Ruby 3.1

  • Released at: Dec 25, 2021 ( file)
  • Status (as of Nov 08, 2023): 3.1.4 is current stable
  • This document first published: Jan 5, 2022
  • Last change to this document: Nov 08, 2023

Note: As already explained in Introduction, this site is dedicated to changes in the language, not the implementation, therefore the list below lacks mentions of lots of important optimization introduced in 3.1, including a new JIT named YJIT. That’s not because they are not important, just because this site’s goals are different.


The first major release after 3.0, Ruby 3.1 mostly dedicated to stabilizing new features like concurrency and pattern-matching, but also introduces a bunch of new features.

See also my this year’s ongoing post series on the making of the changelog and Ruby development process.

Language changes

Values in Hash literals and keyword arguments can be omitted

In hash literals and method calls, x: is now a shortcut for x: x—take hash value from the context, with the same name as a key.

  • Discussion: Feature #14579
  • Documentation: doc/syntax/literals.rdoc#Hash Literals, doc/methods.rdoc#Keyword Arguments, Hash
  • Code:
    x, y = 100, 200
    # => [100, 200]
    # In hash literal
    h = {x:, y:}
    # => {:x=>100, :y=>200}
    # In method call
    p(x:, y:)
    # prints: "{:x=>100, :y=>200}"
    # Any context-available names work too, not only local variables:
    C = 'constant'
    # => "constant"
    def m()='method'
    # => :m
    {C:, m:}
    # => {:C=>"constant", :m=>"method"}
    # Undefined names behave like in any other context:
    # => undefined local variable or method `z' (NameError)
    # Dynamic symbols are not supported:
    name = 'x'
    #          ^
    # syntax error, unexpected '}' (SyntaxError)
    # ...including symbols produced from literal strings:
    #    ^
    # syntax error, unexpected '}' (SyntaxError)
  • Notes: Be aware, that using new feature in a method call without parenthesis might lead to unexpected effect:
    # Useful "puts-debugging" technique: print variable values with names
    p(x:, y:)
    # ...but if you'll write it this way:
    p x:, y:
    x + y # ... some following code
    # It would be interpreted as this, for backward compatibility:
    p(x:, y: x + y)

    Therefore, if you use value omission in method calls, always use parenthesis!

  • Follow-ups: It is still discussed whether Ruby should handle the problem above (unexpected continuation) more gracefully. It is possible that the behavior would change in a migration-friendly manner: first, the warning in 3.2 about key: at the end of the line; then, in 3.3, next line would be considered a separate statement.

Inside “endless” method definitions, method calls without parenthesis are allowed

def some_method = other_method "bar" is now allowed, previously it was required to write other_method("bar") in the method’s body.

  • Reason: When endless methods were introduced in 3.0, this syntax (named “command syntax” internally) wasn’t allowed; but it was not a deliberate decision, and rather a side-effect of a parser complexity. During development of 3.1, the workaround was found to make code more consistent and refactoring from multi-line to single-line methods easier.
  • Discussion: Feature #17398
  • Documentation: — (doc/syntax/methods.rdoc doesn’t mention new or old behavior.)
  • Code:
    def log(msg) = puts "#{}: #{msg}"
    # 3.0: syntax error, unexpected string literal, expecting `do' or '{' or '('
    # 3.1: successfully defined

Anonymous block argument

If method uses its block argument only to pass to another method, it can be marked by anonymous &.

