Bench 1.0.0 released

Bench version 1.0.0 released!
http://bench.rubyforge.org

== DESCRIPTION

Do you remeber how to use the benchmark library from the Ruby standard
lib? I don’t.

Now you need not to remember, there is Bench: A DSL around the benchmark
lib of the Ruby
standard lib with the goal to make benchmarking as easy as possible.

== SYNOPSIS

Adapted example of the benchmark documentation from the pickaxe version
2
page 657

require ‘bench’

string = ‘Stormy Weather’
m = string.method(:length)

benchmark ‘code’ do
m.call
end

benchmark ‘send’ do
string.send(:length)
end

benchmark ‘eval’ do
eval “string.length”
end

run 10_000

== CREDITS

Copyright 2008 by Jan F. ([email protected])

== LICENSE

Ruby’s license.

On Jun 10, 2008, at 12:34 , Jan F. wrote:

Do you remeber how to use the benchmark library from the Ruby standard
lib? I don’t.

Now you need not to remember, there is Bench: A DSL around the
benchmark
lib of the Ruby standard lib with the goal to make benchmarking as
easy as possible.

If I don’t remember how to use the benchmark library, why would I
remember how to use your DSL?

That is what ri is for. ri Benchmark obviates the need for Bench.

On Jun 10, 3:11 pm, Ryan D. [email protected] wrote:

If I don’t remember how to use the benchmark library, why would I
remember how to use your DSL?

That is what ri is for. ri Benchmark obviates the need for Bench.

Agreed. However, I would like a nicer interface:

I’ve submitted “bench-unit” as a RubyForge project.

Regards,

Dan

On Wed, Jun 11, 2008 at 5:04 AM, Jan F.
[email protected] wrote:

Ryan D. wrote:

If I don’t remember how to use the benchmark library, why would I
remember how to use your DSL?
Because it has only 2 commands: benchmark and run. :wink:

Agree :slight_smile:

Good job!

On Jun 10, 3:34 pm, Jan F. [email protected] wrote:

benchmark ‘code’ do

run 10_000

What is the underlying translation of benchmark? I.e. the original
Benchmark library has a few different methods, which did you use? I
like the simplicity of your DSL. In the long run it might be nice to
see this advance beyond a dependency on the original benchmark
library.

T.

Ryan D. wrote:

If I don’t remember how to use the benchmark library, why would I
remember how to use your DSL?
Because it has only 2 commands: benchmark and run. :wink:

That is what ri is for. ri Benchmark obviates the need for Bench.
Maybe, but with bench it’s a pleasure to doing benchmarks interactive
with irb:

require ‘bench’

benchmark ‘simple’ do
?> /ll/ =~ ‘hello world’

end

benchmark ‘freezed’ do
?> /ll/.freeze =~ ‘hello world’

end

run 1000
user system total real
simple 0.000000 0.000000 0.000000 ( 0.003960)
freezed 0.010000 0.000000 0.010000 ( 0.004870)

run 1000
user system total real
simple 0.010000 0.000000 0.010000 ( 0.003969)
freezed 0.000000 0.000000 0.000000 ( 0.004624)

let’s try more iterations

?> run 10000
user system total real
simple 0.060000 0.000000 0.060000 ( 0.058049)
freezed 0.060000 0.000000 0.060000 ( 0.058636)

run 100000
user system total real
simple 0.500000 0.000000 0.500000 ( 0.502427)
freezed 0.540000 0.000000 0.540000 ( 0.533421)

now another benchmark sample

?> RE = /ll/

benchmark ‘constant’ do
?> RE =~ ‘hello world’

end

run
user system total real
simple 0.000000 0.000000 0.000000 ( 0.000031)
freezed 0.000000 0.000000 0.000000 ( 0.000469)
constant 0.000000 0.000000 0.000000 ( 0.000031)

But It’s certainly a question of taste. :slight_smile:

Regards,
Jan F.

Jan F. wrote:

    x.report(bm.name) { count.times {bm.proc.call} }

Suggestion:

       x.report(bm.name) { count.times(&bm.proc) }

This will run faster, and more importantly less of the reported time
will be due to framework overhead.

Trans [email protected] wrote:

What is the underlying translation of benchmark? I.e. the original
Benchmark library has a few different methods, which did you use? I like
the simplicity of your DSL. In the long run it might be nice to see this
advance beyond a dependency on the original benchmark library.
Let’s look at the code:

def run count=1
size = Bench.queue.inject(0) {|max, bm| size = bm.name.size; size >
max ? size : max}
Benchmark.bm(size) do |x|
Bench.queue.each do |bm|
x.report(bm.name) { count.times {bm.proc.call} }
end
end
end

Two methods of the original benchmark are called:

  • Benchmark.bm with an automatic calculated label_width (I don’t want to
    think about that).
  • Benchmark::Report#report

You can call run as often as you want and do GC.start or whatever
before,
so you can simulate the Benchmark.bmbm method. That is enough for my
needs.

Maybe there is a better approach? Feel free to contribute:
http://gitorious.org/projects/bench :slight_smile:

Regards
Jan F.

Joel VanderWerf [email protected] wrote:

       x.report(bm.name) { count.times(&bm.proc) }

This will run faster, and more importantly less of the reported time
will be due to framework overhead.
Thanks a lot. :slight_smile:

Regards
Jan F.