Catch throw vs raise rescue vs baseline return (Ruby v3.2.0)

bench.rb

code1 = -> { catch(:foo) { throw(:foo, 'string') } }
code2 = -> { begin; raise RuntimeError; rescue RuntimeError; 'string' end }
code3 = -> { 'string' }

Benchmark.ips do |x|
  x.report("catch a throw") {
    value = code1.call
  }

  x.report("rescue and exception") {
    value = code2.call
  }

  x.report("return string") {
    value = code3.call
  }

  x.compare!
end

# Warming up --------------------------------------
#        catch a throw   347.786k i/100ms
# rescue and exception   135.262k i/100ms
#        return string   795.463k i/100ms
# Calculating -------------------------------------
#        catch a throw      3.465M (± 0.4%) i/s -     17.389M in   5.019335s
# rescue and exception      1.359M (± 2.0%) i/s -      6.898M in   5.077732s
#        return string      7.949M (± 0.4%) i/s -     39.773M in   5.003773s

# Comparison:
#        return string:  7948778.8 i/s
#        catch a throw:  3464519.8 i/s - 2.29x  slower
# rescue and exception:  1359120.9 i/s - 5.85x  slower

I've investigated this and I think there's a flaw in this code. I'm not sure code is being reassigned correctly between tests. raise/rescue should still be 3x slower even in 3.2. 2x slower if you ignore the backtrace. Try my gist where I numbered each code block. Your results seemed unexpected given other recent reporting.

@silent-e Great find. Wow what a weird behavior with variables. Updated the bench.