Why doesn't Java Map extend Collection?

From the Java Collections API Design FAQ:

Why doesn't Map extend Collection?

This was by design. We feel that mappings are not collections and collections are not mappings. Thus, it makes little sense for Map to extend the Collection interface (or vice versa).

If a Map is a Collection, what are the elements? The only reasonable answer is "Key-value pairs", but this provides a very limited (and not particularly useful) Map abstraction. You can't ask what value a given key maps to, nor can you delete the entry for a given key without knowing what value it maps to.

Collection could be made to extend Map, but this raises the question: what are the keys? There's no really satisfactory answer, and forcing one leads to an unnatural interface.

Maps can be viewed as Collections (of keys, values, or pairs), and this fact is reflected in the three "Collection view operations" on Maps (keySet, entrySet, and values). While it is, in principle, possible to view a List as a Map mapping indices to elements, this has the nasty property that deleting an element from the List changes the Key associated with every element before the deleted element. That's why we don't have a map view operation on Lists.

Update: I think the quote answers most of the questions. It's worth stressing the part about a collection of entries not being a particularly useful abstraction. For example:

Set<Map.Entry<String,String>>

would allow:

set.add(entry("hello", "world"));
set.add(entry("hello", "world 2");

(assuming an entry() method that creates a Map.Entry instance)

Maps require unique keys so this would violate this. Or if you impose unique keys on a Set of entries, it's not really a Set in the general sense. It's a Set with further restrictions.

Arguably you could say the equals()/hashCode() relationship for Map.Entry was purely on the key but even that has issues. More importantly, does it really add any value? You may find this abstraction breaks down once you start looking at the corner cases.

It's worth noting that the HashSet is actually implemented as a HashMap, not the other way around. This is purely an implementation detail but is interesting nonetheless.

The main reason for entrySet() to exist is to simplify traversal so you don't have to traverse the keys and then do a lookup of the key. Don't take it as prima facie evidence that a Map should be a Set of entries (imho).

While you've gotten a number of answers that cover your question fairly directly, I think it might be useful to step back a bit, and look at the question a bit more generally. That is, not to look specifically at how the Java library happens to be written, and look at why it's written that way.

The problem here is that inheritance only models one type of commonality. If you pick out two things that both seem "collection-like", you can probably pick out a 8 or 10 things they have in common. If you pick out a different pair of "collection-like" things, they'll also 8 or 10 things in common -- but they won't be the same 8 or 10 things as the first pair.

If you look at a dozen or so different "collection-like" things, virtually every one of them will probably have something like 8 or 10 characteristics in common with at least one other one -- but if you look at what's shared across every one of them, you're left with practically nothing.

This is a situation that inheritance (especially single inheritance) just doesn't model well. There's no clean dividing line between which of those are really collections and which aren't -- but if you want to define a meaningful Collection class, you're stuck with leaving some of them out. If you leave only a few of them out, your Collection class will only be able to provide quite a sparse interface. If you leave more out, you'll be able to give it a richer interface.

Some also take the option of basically saying: "this type of collection supports operation X, but you're not allowed to use it, by deriving from a base class that defines X, but attempting to use the derived class' X fails (e.g., by throwing an exception).

That still leaves one problem: almost regardless of which you leave out and which you put in, you're going to have to draw a hard line between what classes are in and what are out. No matter where you draw that line, you're going to be left with a clear, rather artificial, division between some things that are quite similar.

I guess the why is subjective.

In C#, I think Dictionary extends or at least implements a collection:

public class Dictionary<TKey, TValue> : IDictionary<TKey, TValue>, 
    ICollection<KeyValuePair<TKey, TValue>>, IEnumerable<KeyValuePair<TKey, TValue>>, 
    IDictionary, ICollection, IEnumerable, ISerializable, IDeserializationCallback

In Pharo Smalltak as well:

Collection subclass: #Set
Set subclass: #Dictionary

But there is an asymmetry with some methods. For instance, collect: will takes association (the equivalent of an entry), while do: take the values. They provide another method keysAndValuesDo: to iterate the dictionary by entry. Add: takes an association, but remove: has been "suppressed":

remove: anObject
self shouldNotImplement 

So it's definitively doable, but leads to some other issues regarding the class hierarchy.

What is better is subjective.