Ruby 2.5/2.6

New Features and Performance Improvements

by Thomas Leitner

Ruby 2.5

Ruby 2.5 New Features

  • rescue/else/ensure are now allowed to be used directly with do/end blocks. Feature #12906
  • Print backtrace and error message in reverse order (oldest call first, most recent call last) Feature #8661
  • Support branch coverage and method coverage measurement. Feature #13901
  • Top-level constant look-up is no longer available. Feature #11547

Ruby 2.5 New Features (cont.)

  • One of our most loved libraries, pp.rb, is now automatically loaded. You no longer have to write require “pp”. Feature #14123
  • Promote cmath, csv, date, dbm, etc, fcntl, fiddle, fileutils, gdbm, ipaddr, scanf, sdbm, stringio, strscan, webrick, zlib from standard libraries to default gems.
  • Onigmo 6.1.3 absence operator (?~exp) (also in 2.4.1)

Ruby 2.5 New Features (cont.)

  • Better RDoc by using Ripper instead of IRB based parser
  • IO#write now receives multiple arguments. Feature #9323
  • Hash#transform_keys Feature #13583
  • -"literal" (String#-@) optimized to return the same object (same as "literal".freeze in Ruby 2.1+) Feature #13295
  • binding.irb automatically requires irb and runs Bug #13099

Ruby 2.5 Performance Improvements

  • 5-10% due to removal of trace instructions (dynamic instrumentation is used instead) Feature #14104
  • Block passing by a block parameter (e.g. def foo(&b); bar(&b); end) is about 3 times faster than Ruby 2.4 by “Lazy Proc allocation” technique. Feature #14045

Ruby 2.5 Performance Improvements (cont.)

  • Improve performance of some built-in methods including Array#concat, Enumerable#sort_by, String#concat, String#index, Time#+, and more.
  • IO.copy_stream uses copy_file_range(2) to provide copy offloading. Feature #13867

Ruby 2.5 Further Reading

Note: The items for the slides “New Features” and “Performance Improvements” were mostly taken verbatim from the first two links.

Ruby 2.6

Ruby 2.6 New Features

  • Add Binding#source_location. Feature #14230
  • Constant names may start with a non-ASCII capital letter. Feature #13770
  • Endless range syntax (1..), same as (1..nil) Feature #12912
  • Range#=== now uses #cover? instead of #include? method (#cover? is faster because it only checks the bounds) Feature #14575

Ruby 2.6 Performance Improvements

Ruby 2.6 JIT (YARV-MJIT)

  • By Takashi Kokubun (and Vladimir Makarov)
  • Use --jit to activate it (there are some other options)
  • Compiles C code via gcc/clang
  • Significant enhancements between preview1 and preview2
  • Follow changes on Github https://github.com/ruby/ruby/commits?author=k0kubun
  • Enhancements measured using optcarrot benchmark which is also used for Ruby 3x3 measurement

Ruby 2.6 Further Reading

Ruby AST Differences

Demo time

ruby -ve "puts RubyVM::InstructionSequence.
          compile('x, y = 1, 2; [x, y].max').disasm"

Benchmarking

Available Tools

  • Built-in library benchmark
  • Better: gem benchmark-ips
  • Newest: gem benchmark-driver

benchmark

  • Very simple

  • Usage:

    Benchmark.bm do |x|
  x.report("name1") { code1 }
  x.report("name2") { code2 }
end
  • Better use Benchmark.bmbm since it warms up the runtime
  • You need to determine the loop count yourself

benchmark-ips

  • Usage like with benchmark:

    Benchmark.ips do |x|
  x.report("name1") { code1 }
  x.report("name2", "code-as-string")
  x.compare!
end
  • Warm-up done by default
  • Determines loop count automatically
  • Helper method for easy comparions
  • Allows to pass code as string to avoid block-calling avoid

benchmark-driver

  • Made by Takashi Kokubun to easily test different versions of Ruby and micro-benchmarks

  • Usage similar:

    Benchmark.driver do |x|
  x.prelude "set-up code here"
  x.report("name1", "code-as-string")
  x.report("name2", "code-as-string")
end
  • In addition to warm-up and loop count determination, it subtracts the loop overhead itself which is important for micro-benchmarks
  • Allows testing multiple Ruby version using rbenv
  • Drive benchmarks using simple YAML files

Benchmark Demo Time

a < b ? b : a

vs

[a, b].max

HexaPDF Benchmarks

  • Same code run on different Ruby versions
  • Alas, Ruby 2.6 in JIT-mode is slower

Raw Text Benchmark

raw text benchmark

Line Wrapping Benchmark

line wrapping benchmark

Optcarrot

The official benchmark for verifying the Ruby 3x3 goal!

git clone git@github.com:mame/optcarrot.git
cd optcarrot
# switch to Ruby 2.6-preview2
ruby -Ilib bin/optcarrot --benchmark examples/Lan_Master.nes
ruby --jit -Ilib bin/optcarrot --benchmark examples/Lan_Master.nes
# switch to TruffleRuby :)
ruby -Ilib -r./tools/shim bin/optcarrot --benchmark examples/Lan_Master.nes

Thank You

You can find the files used during the presentation in the files directory.