String#split behaviour with embedded strings.

embedded_strings.rb

s = "One Two Three" # Shorter than 24 characters.

s.split /\s/ # Uncomment and comment. Why does the original string now occur 2 more times (1 per match?) in the ObjectSpace than without splitting?

times = 0
ObjectSpace.each_object(String) do |str|
  times += 1 if str == s
end
p times


s = "One Two Three Four Five" # Shorter than 24 characters.

s.split /\s/ # Uncomment and comment. Why does the original string now occur 4 more times (1 per match?) in the ObjectSpace than without splitting?

times = 0
ObjectSpace.each_object(String) do |str|
  times += 1 if str == s && str.hash == s.hash
end
p times


s = "One Two Three Four Five Six" # Longer than 24 characters.

s.split /\s/ # Uncomment and comment.

times = 0
ObjectSpace.each_object(String) do |str|
  times += 1 if str == s && str.hash == s.hash
end
p times

Running GC.start after the splits cleans up the garbage left behind by #split.

... This modified version behaves deterministically:
https://gist.github.com/kschiess/6316508