Hello, I restarted new benchmarks for the Free mod...

# arrow-contributors

[PhBastiani]

Hello, I restarted new benchmarks for the Free module...
For these tests I configured JMH in SingleShot mode (i.e. without JVM warmup, and with a single measurement)... and, i compared Cats/Arrow implementations !
Each test uses a similar method to calculate the Fibonacci sequence.
For example :

fun trampolineFibonacci(n: Int): TrampolineF<Int> =
    when {
      n < 2 -> Trampoline.done(n)
      else -> {
        TrampolineF.fx {
          val x = Trampoline.defer { trampolineFibonacci(n - 1) }.bind()
          val y = Trampoline.defer { trampolineFibonacci(n - 2) }.bind()
          x + y
        }
      }
    }

Benchmark (Param) Score (ms/op)
arrow.benchmarks.TrampolineBench.eval 30 7933,298
arrow.benchmarks.TrampolineBench.trampoline 30 9285,310
cats.benchmarks.TrampolineBench.eval 30 655,456
cats.benchmarks.TrampolineBench.stdlib (TailCalls) 30 596,109
cats.benchmarks.TrampolineBench.trampoline 30 1567,035

For Cats the Trampoline / Eval ratio is significant. Overall results on the Arrow side do not seem very good 🤔
I didn't do any tests without fx