Area#add(Area) Optimization
Philip Race
philip.race at oracle.com
Sat Feb 12 22:26:37 UTC 2022
I suppose this could be added but it is something an application also
can easily do for itself.
> curves = rhs.curves;
The straight reference, rather than a copy surprised me, but I suppose
(I am not familiar
with the working of the Area class) that all mutations must create a new
Vector ...
-phil.
On 2/11/22 3:51 PM, Jeremy Wood wrote:
> I’m working a lot with the Area class lately. Is there any interest in
> the following optimization?
>
> Area#add(Area) currently resembles:
>
> public void add(Area rhs) {
> curves = new AreaOp.AddOp().calculate(this.curves, rhs.curves);
> invalidateBounds();
> }
>
> I propose replacing it with:
>
> public void add(Area rhs) {
> if (isEmpty()) {
> curves = rhs.curves;
> } else if (rhs.isEmpty()) {
> return;
> } else {
> curves = new AreaOp.AddOp().calculate(this.curves, rhs.curves);
> }
> invalidateBounds();
> }
>
> If either operand is empty then this is effectively a null op. But the
> current implementation may still take several seconds to return
> depending on the complexity of the non-empty operand.
>
> For example the following two snippets of code may have significantly
> different execution times:
>
> Area sum = new Area();
> for(Shape shape : incomingShapes) {
> sum.add(new Area(shape));
> }
>
> vs
>
> Area sum = new Area(shapes.get(0));
> for(int a = 1; a < incomingShapes.size(); a++) {
> Shape shape = incomingShapes.get(a);
> sum.add(new Area(shape));
> }
>
> And similarly we could review the other operations
> (subtract/intersect/exclusiveOr/transform) for obvious potential null
> ops.
>
> If I understand openjdk rules correctly: I'm not authorized to submit
> an openJDK ticket. Does this interest anyone else on this list enough
> to submit this as an openJDK ticket? (I'm happy to work on it if I can
> help.)
>
> Regards,
> - Jeremy
More information about the client-libs-dev
mailing list