crystalruby

crystalruby is a gem that allows you to write Crystal code, inlined in Ruby. All you need is a modern crystal compiler installed on your system.

You can then turn simple methods into Crystal methods as easily as demonstrated below:

require 'crystalruby'

module MyTestModule
  # The below method will be replaced by a compiled Crystal version
  # linked using FFI.
  crystalize [a: :int, b: :int] => :int
  def add(a, b)
    a + b
  end
end

# This method is run in Crystal, not Ruby!
MyTestModule.add(1, 2) # => 3

With as small a change as this, you should be able to see a significant increase in performance for some Ruby code. E.g.

require 'crystalruby'
require 'benchmark'

module PrimeCounter
  crystalize [n: :int32] => :int32
  def count_primes_upto_cr(n)
    (2..n).each.count do |i|
      is_prime = true
      (2..Math.sqrt(i).to_i).each do |j|
        if i % j == 0
          is_prime = false
          break
        end
      end
      is_prime
    end
  end

  module_function

  def count_primes_upto_rb(n)
    (2..n).each.count do |i|
      is_prime = true
      (2..Math.sqrt(i).to_i).each do |j|
        if i % j == 0
          is_prime = false
          break
        end
      end
      is_prime
    end
  end
end

include PrimeCounter
puts(Benchmark.realtime { count_primes_upto_rb(1000_000) })
puts(Benchmark.realtime { count_primes_upto_cr(1000_000) })

2.8195170001126826 # Ruby
0.3402599999681115 # Crystal

Note: The first run of the Crystal code will be slower, as it needs to compile the code first. The subsequent runs will be much faster.

You can call embedded crystal code, from within other embedded crystal code. E.g.

module Cache

  crystalize [key: :string] => :string
  def redis_get(key)
    rds = Redis::Client.new
    value = rds.get(key).to_s
  end

  crystalize [key: :string, value: :string] => :string
  def redis_set_and_return(key, value)
    redis = Redis::Client.new
    redis.set(key, value)
    Cache.redis_get(key)
  end
end
Cache.redis_set_and_return('test', 'abc')
puts Cache.redis_get('test')

$ abc

Where the Crystal syntax is also valid Ruby syntax, you can just write Ruby. It'll be compiled as Crystal automatically.

E.g.

crystalize [a: :int, b: :int] => :int
def add(a, b)
  puts "Adding #{a} and #{b}"
  a + b
end

Some Crystal syntax is not valid Ruby, for methods of this form, we need to define our functions using a raw: true option

crystalize [a: :int, b: :int] => :int, raw: true
def add(a, b)
  <<~CRYSTAL
    c = 0_u64
    a + b + c
  CRYSTAL
end

The below is a stand-alone one-file script that allows you to quickly see crystalruby in action.

# crystalrubytest.rb
require 'bundler/inline'

gemfile do
  source 'https://rubygems.org'
  gem 'crystalruby'
end

require 'crystalruby'

module Adder
  crystalize [a: :int, b: :int] => :int
  def add(a, b)
    a + b
  end
end

puts Adder.add(1, 2)

Currently primitive types are supported. Composite types are supported using JSON serialization. C-Structures are a WIP. To see the list of currently supported primitive type mappings of FFI types to crystal types, you can check: CrystalRuby::Typemaps::CRYSTAL_TYPE_MAP E.g.

CrystalRuby::Typemaps::CRYSTAL_TYPE_MAP
=> {:char=>"Int8",
 :uchar=>"UInt8",
 :int8=>"Int8",
 :uint8=>"UInt8",
 :short=>"Int16",
 :ushort=>"UInt16",
 :int16=>"Int16",
 :uint16=>"UInt16",
 :int=>"Int32",
 :uint=>"UInt32",
 :int32=>"Int32",
 :uint32=>"UInt32",
 :long=>"Int32 | Int64",
 :ulong=>"UInt32 | UInt64",
 :int64=>"Int64",
 :uint64=>"UInt64",
 :long_long=>"Int64",
 :ulong_long=>"UInt64",
 :float=>"Float32",
 :double=>"Float64",
 :bool=>"Bool",
 :void=>"Void",
 :string=>"String"}

The library allows you to pass complex nested structures using JSON as a serialization format. The type signatures for composite types can use ordinary Crystal Type syntax. Type conversion is applied automatically.

