GNU Trove: High performance collections for Java.

Objectives

The GNU Trove library has two objectives:

1. Provide "free" (as in "free speech" and "free beer"), fast, lightweight implementations of the java.util Collections API. These implementations are designed to be pluggable replacements for their JDK equivalents.
2. Whenever possible, provide the same collections support for primitive types. This gap in the JDK is often addressed by using the "wrapper" classes (java.lang.Integer, java.lang.Float, etc.) with Object-based collections. For most applications, however, collections which store primitives directly will require less space and yield significant performance gains.

Hashtable techniques

The Trove maps/sets use open addressing instead of the chaining approach taken by the JDK hashtables. This eliminates the need to create Map.Entry wrappper objects for every item in a table and so reduces the O (big-oh) in the performance of the hashtable algorithm. The size of the tables used in Trove's maps/sets is always a prime number, improving the probability of an optimal distribution of entries across the table, and so reducing the the likelihood of performance-degrading collisions. Trove sets are not backed by maps, and so using a THashSet does not result in the allocation of an unused "values" array.

Hashing strategies

Trove's maps/sets support the use of custom hashing strategies, allowing you to tune collections based on characteristics of the input data. This feature also allows you to define hash functions when it is not feasible to override Object.hashCode(). For example, the java.lang.String class is final, and its implementation of hashCode() takes O(n) time to complete. In some applications, however, it may be possible for a custom hashing function to save time by skipping portions of the string that are invariant.

Using java.util.HashMap, it is not possible to use Java language arrays as keys. For example, this code:

    char[] foo, bar;
    foo = new char[] {'a','b','c'};
    bar = new char[] {'a','b','c'};
    System.out.println(foo.hashCode() == bar.hashCode() ? "equal" : "not equal");
    System.out.println(foo.equals(bar) ? "equal" : "not equal");
    

    not equal
    not equal
    

In a gnu.trove.THashMap, however, you can implement a TObjectHashingStrategy to enable hashing on arrays:

    class CharArrayStrategy implements TObjectHashingStrategy {
        public int computeHashCode(Object o) {
            char[] c = (char[])o;
            // use the shift-add-xor class of string hashing functions
            // cf. Ramakrishna and Zobel, "Performance in Practice of String Hashing Functions"
            int h = 31; // seed chosen at random
            for (int i = 0; i < c.length; i++) { // could skip invariants
                h = h ^ ((h << 5) + (h >> 2) + c[i]); // L=5, R=2 works well for ASCII input
            }
            return h;
        }

        public boolean equals(Object o1, Object o2) {
            char[] c1 = (char[])o1;
            char[] c2 = (char[])o2;
            if (c1.length != c2.length) { // could drop this check for fixed-length keys
                return false;
            }
            for (int i = 0, len = c1.length; i < len; i++) { // could skip invariants
                if (c1[i] != c2[i]) {
                    return false;
                }
            }
            return true;
        }
    }
    

Iterators in primitive collections

As of release 0.1.7, Trove's primitive mappings include access through Iterators as well as procedures and functions. The API documentation on those classes contains several examples showing how these can be used effectively and explaining why their semantics differ from those of java.util.Iterator.

Miscellaneous

N.B. using Map.entrySet on a Trove Map is supported, but not encouraged. The reason is that this API requires the creation of the Map.Entry Objects that all other parts of Trove manage to avoid. An alternative is to implement the appropriate Procedure interface and use it to invoke the Map's forEachEntry API. Map.keySet and Map.values are not similarly encumbered; nevertheless, the forEachKey, forEachValue, and transformValues APIs will yield slightly better performance at the cost of compatibility with the interface of java.util.Map.