Not the Fruity logo, but I like grapefruit

Another tech review, this time a Ruby library. Fruity is a library for performance testing. You have probably heard of its better-known cousin, Benchmark.

When I’m interested in benchmarking, it’s usually for small methods. I’ll think of a different way to approach a problem, and then the thought will occur to me, “But is it performant?” Fruity is ideal for these kind of situations. But why might I want to import a library when Benchmark is built in?

Let’s take a recent example. I was working through the Minesweeper exercise on Exercism.io. I was thinking if there was a better way to make sure my Minesweeper grid didn’t have any illegal characters in it. The first way was to convert an array of characters into a string, then to use count to make sure the count of all unwanted characters was zero.

After I finished the exercise, I thought about another way. The Array class has an intersection operator &. If I used that, I wouldn’t have to convert the array to a string and maybe save some cycles.

When I’m wanting a Ruby sandbox, I turn to Pry. I don’t use most of Pry’s features, but I like it over IRB for its better syntax highlighting.

First, some test data:

[9] pry(main)> xyzzy = ['x','y','z','z','y']
=> ["x", "y", "z", "z", "y"]
[10] pry(main)> good = ['+','|','*','1','2','3','4',' ','-']
=> ["+", "|", "*", "1", "2", "3", "4", " ", "-"]

Here’s a performance comparison using Benchmark:

[15] pry(main)> n = 100000
=> 100000
[16] pry(main)> Benchmark.bm do |x|
[16] pry(main)*   x.report { xyzzy.join.count('^+|*1234 -') == 0 }
[16] pry(main)*   x.report { xyzzy & good == xyzzy }
[16] pry(main)* end
       user     system      total        real
   0.000000   0.000000   0.000000 (  0.000007)
   0.000000   0.000000   0.000000 (  0.000012)
=> [#<Benchmark::Tms:0x007fff0326ce38
  @cstime=0.0,
  @cutime=0.0,
  @label="",
  @real=7.4260024121031165e-06,
  @stime=0.0,
  @total=0.0,
  @utime=0.0>,
 #<Benchmark::Tms:0x007fff0326c4d8
  @cstime=0.0,
  @cutime=0.0,
  @label="",
  @real=1.1732001439668238e-05,
  @stime=0.0,
  @total=0.0,
  @utime=0.0>]

Note that this output is a little cryptic. It does tell me that the first way is faster, but I need to do a little work. I had to think about the number of iterations I wanted for the test. And, I had to do some mental math to figure out the real answer I wanted, which is “which one is faster and by how much?”

Now, let’s have a look at Fruity:

[11] pry(main)> compare do
[11] pry(main)*   count { xyzzy.join.count('^+|*1234 -') == 0 }
[11] pry(main)*   intersect { xyzzy & good == xyzzy }
[11] pry(main)* end
Running each test 2048 times. Test will take about 1 second.
count is faster than intersect by 3x ± 0.1

Notice a few things about Fruity:

  • There’s less boilerplate code
  • I was able to give each algorithm a name
  • I didn’t have to specify the number of iterations. Fruity came up with an intelligent guess.
  • Fruity reported its results in relative terms

A simple test like this highlights a different in approach between Fruity and other benchmarking tools. For more on that, take a look at the project’s README on Github. In short, Fruity seeks to eliminate noise by not reporting insignificant differences.

Fruity can do more than just anonymous methods, as I’ve shown here. It can compare methods on a class or another kind of callable. It can also do comparisons with parameters.

In cases like mine, clarity trumps precision. It’s the difference between asking LaForge and Data a question. I’d prefer some synthesis over a data dump (no pun intended). Next time you’re wondering about performance in Ruby, give Fruity a try.