Ruby plus equals (+=) versus append/concatenation shovel (<<)

Alec Jacobson

September 08, 2010

weblog/

I was stunned to watch how slow a recent ruby program was. All it was doing was concatenating a bunch of string literals in a big loop. Originally I was using plus equals:

str = ""
1000.times do |i|
  str += "foo bar"
end

On a whim I tried switching to using an array then joining:

str = ""
str_array = []
1000.times do |i|
  str_array << "foo bar"
end
str = str_array.join

Already this was way faster. I wrote up a little benchmarking program to see just how badly "+=" performs compared to "<<". I compare string +=, to the array set up I have above, and just using "<<" on the string:

str = ""
1000.times do |i|
  str << "foo bar"
end

Here's my little test program.

#!/usr/bin/ruby

power = 20

power.times do |p|
  n = 2**p
  str = ""
  start_time = Time.now
  n.times do |i|
    str += "x"
  end
  duration = Time.now - start_time
  #puts "#{n} string appends took: #{duration}s"
  puts "#{n} #{duration}"
end

power.times do |p|
  n = 2**p
  str3 = ""
  start_time = Time.now
  n.times do |i|
    str3 << "x"
  end
  duration = Time.now - start_time
  puts "#{n} #{duration}"
end

power.times do |p|
  n = 2**p
  str2 = ""
  start_time = Time.now
  str_array = []
  n.times do |i|
    str_array << "x"
  end
  str2 = str_array.join
  duration = Time.now - start_time
  puts "#{n} #{duration}"
end

And here are the results:

String += is asymptotically worse than <<. Reading through the ruby doc on strings its clear this is because:

str1 += str2

is syntactic sugar for something like

str1 = str1 + str2

whose "=" creates a new string object, hence the big computational cost. But why?! I can't think of any reason why "+=" shouldn't be syntactic sugar for "<<". Can you? Update: I get it! Here's two short snippets that illustrate the difference:

a = "x"
b = a
b += "y"
a

Which results in "x"

a = "x"
b = a
b << "y"
a

Which results in "xy" It's subjective whether x+=y should mean "append y to x" or always be syntactic sugar for "x = x + y". My vote is for the later, which means I must be content that in Ruby these operators do different things and thus have different speeds.