Very interesting, well it’s good to know that JRuby is at least more or less consistent :).

In general in JRuby the difference between merge and merge it’s even larger..
On my machine:
# On Ruby i686-linux 1.9.1 merge! is 42.8203960017727 faster than merge.
# On Ruby java 1.8.7 merge! is 772.361024112707 faster than merge. (first run)
However, using nailgun (http://sourceforge.net/projects/nailgun/):
# On Ruby java 1.8.7 merge! is 92.3685496183206 faster than merge.
# On Ruby java 1.8.7 merge! is 102.961025482799 faster than merge.
# On Ruby java 1.8.7 merge! is 207.121184088807 faster than merge.
# …

Yeah, if you can restrict it to just doing the merge once, that certainly helps a lot, my example was contrived in that I was creating the small hashes, but with my real problem I already had a bunch of little hashes which I wanted to merge into one, so no matter how I spun it I would have to do a lot of merges.

main_hash = {}
time = Benchmark.realtime do
simple_hash = {}
(1..10000).each do |number|
simple_hash[number.to_s] = number
end
main_hash = main_hash.merge(simple_hash)
end
puts “Time elapsed #{time} seconds”
puts “Main hash key count: #{main_hash.keys.count}”

Time elapsed 0.133629083633423 seconds

The difference can be significant. All though, this will be different from Ruby 1.8 and 1.9. Sketchup uses 1.8. If you are creating variables within the loop then the for loop will be faster because the variables are not locale to the loop. Difference in the test was at times double. But if you init the variables used in the each loop then the differences isn’t very noticeable.