  • Reason: The initial proposal for the feature is 6-year old and focused on avoiding intermediate blocks object allocation on block forwarding. It was considered redundant when block forwarding was optimized in Ruby 2.5; but then the core team decided it is actually a nice and unambiguous shortcut for methods that just pass the block further. As block argument is frequently called just block, the absence of the name doesn’t affect readability.
  • Discussion: Feature #11256
  • Documentation: doc/syntax/methods.rdoc#Block Argument
  • Code:
    def logged_open(filename, &)
      puts "Opening #{filename}...", &)
    logged_open('') { puts }
    # Prints:
    #  Opening
    #  ...first 100 chars of
    # Attempt to use `&` inside a method without block parameter is an error:
    def filter(data)=
    # => no anonymous block parameter (SyntaxError)
    # ...and with named block parameter too
    def filter(data, &block)= data.filter(&)
    # => no anonymous block parameter (SyntaxError)
    # If you temporary want to catch a block in the method, `Kernel#proc` will help:
    def filter(data, &)
      block = proc(&)
      p block,
    # => :filter
    filter(1..10, &:odd?)
    # Prints:
    #  #<Proc:0x00007fe9842fa308(&:odd?) (lambda)>
    #  true
    # => [1, 3, 5, 7, 9]


  • One-line pattern matching (both versions) is no longer experimental

Expressions and non-local variables allowed in pin operator ^

On the initial implementation of the pattern matching, the only patterns that could’ve been used in matching expression were constants, literals and pinned local variables. Since 3.1, any kind of statement can be a pattern, with extended functionality of the ^ pin operator.

  • Discussion: Feature #17411 (pinning expressions), Feature #17724 (pinning class/global variables)
  • Documentation: doc/syntax/pattern_matching.rdoc#Variable pinning
  • Code:
    # In Ruby 3.0, this was possible:
    {version: 12} in {version: 10..15} # range literal as a pattern
    # => true
    # But this was not, despite being intuitively similar:
    {timestamp:} in {timestamp: (}
    # syntax error
    # In 3.1, this is possible with pinning:
    {timestamp:} in {timestamp: ^(}
    # => true
    # Any complex expression is possible
    # ...consider years.txt has lines like "2021-2025", some ranges of years. in ^('years.txt').lines.first.then { _1.split('-') }.then { _1.to_i.._2.to_i })
    # parenthesis are mandatory, this wouldn't work: in ^rand(2021..2023)
    # syntax error

Parentheses can be omitted in one-line pattern matching.

  • Discussion: Feature #16182
  • Documentation: doc/syntax/pattern_matching.rdoc#Patterns
  • Code:
    [0, 1] => _, x
    # 3.0:
    #   syntax error, unexpected ',', expecting `end'
    #   [0, 1] => _, x
    #              ^
    # 3.1: success:
    x #=> 1
    {y: 2} => y:
    # 3.0
    #   syntax error, unexpected label)
    #   {y: 2} => y:
    #             ^~
    # 3.1: success:
    y #=> 2
    # Works with in: syntax, too:
    if {language: 'Ruby', version: '3.1'} in version: ('3.0'..)
      puts "Pattern-matching is cool"
    # 3.0: syntax error
    # 3.1: Prints "Pattern-matching is cool"

Multiple assignment evaluation order change

  • Reason: Multiple assignment evaluation order was unexpected: the right part of the expression was calculated before the left. While rarely crucial, this still could cause subtle bugs.
  • Discussion: Bug #4443
  • Documentation:
  • Code:
    def left1
      puts "left1"
    def left2
      puts "left2"
    def right1
      puts "right1"
    def right2
      puts "right2"
    left1[0] = right1
    # Both Ruby 3.0 and 3.1: prints left1, right1 -- the left side calculated first
    left1[0], left2[0] = right1, right2
    # 3.0: prints right1, right2, left1, left2 -- right side calculated first
    # 3.1: prints left1, left2, right1, right2 -- consistent with singular assignment

Core classes and modules

Class and Module


Returns all direct subclasses of the class.

  • Discussion: Feature #18273
  • Documentation: Class#subclasses
  • Code:
    class A
    class B < A
    class C < B
    A.subclasses #=> [B] -- only direct descendants, no grand-children
    class << A
    A.subclasses #=> [B] -- singleton class is not included either
    # Dynamic subclasses are included:
    c =
    # => [#<Class:0x00007f3e45907858>, B]
    # Note that `.subclasses` doesn't hold references to the descendants, and therefore doesn't prohibit
    # them from being garbage-collected
    c = nil
    # => [B] -- dynamic class is gone
  • Notes: Almost till the final release, method #descendants (returning all descendants, not only direct one) was also been introduced. But then Matz had a change of mind and decided #subclasses covers most of the reasonable needs, and #descendants introduces too much complexity. It still might be considered for the future versions.

