Cop idea: Native Hash#slice vs other slice implementations before native code

[ydakuka]

https://github.com/fastruby/fast-ruby#native-hashslice-vs-other-slice-implementations-before-native-code

Describe the solution you'd like

HASH = {
  title: "awesome",
  description: "a description",
  author: "styd",
  published_at: Time.now
}

KEYS = %i[title author other].freeze

# bad
def my_method
  HASH.select { |k, _| KEYS.include? k }
end

# bad
def my_method
  memo = {}
  KEYS.each { |k| memo[k] = HASH[k] if HASH.key?(k) }
  memo
end

# bad
def my_method
  KEYS.each_with_object({}){ |k, h| h[k] = HASH[k] if HASH.key?(k) }
end

# good
def my_method
  HASH.slice(*KEYS)
end

Rubocop

ydakuka@yauhenid:~/Work/project$ bin/rails_docker rubocop -V
1.57.2 (using Parser 3.2.2.4, rubocop-ast 1.29.0, running on ruby 2.7.8) [x86_64-linux]
  - rubocop-capybara 2.19.0
  - rubocop-factory_bot 2.24.0
  - rubocop-performance 1.19.1
  - rubocop-rails 2.22.0
  - rubocop-rake 0.6.0
  - rubocop-rspec 2.25.0
  - rubocop-thread_safety 0.5.1

[ydakuka]

I've benchmarked it (ruby 3.3.0).

ydakuka@yauhenid:~/ruby-docker-app$ docker run ruby-app  
ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [x86_64-linux]
Warming up --------------------------------------
Hash#native-slice      342.197k i/100ms
Array#each             180.535k i/100ms
Array#each_w/_object   115.737k i/100ms
Hash#select-include     56.865k i/100ms
Calculating -------------------------------------
Hash#native-slice         3.588M (± 1.5%) i/s -     18.136M in   5.056031s
Array#each                1.681M (± 1.1%) i/s -      8.485M in   5.048040s
Array#each_w/_object      1.336M (± 4.9%) i/s -      6.713M in   5.041132s
Hash#select-include     540.935k (± 4.7%) i/s -      2.730M in   5.057020s

Comparison:
Hash#native-slice   :  3587866.5 i/s
Array#each          :  1681076.6 i/s - 2.13x  slower
Array#each_w/_object:  1335545.9 i/s - 2.69x  slower
Hash#select-include :   540935.0 i/s - 6.63x  slower

[ydakuka]

I've dug into the real-world- repos and found that the pattern (Hash#select-include) is pretty popular.