E.g.

crystalize [a: json{ Int64 | Float64 | Nil }, b: json{ String | Array(Bool)  } ] => :void
def complex_argument_types
  puts "Got #{a} and #{b}"
end

crystalize [] => json{ Int32 | String | Hash(String, Array(NamedTuple(hello: Int32)) | Time)}
def complex_return_type
  return {
    "hello" => [
      {
        hello: 1,
      },
    ],
    "world" => Time.utc
  }
end

Type signatures validations are applied to both arguments and return types.

[1] pry(main)> Foo.complex_argument_types(nil, "test")
Got  and test
=> nil

[2] pry(main)> Foo.complex_argument_types(88, [true, false, true])
Got 88 and [true, false, true]
=> nil

[3] pry(main)> Foo.complex_argument_types(88, [true, false, 88])
ArgumentError: Expected Bool but was Int at line 1, column 15
from crystalruby.rb:303:in `block in compile!'

You can name your types, for more succinct method signatures. The type names will be mirrored in the generated Crystal code. E.g.

IntArrOrBoolArr = crtype{ Array(Bool) | Array(Int32) }

crystalize [a: json{ IntArrOrBoolArr }] => json{ IntArrOrBoolArr }
def method_with_named_types(a)
  return a
end

Exceptions thrown in Crystal code can be caught in Ruby.

You can use any Crystal shards and write ordinary, stand-alone Crystal code.

The default entry point for the crystal shared library generated by the gem is inside ./crystalruby/{library_name}/src/main.cr. {library_name} defaults to crystalruby if you haven't explicitly specific a different library target.

This file is not automatically overridden by the gem, and is safe for you to define and require new files relative to this location to write additional stand-alone Crystal code.

You can define shard dependencies inside ./crystalruby/{library_name}/src/shard.yml Run the below to install new shards

bundle exec crystalruby install

Remember to also require these dependencies after installing them to make them available to crystalruby code. E.g. inside ./crystalruby/{libraryname}/src/main.cr

You can edit the default paths for crystal source and library files from within the ./crystalruby.yaml config file.

Sometimes you may want to wrap a Crystal method in Ruby, so that you can use Ruby before the Crystal code to prepare arguments, or after the Crystal code, to apply transformations to the result. A real-life example of this might be an ActionController method, where you might want to use Ruby to parse the request, perform auth etc., and then use Crystal to perform some heavy computation, before returning the result from Ruby. To do this, you simply pass a block to the crystalize method, which will serve as the Ruby entry point to the function. From within this block, you can invoke super to call the Crystal method, and then apply any Ruby transformations to the result.

module MyModule
  crystalize [a: :int32, b: :int32] => :int32 do |a, b|
    # In this example, we perform automated conversion to integers inside Ruby.
    # Then add 1 to the result of the Crystal method.
    result = super(a.to_i, b.to_i)
    result + 1
  end
  def add(a, b)
    a + b
  end
end

MyModule.add("1", "2")

crystalruby also allows you to write inline Crystal code that does not require binding to Ruby. This can be useful for e.g. performing setup operations or initializations.

Follow these steps for a toy example of how we can use crystalized ruby and inline chunks to expose the crystal-redis library to Ruby.

