I always was curious on how effective is visitor pattern vs instanceof when it comes to detect whether the given base implements certain interface or not.
I had 7 interface classes and visitor for all of these, and 7 concrete implementations, then I tested instanceof and visitor pattern on a large collection of randomly picked instances of these classes, the timing for 100000 objects was as follows on my machine (best value is taken for several interations of both benchmarks, the less is of course the better):
instanceof: 5.055 sec visitor: 3.511 sec
The testing code snippets are as follows:
// for instanceof:
public static int measureInstanceOf(Node[] nodes) {
int litCount = 0;
for (int i = 0; i < N; ++i) {
if (nodes[i] instanceof Literal) {
++litCount;
}
}
return litCount;
}
// for visitor:
public static final class LiteralCheckerVisitor implements NodeVisitor {
boolean isLiteral = false;
public void visitMarker(Marker marker) {}
// other methods are similar to the given above and skipped
public void visitLiteral(Literal literal) { isLiteral = true; }
}
public static int measureVisitor(Node[] nodes) {
int litCount = 0;
final LiteralCheckerVisitor visitor = new LiteralCheckerVisitor();
for (int i = 0; i < N; ++i) {
// equivalent of instanceof Literal
visitor.isLiteral = false;
nodes[i].accept(visitor);
if (visitor.isLiteral) {
++litCount;
}
}
return litCount;
}
As for java version:
java version "1.6.0_24" Java(TM) SE Runtime Environment (build 1.6.0_24-b07-334-10M3326) Java HotSpot(TM) 64-Bit Server VM (build 19.1-b02-334, mixed mode)
The resolution is quite simple - try to avoid instanceof, at least in time-critical code. It's usage often indicates a misunderstanding of the OOP principles and often may result in bad design (and bad performance as well as higher maintenance costs).
Posts