Refinement class

The class of the context (self) available inside the refine SomeClass do ... end, and a new private method: Refinement#import_methods

  • Reason: The include/extend in refinements never worked (refined class wouldn’t have the modules included) but were allowed. import_methods were introduced as a compromise replacement, limited in functionality (see Notes).
  • Discussion: Bug #17429
  • Documentation: Refinement
  • Code:
    module AsJSON
      def as_json_data
        JSON.parse(to_json, symbolize_names: true)
    Warning[:deprecated] = true # To see the warning below
    module JSONizer
      refine Array do
        include AsJSON
        # warning: Refinement#include is deprecated and will be removed in Ruby 3.2
        import_methods AsJSON
      refine Hash do
        import_methods AsJSON
    using JSONizer
    [:bar, /baz/,].as_json_data
    # => ["bar", "(?-mix:baz)", "2021-12-25 20:47:29 +0200"]
    {'value' => 1/3r}.as_json_data
    # => {:value=>"1/3"}
  • Notes: Unlike module including, import_methods works by copying the methods bytecode to refined objects. Thus only methods defined in Ruby can be imported (because methods defined in C—including core methods—don’t have bytecode). This somewhat limits the usability of the feature:
    module Binread
      # try to temporarily make string into a byte-sequence...
      refine String do
        def each(&)
        import_methods Enumerable
        # Can't import method which is not defined with Ruby code: Enumerable#drop (ArgumentError)
  • Follow-ups: In Ruby 3.2, several usability improvments using the new class were introduced: ability to ask which refinements some module defines, which refinements are active in the current context, and what class or module the particular refinement refines.

Module#prepend behavior change

Now, when module is prepended to the class, it always becomes first in the ancestor chain; even if it was already included.

  • Reason: It was noted that prepend is a mechanism of method redefinition that is relied upon to be performed strictly before the module of the class affected; but if the class had the module already ws included (especially in complex systems and indirectly), this becomes unreliable.
  • Discussion: Bug #17423
  • Documentation: Module#prepend (not that it actually documents new—or old—behavior)
  • Code:
    module M
      def foo
        ["In M", *(super rescue nil)]
    class A
      def foo
        ["in A", *(super rescue nil)]
    A.include M
    A.prepend M
    # 3.0: => [A, M, Object, Kernel, BasicObject]
    # 3.1: => [M, A, M, Object, Kernel, BasicObject]
    # 3.0: => ["in A", "In M"] -- only method from included module is executed
    # 3.1: => ["In M", "in A", "In M"] -- both included AND prepended methods are executed
    # Note that relation is not symmetric: if the `prepend` happened first, `include` wouldn't affect
    # ancestors chain at all:
    class B
      def foo
        ["in B", *(super rescue nil)]
    B.prepend M
    B.include M
    # => [M, B, Object, Kernel, BasicObject] -- both in 3.0 and 3.1
    # => ["In M", "in B"]

Module#private, #public, #protected, and #module_function return their arguments

  • Reason: Allows to chain method decoration DSLs, like memoize private :method_name. Follows similar in spirit changes of attr_reader/attr_writer behavior in Ruby 3.0
  • Discussion: Feature #12495
  • Documentation: Module#private, Module#public, Module#protected, Module#module_function
  • Code:
    p private # global private making all the method below private
    # => nil
    p private def m; end
    # => :m
    def m1; end
    def m2; end
    p private :m1, :m2
    # => :m1, :m2
    # Usage:
    require 'memoist'
    extend Memoist
    # This works: because memoize returns :m, and private receives it
    private memoize def m
      # ...
    # In 3.0 this didn't work, but in 3.1, this does:
    memoize private def m
      # ...

Method/UnboundMethod: #public?, #private?, #protected?

