Benchmarking using the criterion library

Profiling aside, benchmarking the time a calculation takes to perform is a direct indicator of real-world performance. Benchmarking Haskell applications is pretty much dominated by the criterion library. There is a system called nofib, which is used to benchmark GHC itself, but for applications criterion is superior. Criterion even produces interactive web pages describing the results of benchmarks, which is a nice feature.

This text is written for criterion-1.1.1.0. Obviously, the criterion package needs to be installed:

cabal install criterion       # or: stack install criterion

A criterion benchmark suite is created as a normal Haskell program. An example is this:

–– file: benchmark.hs

import Criterion.Main
import Data.List (foldl')

main = defaultMain [
  bgroup "sum" [ bench "sum"    $ whnf sum            [1..1000000]
    , bench "foldr"  $ whnf (foldr (+) 0)  [1..1000000]
    , bench "foldl...