Deep dive into Ruby's require - RubyConf Taiwan 2023

Copyright © 2020 Present ANDPAD Inc. This information is confidential and was prepared by ANDPAD Inc. for the use of our client. It is not to be relied on by and 3rd party. Proprietary & Confidential 無断転載・無断複製の禁止
Deep dive into
Ruby's require
Hiroshi SHIBATA @hsbt
2023/12/15 RubyConf Taiwan 2023

Hiroshi SHIBATA
https://hsbt.org
@hsbt
Ruby core team
RubyGems/Bundler team
Technical fellow at ANDPAD
Self introduction

Introduction of ANDPAD

What's feature of ANDPAD?

Learning Ruby's
require

Inside of `require` of Ruby

Basic knowlege of require
• `require` is major method in Ruby.
• `require` can handle Ruby and
C/Rust extension with your
platform like linux or macOS.
• `require` find $LOAD_PATH by your
installation path originally.
• `require_relative` find only
current path.
>> require 'rss'
=> true
>> require 'rss'
=> false
>> require "bigdecimal"
=> true
#
>> require "bigdecimal.bundle"
=> true
#
>> require "bigdecimal.so"
=> true
>> require_relative "rss"
...cannot load such file -- ...

load is a few different with require
• Ruby also has `load`. It's alternative of require.
• `load` didn't find your installation path. It find only current path.
• `load` need to extension of file like “.rb”.
• `load` can load ruby code many times.
>> load "psych"
(irb):1:in `load': cannot load such file -- psych (LoadError)
>> load "psych.rb"
=> true
>> load "psych.rb"
...lib/psych.rb:236: warning: already initialized constant
Psych::LIBYAML_VERSION
(snip)
=> true

autoload is more Ruby'ish way
• `require` and `load` are immediately evaluated by Ruby.
• But `autoload` is only load Ruby code when constant is
referenced.
• Ex. You call `Bundler::Definition`, Ruby will call `bundler/dependency.rb`
module Bundler
environment_preserver = EnvironmentPreserver.from_env
ORIGINAL_ENV = environment_preserver.restore
environment_preserver.replace_with_backup
SUDO_MUTEX = Thread::Mutex.new
autoload :Checksum, File.expand_path("bundler/checksum", __dir__)
autoload :Definition, File.expand_path("bundler/definition", __dir__)
autoload :Dependency, File.expand_path("bundler/dependency", __dir__)
(snip)

How load Ruby code from outside
of Ruby installation path?

We have RubyGems
• RubyGems is a package/library for the Ruby programming language
• We can install gems from rubygems.org
• gemspec is a file describing Gem::Specification
• This class for defining metadata including name, version,
platform, etc.
• RubyGems provide extended `require` and `gem` method for loading
gem for us.

gemspec in your code
• You can see gemspec with `Gem.loaded_specs` like this:
• `Gem::Speci
fi
cation#dependencies` is important parts of your application.
>> Gem.loaded_specs["rack"]
=>
Gem::Speci
fi
cation.new do |s|
s.name = "rack"
s.version = Gem::Version.new("2.2.8")
s.installed_by_version = Gem::Version.new("3.4.10")
s.authors = ["Leah Neukirchen"]
s.date = Time.utc(2023, 7, 31)
s.dependencies = [Gem::Dependency.new("minitest", Gem::Requirement.new(["~> 5.0"]), :development),
Gem::Dependency.new("minitest-sprint", Gem::Requirement.new([">= 0"]), :development),
Gem::Dependency.new("minitest-global_expectations", Gem::Requirement.new([">= 0"]), :development),
Gem::Dependency.new("rake", Gem::Requirement.new([">= 0"]), :development)]
s.description = "Rack provides a minimal, modular and adaptable interface for developingnweb
applications in Ruby. By wrapping HTTP requests and responses innthe simplest way possible, it uni
fi
es
and distills the API for webnservers, web frameworks, and software in between (the so-
callednmiddleware) into a single method call.n"
(...snip...)
end

Performance issues with RubyGems
• RubyGems extends `require` n `kernel_require.rb`
def require(path) # :doc:
return gem_original_require(path) unless
Gem.discover_gems_on_require
begin
RUBYGEMS_ACTIVATION_MONITOR.enter
path = path.to_path if path.respond_to? :to_path
if spec = Gem.
fi
nd_unresolved_default_spec(path)
# Ensure -I beats a default gem
resolved_path = begin
rp = nil
(snip)
$ wc -l *
165 kernel_require.rb
49 kernel_warn.rb

