gistfile1.rb

# Compile CityHash
#g++ -shared city.cc -o libcity.dylib
#

# Figure out the symbol names.. this doesn't look right. Probably user error.
# nm -g libcity.dylib 
#
=begin
0000000000000db7 T __Z10CityHash64PKcm
0000000000001800 T __Z11CityHash128PKcm
...
0000000000001a55 T __ZSt9make_pairIyyESt4pairIT_T0_ES1_S2_
000000000000114e T __Z18CityHash64WithSeedPKcmy
000000000000117f T __Z19CityHash128WithSeedPKcmSt4pairIyyE

                 U ___gxx_personality_v0
                 U dyld_stub_binder
=end

module CityHash
  require "rubygems"
  require "ffi"
  extend FFI::Library
  ffi_lib "city"

  attach_function :_Z10CityHash64PKcm,  [:string, :size_t], :uint64
  attach_function :_Z11CityHash128PKcm, [:string, :size_t], :ulong
  attach_function :_Z18CityHash64WithSeedPKcmy, [:string, :size_t, :ulong], :ulong
end

if __FILE__ == $0
  require "benchmark"
  require "zlib"

  str = ''
  # You can preseed the CityHash functions
  seed = 1_100_200_300_400.1

  # Build a large-ish string
  
  100.times{str << (("A".."Z").to_a + (0..9).to_a).flatten.join}
  size = str.size
  puts "=======TO HASH=========="
  puts str
  puts "=======TO HASH=========="

  Benchmark.bmbm do |x|
    x.report("cityhash64")  { 10_000.times{CityHash._Z10CityHash64PKcm(str, size)} }
    x.report("cityhash128") { 10_000.times{CityHash._Z11CityHash128PKcm(str, size)} }
    x.report("cityhash64 (w/ seed)") { 10_000.times{CityHash._Z18CityHash64WithSeedPKcmy(str, size, seed)} }
    x.report("string#hash") { 10_000.times{str.hash}  }
    x.report("Zlib#crc32") { 10_000.times{ Zlib.crc32(str) } }
  end
  puts
  puts "Ruby Version is #{RUBY_VERSION}"
  puts "City Hash (64):    " + CityHash._Z10CityHash64PKcm(str, size).to_s
  puts "City Hash (128):   " + CityHash._Z11CityHash128PKcm(str, size).to_s
  # the String#hash value changes everytime you run the script (???)
  puts "String#hash:       " + str.hash.to_s
  puts "Zlib#crc32         " + Zlib.crc32(str).to_s
end

Actually, I've added

extern "C"

before these functions and now can use it with normal names. But I've tried without extern "C' stuff too and it still crushes .