Typically, a cache should keep objects around some time and shall expose of them some time later. What is a good time to hold an object depends on the use case. I wanted this thing to be simple, no threads or schedulers. This approach works for me. Unlike SoftReferences, objects are guaranteed to be available some minimum amount of time. However, the do not stay around in memory until the sun turns into a red giant.
As useage example think of a slowly responding system that shall be able to check if a request has been done quite recently, and in that case not to perform the requested action twice, even if a hectic user hits the button several times. But, if the same action is requested some time later, it shall be performed again.
class Cache<T> {
long avg, count, created, max, min;
Map<T, Long> map = new HashMap<T, Long>();
/**
* @param min minimal time [ns] to hold an object
* @param max maximal time [ns] to hold an object
*/
Cache(long min, long max) {
created = System.nanoTime();
this.min = min;
this.max = max;
avg = (min + max) / 2;
}
boolean add(T e) {
boolean result = map.put(e, Long.valueOf(System.nanoTime())) != null;
onAccess();
return result;
}
boolean contains(Object o) {
boolean result = map.containsKey(o);
onAccess();
return result;
}
private void onAccess() {
count++;
long now = System.nanoTime();
for (Iterator<Entry<T, Long>> it = map.entrySet().iterator(); it.hasNext();) {
long t = it.next().getValue();
if (now > t + min && (now > t + max || now + (now - created) / count > t + avg)) {
it.remove();
}
}
}
}