HashSets and HashMaps are implemented using a data structure known as a hash table. You don’t need to understand hash tables to use HashSets or HashMaps, but any computer programmer should be familiar with hash tables and how they work.
Hash tables are an elegant solution to the search problem. A hash table, like a HashMap, stores key/value pairs. Given a key, you have to search the table for the corresponding key/value pair. When a hash table is used to implement a set, the values are all null, and the only question is whether or not the key occurs in the set. You still have to search for the key to check whether it is there or not.
In most search algorithms, in order to find the item you are interested in, you have to look through a bunch of other items that don’t interest you. To find something in an unsorted list, you have to go though the items one-by-one until you come to the one you are looking for. In a binary sort tree, you have to start at the root and move down the tree until you find the item you want. When you search for a key/value pair in a hash table, you can go directly to the location that contains the item you want. You don’t have to look through any other items. (This is not quite true, but it’s close.) The location of the key/value pair is computed from the key: You just look at the key, and then you go directly to the location where it is stored.
How can this work? If the keys were integers in the range 0 to 99, we could store the key/value pairs in an array, A, of 100 elements. The key/value pair with keyK would be stored in A[K]. The key takes us directly to the location of the key/value pair. The problem is that there are usually far too many different possible keys for us to be able to use an array with one location for each possible key. For example, if the key can be any value of type int, then we would need an array with over four billion locations — quite a waste of space if we are only going to store, say, a few thousand items! If the key can be a string of any length, then the number of possible keys is infinite, and using an array with one location for each possible key is simply impossible.
Nevertheless, hash tables store their data in an array, and the array index where a key is stored is based on the key. The index is not equal to the key, but it is computed from the key. The array index for a key is called the hash code for that key. A function that computes a hash code, given a key, is called a hash function. To find a key in a hash table, you just have to compute the hash code of the key and go directly to the array location given by that hash code. If the hash code is 17, look in array location number 17.
Now, since there are fewer array locations than there are possible keys, it’s possible that we might try to store two or more keys in the same array location. This is called a collision. A collision is not an error. We can’t reject a key just because another key happened to have the same hash code. A hash table must be able to handle collisions in some reasonable way. In the type of hash table that is used in Java, each array location actually holds a linked list of key/value pairs (possibly an empty list). When two items have the same hash code, they are in the same linked list. The structure of the hash table looks something like this:
In this diagram, there is one item with hash code 0, no items with hash code 1, two items with hash code 2, and so on. In a properly designed hash table, most of the linked lists are of length zero or one, and the average length of the lists is less than one. Although the hash code of a key doesn’t necessarily take you directly to that key, there are probably no more than one or two other items that you have to look through before finding the key you want. For this to work properly, the number of items in the hash table should be somewhat less than the number of locations in the array. In Java’s implementation, whenever the number of items exceeds 75% of the array size, the array is replaced by a larger one and all the items in the old array are inserted into the new one. (This is why adding one new item will sometimes cause the ordering of all the items in the hash table to change completely.)
There is still the question of where hash codes come from. Every object in Java has a hash code. The Object class defines the method hashCode(), which returns a value of type int. When an object, obj, is stored in a hash table that has N locations, a hash code in the range 0 to N-1 is needed. This hash code is computed as Math.abs(obj.hashCode()) % N, the remainder when the absolute value of obj.hashCode() is divided by N. (The Math.abs is necessary because obj.hashCode() can be a negative integer, and we need a non-negative number to use as an array index.)
For hashing to work properly, two objects that are equal according to the equals() method must have the same hash code. In the Object class, this condition is satisfied because both equals() and hashCode() are based on the address of the memory location where the object is stored. However, many classes redefine the equals() method. If a class redefines the equals() method, and if objects of that class will be used as keys in hash tables, then the class should also redefine the hashCode() method. For example, in the String class, the equals() method is redefined so that two objects of type String are considered to be equal if they contain the same sequence of characters. The hashCode() method is also redefined in the String class, so that the hash code of a string is computed from the characters in that string rather than from its location in memory. For Java’s standard classes, you can expect equals() and hashCode() to be correctly defined. However, you might need to define these methods in classes that you write yourself.
Next: Testing Java Programs