1. Start our toy project

mkdir crystalredis
cd crystalredis
bundle init

2. Add dependencies to our Gemfile and run bundle install

# frozen_string_literal: true

source "https://rubygems.org"

gem 'crystalruby'

# Let's see if performance is comparable to that of the redis gem.
gem 'benchmark-ips'
gem 'redis'

3. Write our Redis client

# Filename: crystalredis.rb
require 'crystalruby'

module CrystalRedis

  crystal do
    CLIENT = Redis.new
    def self.client
      CLIENT
    end
  end

  crystalize [key: :string, value: :string] => :void
  def set(key, value)
    client.set(key, value)
  end

  crystalize [key: :string] => :string
  def get(key)
    client.get(key).to_s
  end
end

4. Load the modules (without running them) to generate our Crystal project skeleton.

bundle exec ruby crystalredis.rb

5. Add the missing Redis dependency to our shard.yml

# filename:  crystalruby/src/shard.yml
dependencies:
  redis:
    github: stefanwille/crystal-redis

# filename: main.cr
require "redis"
require "./generated/index"

bundle exec crystalruby install

6. Compile and benchmark our new module in Ruby

# Filename: benchmark.rb
# Let's compare the performance of our CrystalRedis module to the Ruby Redis gem
require_relative "crystalredis"
require 'redis'
require 'benchmark/ips'

Benchmark.ips do |x|
  rbredis = Redis.new

  x.report(:crredis) do
    CrystalRedis.set("hello", "world")
    CrystalRedis.get("hello")
  end

  x.report(:rbredis) do
    rbredis.set("hello", "world")
    rbredis.get("hello")
  end
end

7. Run the benchmark

$ bundle exec ruby benchmark.rb

#crystalredis wins! (Warm up during first run will be slow for crredis, due to first compilation)

ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [arm64-darwin22]
Warming up --------------------------------------
             crredis     1.946k i/100ms
             rbredis     1.749k i/100ms
Calculating -------------------------------------
             crredis     22.319k (± 1.7%) i/s -    112.868k in   5.058448s
             rbredis     16.861k (± 9.1%) i/s -     83.952k in   5.024941s

You can control whether CrystalRuby builds in debug or release mode by setting following config option

CrystalRuby.configure do |config|
  config.debug = false
end

By default, Crystal code is only JIT compiled. In production, you likely want to compile the Crystal code ahead of time. To do this, you can create a dedicated file which

• Preloads all files Ruby code with embedded crystal
• Forces compilation.

E.g.

# E.g. crystalruby_build.rb
require "crystalruby"

CrystalRuby.configure do |config|
  config.debug = false
end

require_relative "foo"
require_relative "bar"

CrystalRuby.compile!

Then you can run this file as part of your build step, to ensure all Crystal code is compiled ahead of time.

While Ruby programs allow multi-threading, Crystal (if not using experimental multi-thread support) uses only a single thread and utilises Fiber based cooperative-multitasking to allow for concurrent execution. This means that by default, Crystal libraries can not safely be invoked in parallel across multiple Ruby threads.

To safely utilise crystalruby in a multithreaded environment, crystalruby implements a Reactor, which multiplexes all Ruby calls to Crystal across a single thread.

By default crystalruby methods are blocking/synchronous, this means that for blocking operations, a single crystalruby call can block the entire reactor across all threads.

To allow you to benefit from Crystal's fiber based concurrency, you can use the async: true option on crystalized ruby methods. This allows several Ruby threads to invoke Crystal code simultaneously.

E.g.

module Sleeper
  crystalize [] => :void
  def sleep_sync
    sleep 2
  end

  crystalize [] => :void, async: true
  def sleep_async
    sleep 2
  end
end

5.times.map{ Thread.new{ Sleeper.sleep_sync } }.each(&:join) # Will take 10 seconds
5.times.map{ Thread.new{ Sleeper.sleep_async } }.each(&:join) # Will take 2 seconds (the sleeps are processed concurrently)

There is a small amount of synchronization overhead to multiplexing calls across a single thread. Ad-hoc testing on a fast machine amounts this to be within the order of 10 microseconds per call. For most use-cases this overhead is negligible, especially if the bulk of your CPU heavy task occurs exclusively in Crystal code. However, if you are invoking very fast Crystal code from Ruby in a tight loop (e.g. a simple 1 + 2) then the overhead of the reactor can become significant.