  • Reason: For exploration, documenting, and metaprogramming (like, defining a wrapper method that would have the same visibility as original method).
  • Discussion: Feature #11689
  • Documentation: Method#private?, UnboundMethod#private?
  • Code:
    class A
      def m1
      private def m2
    m1 = A.instance_method(:m1)
    # => #<UnboundMethod: A#m1() >
    # => false
    # => true
    m2 =
    # => #<Method: A#m2()>
    # => true
  • Notes: Another idea discussed was just having Method#visibility, but it was hard to decide on the method’s name and what it should return, while three predicates are more or less obvious.
  • Follow-up: Reverted in Ruby 3.2: it was considered after all that visibility is not a property of the method, but rather the responsibility of its owner.

Kernel#load: module as a second argument

load(filename, module) now evaluates filename’s code in the context of the module.

  • Reason: The long-requested feature (the request is created almost 10 years ago!) might be useful for implementing simple helper libraries with a few unscoped methods/classes, and then scoping them while loading.
  • Discussion: Feature #6210
  • Documentation: Kernel#load
  • Code:
    # Assuming the following file test-load.rb
    ENV={a: 1}
    def puts(*)
      raise NotImplementedError
    class String
      def debug = p("Hello from test string")
    print "Additional module loaded"
    # This would replace `ENV`, `puts` and extend core `String`, and pollute global scope with methods
    load 'test-load.rb'
    # prints "Additional module loaded"
    # => {:a=>1}
    # prints "Hello from test string"
    puts 'bar'
    # NotImplementedError
    # This would not affect `ENV`, `puts`, and `String`, but the only use of the call would be for side-effects
    load 'test-load.rb', true
    # prints "Additional module loaded"
    # => ...system ENV...
    # undefined method `debug' for "test":String
    puts 'bar'
    # ...prints as usual
    # 3.1+
    # This would add `ENV`, `puts` and `String` to `Context`, and will make methods available as `foo`
    module M; end
    load 'test-load.rb', M
    # prints "Additional module loaded"
    # => ...system ENV...
    # => {:a=>1}
    # prints "Hello from test string"
    # => #<UnboundMethod: M#puts() test-load.rb:3> -- can be used by including M into some other module/class
    # Note that `load` wouldn't load into class:
    class C; end
    load 'test-load.rb', C # the call is successful, but C treated as just truthy value (wrap=true)
    # uninitialized constant C::ENV (NameError)
  • Notes: Scoping into a module is transitive for other load calls: if the loaded file loads more files, they also will be scoped into the module. But it is not transitive for require:
    # assuming in test-load2.rb
    class Integer
      def debug = p "debug Int"
    # ...and in test-load.rb
    load 'test-load2.rb'
    # then if we do
    module M; end
    load 'test-load.rb', M
    # prints "debug Int"
    # ...but if in test-load.rb we'll have this:
    require './test-load2'
    # ...then
    load 'test-load.rb', M
    # uninitialized constant M::Integer (NameError)
    # prints "debug Int"


Safe explicit conversion methods, which calls #to_int if the argument has it, and returns nil otherwise.

  • Reason: The similar methods exist on a few other core classes, like String and Array. They are useful for implementing polymorphic methods (like #[] behaving differently for integer, string and range), as a safe way of type-checking/type-coercion.
  • Discussion: Feature #15211
  • Documentation: Integer.try_convert
  • Code:
    Integer.try_convert(1) #=> 1
    Integer.try_convert(1.3) #=> 1
    Integer.try_convert(3/2r) #=> 1
    require 'bigdecimal'
    #=> 100000000000000000000000000000000000000000000000000000000000000
    # If the object doesn't have #to_int, returns `nil`
    Integer.try_convert('1') #=> nil
    class MyNumber
      def to_int
    # If #to_int doesn't return Integer, the error is raised (even if the result is convertable)
    #=> `try_convert': can't convert MyNumber to Integer (MyNumber#to_int gives Float) (TypeError)
    # Usage in some array-alike slicing API:
    def slice(index)
      idx = Integer.try_convert(index) or
        raise ArgumentError, "Expected argument convertible to Integer, got #{index}"
      # Now `idx` is safe to use even for low-level APIs like memory addressing, it is definitely
      # Integer, even if some Float was passed initially.

