1brc_noalloc.rb

require 'strscan'

stats = Hash.new { |h, k| h[k] = [10_000, -10_000, 0, 0] }

SEMI = String.new(';', encoding: Encoding::BINARY)
NEWLINE = String.new("\n", encoding: Encoding::BINARY)

BUFLEN = 128 * 1024 * 1024
buf = ' ' * BUFLEN
buf.force_encoding(Encoding::BINARY)

city = String.new(' ', encoding: Encoding::BINARY) * 1024
temp_s = String.new(' ', encoding: Encoding::BINARY) * 1024
remainder = String.new(' ', encoding: Encoding::BINARY) * 1024
empty = String.new('', encoding: Encoding::BINARY)

citylen = 1024
templen = 1024

File.open('measurements.txt', encoding: Encoding::BINARY) do |f| # rubocop:disable Metrics/BlockLength
  scanner = StringScanner.new(buf)
  remainder = empty
  loop do # rubocop:disable Metrics/BlockLength
    f.readpartial(BUFLEN, buf)
    citypos = 0
    temppos = 0
    scanner.reset

    # special handling for first line of block
    fcity, ftemp_s = (remainder + scanner.scan_until(/\n/)).split(SEMI, 2)
    ftemp = ftemp_s.to_f

    existing = stats[fcity]
    min, max, sum, count = existing

    existing[0] = ftemp if ftemp < min
    existing[1] = ftemp if ftemp > max
    existing[2] = sum + ftemp
    existing[3] = count + 1

    citypos = scanner.pos

    loop do
      break unless scanner.skip_until(/;/)

      newcitylen = scanner.pos - citypos - 1
      city.bytesplice(0, citylen, buf, citypos, newcitylen)
      citylen = newcitylen
      temppos = scanner.pos

      break unless scanner.skip_until(/\n/)

      newtemplen = scanner.pos - temppos - 1
      temp_s.bytesplice(0, templen, buf, temppos, newtemplen)
      templen = newtemplen
      citypos = scanner.pos

      temp = temp_s.to_f

      existing = stats[city]
      min, max, sum, count = existing

      existing[0] = temp if temp < min
      existing[1] = temp if temp > max
      existing[2] = sum + temp
      existing[3] = count + 1
    end
    remainder.bytesplice(0..-1, buf, citypos..-1)
  end
rescue EOFError
  # all good
end

stats.each do |city, cstats|
  puts "#{city}=#{cstats[0]}/#{cstats[2] / cstats[3]}/#{cstats[1]}"
end