Le 03/11/2009 03:29, Martin Buchholz a écrit :
I would like to contribute an improvement to LinkedList
webrev:
http://cr.openjdk.java.net/~martin/webrevs/openjdk7/LinkedList/
Josh, Bill and I have discussed this offline,
and they have given their blessing.
(i.e. I am not asking for reviewers)
I apologize for this - there are so many more important things to work on,
and no one is really using LinkedList
in performance-sensitive production code, eh?
Chris, could you file a bug?
Synopsis: LinkedList facelift
Description:
- A list of size N should create only N+1 objects, not N+2.
- Use of null as sentinel value instead of a sentinel (Null Object)
is slightly less convenient, but is measurably faster, since hotspot
has to insert null checks (with NPE throws) anyways.
- generify
- warning removal
- gratuitous tests added.
- cosmetic doc improvements
- the original motivation for this was to have a O(1) clear,
but we eventually decided against that
- but we documented our decision.
Below is a stupid microbenchmark demonstrating app. a factor of 2 improvement in
"throughput", and a 50% increase in memory footprint.
public class LinkedListBench {
public static void main(String[] args) {
LinkedList spine = new LinkedList();
long i = 0;
long t0 = System.nanoTime();
try {
for (i = 0; ; i++) {
spine.add(new LinkedList());
}
} catch (OutOfMemoryError e) {
spine.clear();
System.out.printf("allocated = %d%n", i);
System.out.printf("ns/object = %d%n", (System.nanoTime()-t0)/i);
}
}
}
Thanks,
Martin
Hi Martin,
Your patch can break backward compat.
add(E) now delegates to a public method (an overridable one)
(addLast). So If I have a class that inherits from LinkedList
and overrides addLast() the current semantics of add(E) is not altered,
with your patch, add(E) semantics is also changed.
I think the code of addLast() should be in a private method and
addLast() and add(E) delegate to it.
Rémi