String#unpack and #unpack1: offset: argument

The new argument allows to unpack data starting from the specified offset.

  • Reason: Handy for working with binary data of complicated shape, flow like: read one byte, decide what type/size is next data, then read this number of bytes starting after the type tag, then read next chunk starting after the data already read. Before the parameter introduction, this type of processing required string slicing, which is cumbersome and might’ve been inefficient.
  • Discussion: Feature #18254
  • Documentation: String#unpack, String#unpack1
  • Code:
    # imagine it is "type tag 1 + data in four 32-bit integers"
    data = [1, 1991, 1995, 1998, 2003].pack('CL*')
    # => "\x01\xC7\a\x00\x00\xCB\a\x00\x00\xCE\a\x00\x00\xD3\a\x00\x00"
    tag = data.unpack1('c')
    # => 1
    data.unpack('L*', offset: 1)
    # => [1991, 1995, 1998, 2003]
    # Negative offsets or out-of-bounds ones don't work
    data.unpack('L*', offset: -1)
    # in `unpack': offset can't be negative (ArgumentError)
    data.unpack('L*', offset: 20)
    # in `unpack':  offset outside of string (ArgumentError)
    # 0 works, being the same as no offset
    data.unpack('c', offset: 0)
    # => [1]

MatchData#match and #match_length

#match(idx) is the same as MatchData#[], but only works for singular numeric or symbolic index. #match_length(idx) returns the length of the idxs match.

  • Reason: It was argued that in the performance-critical code sometimes there is a need to just know the length of the substring matched, and doing it via match[index].length leads to creation of an intermediate string soon to be dropped, while #match_length doesn’t do that. #match is added mostly for completeness (it would be weird to not have it when #match_length method exists).
  • Discussion: Feature #18172
  • Documentation: MatchData#match, MatchData#match_length
  • Code:
    m = "Ruby 3.1 is released December 25, 2021".match(/Ruby (?<version>\d\.\d) .* (?<year>\d{4})/)
    # => #<MatchData "Ruby 3.1 is released December 25, 2021" version:"3.1" year:"2021">
    # => "3.1"
    # => "3.1"
    # => "3.1"
    # => 3
    # => "2021"
    # => 4
    # in `match': index 3 out of matches (IndexError)
    # in `match_length': index 3 out of matches (IndexError)
    # ...unlike #[]
    # => nil
    # Realistic usage:
    text = "My credit card number is 5221-1234-5678-9000 (not really)"
    text.gsub!(/(\d{4})-(\d{4})-(\d{4})-(\d{4})/) {
      m = Regexp.last_match
      "#{m.match(1)}-#{'*' * m.match_length(2)}-#{'*' * m.match_length(3)}-#{m.match(4)}"
     # => "My credit card number is 5221-****-****-9000 (not really)"


in: parameter for constructing time

The new parameter is accepting offsets or timezone objects, and (finally!) allows to easily construct time in desired timezone.

  • Discussion: Feature #17485
  • Documentation:,,
  • Code:, 12, 24, in: '+09:00')
    # => 2021-12-24 00:00:00 +0900, 12, 24, in: 'UTC')
    #  => 2021-12-24 00:00:00 UTC
    # This was previously possible only with "pass all params" signature
    # Ruby 3.0: tz was only a positional argument:, 12, 24, 0, 0, 0, '+09:00')
    # => 2021-12-24 00:00:00 +0900
    # This was previously not possible to do in one call: ::now with timezone '+09:00')
    # => 2021-12-21 18:33:51.583436226 +0900
    # => 2021-12-21 11:34:00.258235705 +0200 -- my local timezone, for comparison
    # This was not possible either: ::at with timezone
    timestamp =
    # => 1640079354, in: '+09:00')
    # => 2021-12-21 18:35:54 +0900
    # Timezone objects are supported, too
    require 'tzinfo'
    zone = TZInfo::Timezone.get('America/New_York') zone)
    # => 2021-12-21 04:37:25.522169768 -0500 zone).zone
    # => #<TZInfo::DataTimezone: America/New_York>

#strftime supports -00:00 offset

%-z allows to format “negative zero” offset for Time.

