Performance drop in 0.2.6

[janlelis]

First of all, thank you a lot for this great gem. Being inspired, I build a very similar one, with some differences in syntax and features. While doing so, I noticed a huge performance drop in the latest version of rubype.

Here is a benchmark using rubype 0.2.5:

ruby version: 2.2.1
sig version: 1.0.0
rubype version: 0.2.5
contracts version: 0.8
Calculating -------------------------------------
                pure    57.194k i/100ms
                 sig     7.816k i/100ms
              rubype     8.756k i/100ms
           contracts     5.771k i/100ms
-------------------------------------------------
                pure      2.855M (± 1.1%) i/s -     14.299M
                 sig    102.990k (± 2.5%) i/s -    515.856k
              rubype    116.062k (± 1.0%) i/s -    586.652k
           contracts     70.575k (± 1.7%) i/s -    357.802k

Comparison:
                pure:  2854866.4 i/s
              rubype:   116062.3 i/s - 24.60x slower
                 sig:   102990.1 i/s - 27.72x slower
           contracts:    70575.3 i/s - 40.45x slower

And here one using rubype 0.2.6:

ruby version: 2.2.1
sig version: 1.0.0
rubype version: 0.2.6
contracts version: 0.8
Calculating -------------------------------------
                pure    53.905k i/100ms
                 sig     7.808k i/100ms
              rubype   417.000  i/100ms
           contracts     5.699k i/100ms
-------------------------------------------------
                pure      2.956M (± 2.4%) i/s -     14.770M
                 sig    105.957k (± 0.5%) i/s -    530.944k
              rubype      4.313k (± 5.3%) i/s -     21.684k
           contracts     69.450k (± 2.3%) i/s -    347.639k

Comparison:
                pure:  2956225.5 i/s
                 sig:   105956.8 i/s - 27.90x slower
           contracts:    69450.1 i/s - 42.57x slower
              rubype:     4312.8 i/s - 685.45x slower

The code used for the benchmark: https://github.com/janlelis/sig/blob/v1.0.0/Rakefile#L32-L131

[jonatack]

Interesting 👍

You might consider using the times form for benchmark-ips, shown below (reproduced from the readme). The results when operations are small are often different:

  # To reduce overhead, the number of iterations is passed in
  # and the block must run the code the specific number of times.
  # Used for when the workload is very small and any overhead
  # introduces incorrectable errors.

  x.report("addition2") do |times|
    i = 0
    while i < times
      1 + 2
      i += 1
    end
  end

[gogotanaka]

@janlelis hi! Looking at your great gem sig, I'm inspired as well. Sig has many interesting features or implementations which Rubype doesn't have.
I'm gonna steal some of them ;)

Aside from that, thank for letting me know about this issue, I found this line causes performance drop.
https://github.com/gogotanaka/Rubype/blob/develop/lib/rubype.rb#L14

By assigning result of Kernel#caller to a variable, made it twice faster but it's still slow...