String#split, String#scan, and String#each_byte which is faster?

irb(main):026:0> Benchmark.bm do |bm|
irb(main):027:1* bm.report("split:") {
10000.times do a = "1234567890".split('') end }
irb(main):028:1> bm.report(" scan:") {
10000.times do a = "1234567890".scan(/./) end }
irb(main):029:1> bm.report("   eb:") {
10000.times do "1234567890".each_byte { |by| (a ||= []) << by } end }
irb(main):030:1> end
     user     system      total        real
split:  0.320000   0.000000   0.320000 (  0.321568)
scan:  0.200000   0.000000   0.200000 (  0.210951)
  eb:  0.260000   0.030000   0.290000 (  0.345428)

So, I am surprised that scan was faster, did you guess that? I wonder if pre-compiling the regex will make it even faster?

irb(main):033:0> Benchmark.bm do |bm|
irb(main):034:1* bm.report("split:") {
10000.times do a = "1234567890".split('') end }
irb(main):035:1> bm.report(" scan:") {
10000.times do a = "1234567890".scan(rx) end }
irb(main):036:1> bm.report("   eb:") {
10000.times do "1234567890".each_byte { |by| (a ||= []) << by } end }
irb(main):037:1> end
     user     system      total        real
split:  0.280000   0.010000   0.290000 (  0.292449)
scan:  0.180000   0.000000   0.180000 (  0.180988)
  eb:  0.280000   0.050000   0.330000 (  0.367461)

Interesting, I may need to dig deep into the ruby core and see what makes the speed difference. I believe split and scan are implemented as native code. I can understand why each_byte is the slowest given the loop complexity. More may follow as I dig in just for fun and the learning value.