  • Reason: RFC 3339 defines a nuance of offset formatting: offset "+00:00" in some service response declares that UTC is the preferred timezone for some context; while "-00:00" declares that preferred timezone is unknown, that’s why data is in UTC.
  • Discussion: Feature #17544
  • Documentation: Time#strftime
  • Code:
    t =
    # => 2021-12-26 18:47:52.979940556 UTC -- yeah, I am still working on a changelog a day after the release :(
    # => "18:47+0000"
    # => "18:47-0000"
    # => "18:47-00:00"
    # => "18:47-00:00:00"


Warning on passing keywords to a non-keyword-initialized struct

  • Reason: After keyword-inited stucts have been introduced, it is easy to forget the actual protocol of some struct, and try to initialized it with keyword arguments—which, if the struct is not keyword-inited, would raise no error, but produce an unexpected result.
  • Discussion: Feature #16806
  • Code:
    Person =, :age)
    # Proper initialization:'Jane', 32)
    #=> #<struct Person name="Jane", age=32>
    # I forgot it is not initialized by keywords, expected to initialize both fields,
    # but it would actually be treated as{name: 'Jane', age: 32}, nil) 'Jane', age: 32)
    # warning: Passing only keyword arguments to Struct#initialize will behave differently from Ruby 3.2. Please use a Hash literal like .new({k: v}) instead of .new(k: v).
    #=> #<struct Person name={:name=>"Jane", :age=>32}, age=nil>
    # If the struct is really initialized by a hash, use {} to signify you know what you do
    Locale ={user: 'користувач', application: 'застосунок'})
    # => #<struct Locale translations={:user=>"користувач", :application=>"застосунок"}>
    # warning produced
    # The warning is produced only if the hash is the ONLY argument
    Locale =, :translations), user: 'користувач', application: 'застосунок')
    #=> #<struct Locale lang=:ua, translations={:user=>"користувач", :application=>"застосунок"}>
    # braces necessary, no warning
  • Follow-up: Since 3.2, following the warning, the structs can be initialized by both keyword and positional argument when keyword_init: true is not defined.


The method allows to check whether the Struct-produced class is initialized by keywords.

  • Reason: Helpful for meta-programming and generic processing of data, like serializing/deserializing.
  • Discussion: Feature #18008
  • Documentation: Struct.keyword_init?
  • Code:, :age).keyword_init?                       #=> nil, :age, keyword_init: true).keyword_init?   #=> true, :age, keyword_init: false).keyword_init?  #=> false
    # Potential usage:
    def deserialize(struct_class, **data)
      if struct_class.keyword_init?**data)

Collections and sequences

Enumerable#compact and Enumerator::Lazy#compact

  • Reason: #compact was previously defined only on a particular classes like Array and Hash, but being a synonym for reject(&:nil?), it is reasonable to have in generic Enumerable. Especially useful in lazy enumeration to produce still-lazy sequences with compact applied.
  • Discussion: Feature #17312
  • Documentation: Enumerable#compact, Enumerator::Lazy#compact
  • Code:
    # quickly imitate some slow or infinite stream producing data with nils possible:
    stream = Enumerator.produce { [nil, 1, 'test'].sample }.lazy
    # 3.0: NoMethodError (undefined method `compact' for #<Enumerator::Lazy: #<Enumerator: #<Enumerator::Producer:0x000055e8ffd38388>:each>>)
    # 3.1: => ["test", 1, 1, "test", "test"]
    # In 3.0, this already worked, which made absence of .compact even more confusing:
    # => ["test", "test", 1, "test", 1]
    # Usage in regular enumerable objects:
    require 'set'
    Set[1, nil, 2].compact
    # 3.0: NoMethodError (undefined method `compact' for #<Set: {1, nil, 2}>)
    # 3.1: => [1, 2] -- not ideal (it would be better to see Set[1, 2]), but better
  • Notes: Hash, despite being Enumerable, has its own [#compact] method definition since Ruby 2.4, which just removes all key-value pairs where values are nil.