Why slow down with RubyGems?
• Find all of your gems: Gem::specification.find_by_path
def self.find_by_path(path)
path = path.dup.freeze
spec = @@spec_with_requirable_file[path] ||= stubs.find do |s|
s.contains_requirable_file? path
end || NOT_FOUND
spec.to_spec
end
This returns all of your gemspec
$ ruby -ve "t = Time.now; require 'bigdecimal'; p Time.now - t"
0.272687
$ set -e GEM_HOME
$ ruby -ve "t = Time.now; require 'bigdecimal'; p Time.now - t"
0.000786
$ ruby --disable-gems -ve "t = Time.now; require 'bigdecimal'; p Time.now - t"
0.000512
• My environment have 2800+ gems. RubyGems search them.

Introduce Kernel#gem method
• RubyGems provide `gem` method to
handle name and requirements.
• It's different with gem method on
Gemfile.
• This method inject gem path into your
$LOAD_PATH with requirements
>> require "rake"
=> true
>> Rake::VERSION
=> "13.1.0"
#
>> gem "rake", "13.0.0"
=> true
>> require "rake"
=> true
>> Rake::VERSION
=> "13.0.0"
#
>> gem "rake", "> 12", "< 13"
=> true
>> require "rake"
=> true
>> Rake::VERSION
=> "12.3.3"
>> orig_paths = $LOAD_PATH.dup
>> gem "rake", "> 12", "< 13"
=> true
>> $LOAD_PATH - orig_paths
=> ["/Users/hsbt/.local/share/gem/gems/rake-12.3.3/lib"]

Inside of gem method
• Kernel#gem create `Gem::Dependency` instance with name and
requirements. After that activate it.
def gem(gem_name, *requirements) # :doc:
(snip)
dep = Gem::Dependency.new(gem_name, *requirements)
loaded = Gem.loaded_specs[gem_name]
return false if loaded && dep.matches_spec?(loaded)
spec = dep.to_spec
if spec
if Gem::LOADED_SPECS_MUTEX.owned?
spec.activate
else
Gem::LOADED_SPECS_MUTEX.synchronize { spec.activate }
end
end

RubyGems has some problems
• RubyGems didn't have resolution mechanism with multiple libraries.
• If you activated Rails 7, you should specify all of related libraries
that support Rails 7.
• It's difficult to resolve only `gem` method.
• Extended `require` is slow if you have a lot of gems like Rails
application.

How works bundler?

Why Bundler come
here?

What’s Bundler?
# frozen_string_literal: true
source "https://rubygems.org"
gem "rake", ">= 11.1”
• Bundler is also package manager of Ruby language
• Bundler focused version locking feature
• Bundler extends a lot of RubyGems resources like gemspec.
• Bundler works with Gemfile like:

Introduction of Lockfile
• Bundler generates lockfile named Gemfile.lock from Gemfile with dependency
resolution.
• Bundler use this lockfile for version locking in your applicaiton
# Gemfile
# frozen_string_literal: true
gem "rss"
# Gemfile.lock
GEM
remote: https://rubygems.org/
specs:
rexml (3.2.5)
rss (0.2.9)
rexml
PLATFORMS
arm64-darwin-23
DEPENDENCIES
rss
BUNDLED WITH
2.5.0.dev

Performance issues
• Bundler also extends RubyGems classes/methods. It's enabled when we used Bundler.
• `Gem::Specification#extension_dir` needs to handle git resource of Gemfile like this:
# for gem “rails”, git: “https://github.com/rails/rails"
alias_method :rg_extension_dir, :extension_dir
def extension_dir
# following instance variable is already used in original method
# and that is the reason to prefix it with bundler_ and add rubocop exception
@bundler_extension_dir ||= if source.respond_to?(:extension_dir_name)
unique_extension_dir = [source.extension_dir_name,
File.basename(full_gem_path)].uniq.join("-")
File.expand_path(File.join(extensions_dir, unique_extension_dir))
else
rg_extension_dir
end
end

Basic case of Gemfile
and bundle exec

Basic case of Gemfile and Bundler
• How Bundler resolve to slow
down with RubyGems and
Bundler?
• I'll introduce it with
`bundle exec` command
with simple rss examples.
# Gemfile.lock
GEM
remote: https://rubygems.org/
specs:
rexml (3.2.5)
rss (0.2.9)
rexml
PLATFORMS
arm64-darwin-23
DEPENDENCIES
rss
BUNDLED WITH
2.5.0.dev