In this case you can use the crystalruby in a single-threaded mode to avoid the reactor overhead and greatly increase performance, with the caveat that all calls to Crystal must occur from a single thread. If your Ruby program is already single-threaded this is not a problem.

CrystalRuby.configure do |config|
  config.single_thread_mode = true
end

crystalruby supports live reloading of Crystal code. It will intelligently recompile Crystal code only when it detects changes to the embedded function or block bodies. This allows you to iterate quickly on your Crystal code without having to restart your Ruby process in live-reloading environments like Rails.

Large Crystal projects are known to have long compile times. To mitigate this, crystalruby supports splitting your Crystal code into multiple libraries. This allows you to only recompile any libraries that have changed, rather than all crystal code within the project. To indicate which library a piece of embedded Crystal code belongs to, you can use the lib option in the crystalize and crystal methods. If the lib option is not provided, the code will be compiled into the default library (simply named crystalruby).

module Foo
  crystalize lib: "foo"
  def bar
    puts "Hello from Foo"
  end

  crystal lib: "foo" do
    REDIS = Redis.new
  end
end

Naturally, Crystal methods must reside in the same library to natively interact. Cross library interaction can be facilitated via Ruby code.

In cases where compiled assets are in left an invalid state, it can be useful to clear out generated assets and rebuild from scratch.

To do this execute:

bundle exec crystalruby clean

crystalruby's primary purpose is to provide ergonomic access to Crystal from Ruby, over FFI. For simple usage, advanced knowledge of Crystal should not be required.

However, the abstraction it provides should remain simple, transparent, and easy to hack on and it should not preclude users from supplementing its capabilities with a more direct integration using ffi primtives.

It should support escape hatches to allow it to coexist with code that performs a more direct FFI integration to implement advanced functionality not supported by crystalruby.

The library is currently in its infancy. Planned additions are:

• Simple mixin/concern that utilises FFI::Struct for bi-directional passing of Ruby objects and Crystal objects (by value).
• Install command to generate a sample build script, and supports build command (which simply verifies then invokes this script)
• Call Ruby from Crystal using FFI callbacks (implement .expose_to_crystal)
• Support long-lived synchronized objects (through use of synchronized memory arena to prevent GC).
• Support for passing crystalruby types by reference (need to contend with GC).
• Explore mechanisms to safely expose true parallelism using FFI over Ractors

To get started, add this line to your application's Gemfile:

gem 'crystalruby'

And then execute:

$ bundle

Or install it yourself as:

$ gem install crystalruby

crystalruby supports some basic configuration options, which can be specified inside a crystalruby.yaml file in the root of your project. You can run crystalruby init to generate a configuration file with sane defaults.

crystalruby init

crystal_src_dir: "./crystalruby"
crystal_codegen_dir: "generated"
crystal_main_file: "main.cr"
crystal_lib_name: "crlib"
crystal_codegen_dir: "generated"
debug: true

Alternatively, these can be set programmatically, e.g:

CrystalRuby.configure do |config|
  config.crystal_src_dir = "./crystalruby"
  config.crystal_codegen_dir = "generated"
  config.debug = true
  config.verbose = false
  config.colorize_log_output = false
  config.log_level = :info
end

After checking out the repo, run bin/setup to install dependencies. Then, run rake test to run the tests. You can also run bin/console for an interactive prompt that will allow you to experiment.

To install this gem onto your local machine, run bundle exec rake install. To release a new version, update the version number in version.rb, and then run bundle exec rake release, which will create a git tag for the version, push git commits and the created tag, and push the .gem file to rubygems.org.

Bug reports and pull requests are welcome on GitHub at https://github.com/wouterken/crystalruby. This project is intended to be a safe, welcoming space for collaboration, and contributors are expected to adhere to the code of conduct.

The gem is available as open source under the terms of the MIT License.

Everyone interacting in the crystalruby project's codebases, issue trackers, chat rooms and mailing lists is expected to follow the code of conduct.