[Test] defn.count_instances vs defn.instances.lenght

thomthom

I was curious of why there was a method definition.count_instances - as I've often just used definition.instances.length. Did a quick test:

d=Sketchup.active_model.definitions[0]; c=1000; puts "Counting #{d.count_instances} instances..."; t=Time.now; c.times { d.count_instances }; puts "Count: #{Time.now-t}s"; t=Time.now; c.times { d.instances.length }; puts "Count: #{Time.now-t}s";

Counting 8642 instances... Count: 0.0s Count: 1.748s

Significant difference.
In order to get .count_instances to be of notisable lag I needed to go over 1,000,000 iterations ( 0.38s). With 10,000,000 it took 3.678s. I did not even try to test with definition.instances.length with that many iterations.

Dan Rathbun

Reason: The count_instances method queries the C++ side for an integer.

Each time you call the instances method, the C side API has to allocate a Ruby Array object which has as it's members, pointers to all the definition's C++ instances collection.

Your test might get closer if you have a statement like i=defn.instances outside the loop. Ie, create the Array only once, then have i.length inside the loop's block.

If you do not need a Ruby Array, then count_instances should always be faster.

This is a candidate for the Optimization topic.

thomthom

Duh! Should off course have eliminated new object creations. (They are all too expensive.)

Changing it and the numbers are quite different:
d=Sketchup.active_model.definitions[0]; inst=d.instances; c=10000000; puts "Counting #{d.count_instances} instances..."; t=Time.now; c.times { d.count_instances }; puts "Count: #{Time.now-t}s"; t=Time.now; c.times { inst.length }; puts "Count: #{Time.now-t}s"

Counting 8642 instances... Count: 3.562s Count: 1.795s