What's Bundler.setup
def setup(*groups)
@definition.ensure_equivalent_gemfile_and_lockfile if
Bundler.frozen_bundle?
# Has to happen first
clean_load_path
specs = @definition.specs_for(groups)
SharedHelpers.set_bundle_environment
Bundler.rubygems.replace_entrypoints(specs)
# Activate the specs
load_paths = specs.map do |spec|
check_for_activated_spec!(spec)
Bundler.rubygems.mark_loaded(spec)
spec.load_paths.reject {|path| $LOAD_PATH.include?(path) }
end.reverse.flatten
Bundler.rubygems.add_to_load_path(load_paths)
setup_manpath
lock(:preserve_unknown_sections => true)
self
end
• `Bundler.setup` and
`Bundler.require` is most
important parts of Bundler
• These methods
defined at `runtime.rb`.
• `bundle exec` call
`Bundler.setup` and
`Kernel.exec`.

Reset your environment with Bundler
@definition.ensure_equivalent_gemfile_and_lockfile if Bundler.frozen_bundle?
# Has to happen first
clean_load_path
• At first, Bundler update your lockfile and install new versions if it's
needed. After that, Reject gem paths that are not `require` yet.
def clean_load_path
loaded_gem_paths = Bundler.rubygems.loaded_gem_paths
$LOAD_PATH.reject! do |p|
resolved_path = resolve_path(p)
next if $LOADED_FEATURES.any? {|lf| lf.start_with?(resolved_path) }
loaded_gem_paths.delete(p)
end
$LOAD_PATH.uniq!
end

• `Bundler.definition` and `Bundler::Resolver` is core parts for this.
• Bundler.definition create instance of `Bundler::Resolver` and call
resolution methods inside `specs_for`.
• `specs` is instance of Bundler::SpecSet.
• Bundler inject `bundler` as dependency into Gemfile.
How select dependencies by Bundler.definition
specs = @definition.specs_for(groups)
SharedHelpers.set_bundle_environment
Bundler.rubygems.replace_entrypoints(specs)
>> specs.map(&:name)
=> ["bundler", "rexml", "rss"]
This method revert extended require by rubygems

Update your LOAD_PATH with scratch
~/.rbenv/versions/master/lib/ruby/gems/3.3.0+0/gems/rss-0.2.9/lib
~/.rbenv/versions/master/lib/ruby/gems/3.3.0+0/gems/rexml-3.2.5/lib
~/.rbenv/versions/master/lib/ruby/gems/3.3.0+0/gems/bundler-2.5.0.dev/lib
# Activate the specs
load_paths = specs.map do |spec|
check_for_activated_spec!(spec)
Bundler.rubygems.mark_loaded(spec)
spec.load_paths.reject {|path| $LOAD_PATH.include?(path) }
end.reverse.flatten
Bundler.rubygems.add_to_load_path(load_paths)
setup_manpath
lock(:preserve_unknown_sections => true)
self
• These logic is easy to understand. We generate paths generated by resolved gemspec.
`Gem::Specification#load_paths` returns load paths from gemspec.
• `load_paths` returns like this:

Future of standard libraries

What are default and
bundled gems?

What's Default gems
• The Ruby core team released "Default gems" to the rubygems.org.
• You can install standard libraries of Ruby via RubyGems.
• Default gems are openssl, psych, json, etc… You can see all of
default gems at https://stdgems.org/
• Rubygems have a detection method for default gems.
>> require 'rss'
=> true
>> Gem.loaded_specs["rss"].default_gem?
=> false
>> require 'openssl'
=> true
>> Gem.loaded_specs["openssl"].default_gem?
=> true

Basic scenario of Bundler resolution
• `Bundler.rubygems.replace_entrypoints` inject gemspecs of default
gems into dependencies of Gemfile.
specs = Declared in Gemfile: rails, nokogiri, sidekiq, etc...
default_spec = Default gems: csv, psych, json, etc...
+
Bundler.rubygems.default_stubs.each do |stub|
default_spec = stub.to_spec
default_spec_name = default_spec.name
next if specs_by_name.key?(default_spec_name)
specs << default_spec
specs_by_name[default_spec_name] = default_spec
end

What's Bundled gems
• We bundled *.gem and unpacked
fi
les to tarball package for Bundled
gems with `gems/bundled_gems` in ruby/ruby repository like this:
• `make install` installed Bundled gem your box.

