, etc. (excluding generics vs. templates)?

Answer 1

multimap.count: O(log n + m) where n is number of keys and m is number of items associated with a given key.

For a Dictionary<TKey, List<TValue>> the equivalent functionality would be:

int count = dictionary[key].Count;

And safer is to say

int count;
List<TValue> list;
if(dictionary.TryGetValue(key, out list)) {
    int count = list.Count;
}

This is an O(1) operation because lookup is O(1)¹ and List<T>.Count is O(1).

multimap.find: O(log n) where n is number of keys

For a Dictionary<TKey, List<TValue>> the equivalent functionality would be:

List<TValue> elements = dictionary[key];

And safer is to say

List<TValue> list;
if(dictionary.TryGetValue(key, out list)) {
    // safe to iterate list
}

This is O(1). See the previous remark on lookup by key in a Dictionary<TKey, TValue>.

multimap.insert: O(log n) where n is the number of keys.

For a Dictionary<TKey, List<TValue>> the equivalent functionality would be:

// value is TValue to insert
List<TValue> list;
if(!dictionary.TryGetValue(key, out list)) {
    list = new List<TValue>();
    dictionary.Add(key, list);
}
list.Add(value);

This is usually O(1) but can be O(n) when the capacity of the dictionary must be increased to accomodate the new element.

multimap.remove: There are three overloads of this method; I will only consider the one that accepts a key and removes all occurrences of that key from the multimap. This is an O(log n + m) operation where there n keys and m objects associate with a given key.

For a Dictionary<TKey, List<TValue>> the equivalent functionality would be:

 dictionary.Remove(key);

From the documentation: “This method approaches an O(1) operation.” Same comment applies.

¹: From the documentation: “Retrieving a value by using its key is very fast, close to O(1).” Why the documentation is vague on this point is confusing to me. Either an operation is O(1) or it isn’t. There is no such thing as “close” to O(1).