The major problem for the bundled gems
If you use Bundler, you need to add the bundled gems into your Gem
fi
le.
gem “rss”
gem "bigdecimal" # You need to this always after Ruby 3.4
…
I need to consider to transition and migration plan for this.
But I have no idea yet. Maybe, I will add the some mechanism to Bundler internal
to care about this.

Warning feature of bundled gems
• I added warning feature about bundled
gems.
• You can see how handle bundled gems
like this.
module Gem::BUNDLED_GEMS
SINCE = {
"rexml" => "3.0.0",
"rss" => "3.0.0",
"webrick" => "3.0.0",
"matrix" => "3.1.0",
"net-ftp" => "3.1.0",
"net-imap" => "3.1.0",
"net-pop" => "3.1.0",
"net-smtp" => "3.1.0",
"prime" => "3.1.0",
"abbrev" => "3.4.0",
"base64" => "3.4.0",
"bigdecimal" => "3.4.0",
"csv" => "3.4.0",
"drb" => "3.4.0",
"getoptlong" => "3.4.0",
"mutex_m" => "3.4.0",
"nkf" => "3.4.0",
"observer" => "3.4.0",
"racc" => "3.4.0",
"resolv-replace" => "3.4.0",
"rinda" => "3.4.0",
"syslog" => "3.4.0",
}.freeze
$ cat -p Gemfile
gem "rss"
$ bundle exec irb
>> require "csv"
(irb):1: warning: csv which will no longer be part of
the default gems since Ruby 3.4.0. Add csv to your
Gemfile or gemspec.
=> true

How extend require in your use case?
We easily extend `require` like this. I only enabled this extension under
the bundler. I added this to monkey patch collection of bundler.
def replace_require(specs)
return if [::Kernel.singleton_class, ::Kernel].any? {|klass| klass.respond_to?(:no_warning_require) }
[::Kernel.singleton_class, ::Kernel].each do |kernel_class|
kernel_class.send(:alias_method, :no_warning_require, :require)
kernel_class.send(:define_method, :require) do |name|
if message = ::Gem::BUNDLED_GEMS.warning?(name, specs: specs)
warn message, :uplevel => 1
end
kernel_class.send(:no_warning_require, name)
end
if kernel_class == ::Kernel
kernel_class.send(:private, :require)
else
kernel_class.send(:public, :require)
end
end
end

We want Namespace feature

Activation issue of default gems
We still have a activation issue
with RubyGems and the default
gems like `json`.
How handle this problem?
You have already activated timeout 0.3.1, but
your Gemfile requires timeout 0.3.2. Since
timeout is a default gem, you can either remove
your dependency on it or try updating to a
newer version of bundler that supports timeout
as a default gem.

How handle activation problem?
1. Rewrite default gems to simple Ruby code
→ 😆
2. Vendoring approach when RubyGems uses only pure ruby libraries like URI.
→ 😩
3. Rewrite default gems written by C extension to pure ruby code.
→ 😩

Example for rewrite C to Ruby
You can use any versions of Psych.
Because RubyGems and Bundler use the
original Ruby implementation for YAML
serializer.
When RubyGems uses json-2.1.0, You
couldn’t use json 1.8.x. Because ruby
gems and rubygems.org never uses
JSON format.

We considered namespace feature with RubyGems and Bundler
tagomoris-san proposed namespace
feature for package system of Ruby.
It may help our activation problem.
Ex. RubyGems and Bundler will use JSON
under only their namespace. You can
activate any version of JSON on your
application.

Writing/Thinking Ruby continuously in 2024
RubyKaigi 2024 will be coming Okinawa island in Japan at May, 2024.
Looking forward seeing you again at RubyKaigi 2024 :wave:

Wrap up

Conclusion
• I talked about...
• Knowledge require, RubyGems, Bundler
• How works Bundler modify LOAD_PATH of Ruby
• How works extends require for use-cases
< Ruby is a programmer's best friend

Deep dive into Ruby's require - RubyConf Taiwan 2023

More Related Content

Similar to Deep dive into Ruby's require - RubyConf Taiwan 2023 (20)

More from Hiroshi SHIBATA (20)

Recently uploaded (20)

Deep dive into Ruby's require - RubyConf Taiwan 2023