Enumerable#tally now accepts an optional hash to count

The hash is updated with the new tally numbers, allowing to accumulate stats from several enumerables efficiently and succinctly.

  • Discussion: Feature #17744
  • Documentation: Enumerable#tally
  • Code:
    week1_shifts = ['John', 'Jane', 'Hassim', 'John', 'Olexa', 'Olexa', 'Jane']
    week2_shifts = ['Jane', 'Jane', 'Miranda', 'John', 'Hassim', 'Olexa', 'John']
    stats = {}
    #=> {"John"=>2, "Jane"=>2, "Hassim"=>1, "Olexa"=>2}
    week2_shifts.tally(stats) # add it to the previous stats
    #=> {"John"=>4, "Jane"=>4, "Hassim"=>2, "Olexa"=>3, "Miranda"=>1}
    # Only hashes or objects implicitly convertible into hash are allowed:
    require 'ostruct'
    week1_shifts.tally( # maybe having #[]= is enough?..
    # in `tally': can't convert OpenStruct to Hash (OpenStruct#to_hash gives NilClass) (TypeError)
    # Hash descendants can be used as argument
    gem 'hashie'
    require 'hashie'
    stats = # Subclass of Hash which allows `.key` access
    res = week1_shifts.tally(stats)
    # => {"John"=>2, "Jane"=>2, "Hassim"=>1, "Olexa"=>2}
    # => Hashie::Mash
    # => 2

Enumerable#each_cons and #each_slice return a receiver

  • Reason: Before 3.1, both methods returned nil, making it inconvenient to use in chained computations. This was also inconsistent with #each, which returns a receiver.
  • Discussion: [GH-1509]
  • Documentation: Enumerable#each_cons, Enumerable#each_slice
  • Code:
    Item =, :next)
    list = ('a'..'e').map { }
                     .each_cons(2) { |item, next_item| = next_item }
    # => [
    #   #<struct Item value="a", next=#<struct Item value="b", ...>,
    #   ...
    #   ...
    # Before Ruby 3.1, one would need to write:
    list = ('a'..'e').map { }
    # this would return nil, so can't be chained
    list.each_cons(2) { |item, next_item| = next_item }


Method checks if two arrays intersect, e.g. have at least one item in common.

  • Reason: Checking intersection is typical operation (is “current user roles intersect with admin roles?”), but without this method, it typically required to write !(ary1 & ary2).empty?, which is cumbersome and produces an intermediate array.
  • Discussion: Feature #15198
  • Documentation: Array#intersect?
  • Code:
    [1, 2, 3, 4].intersect?([1, 11, 111]) #=> true
    [1, 2, 3, 4].intersect?([11, 111]) #=> false
    # As with other set-alike method, items comparison are performed with #eql? which is in some cases
    # stricter than #==
    [1] == [1.0]
    # => true
    # => false
  • Notes:
    • As it is with &, the new method treats arrays as sets of values with order ignored, so the semantic is just “if any value of array 1 is present in array 2”, not some more strict check (like “if elements at the end of array 1 are the same as elements at the beginning of array 2”).
    • The complexity of the method is roughly O(size1+size2).

Thread::Queue#initialize: initial values can be passed to initializer

  • Reason: Despite the fact that Thread::Queue is a specialized collection for safe data exchange in a multi-threaded context, it is considered to frequently have some initial set of values to process (like list of workers), new method call-sequence simplifies that.
  • Discussion: Feature #17327
  • Documentation:
  • Code:
    queue =['task1', 'task2', 'task3'])
    # => #<Thread::Queue:0x00007f434de1b800>
    producer = do
      while !queue.empty?
        task = queue.pop
        puts "procesing #{task}"
    # procesing task1
    # procesing task2
    # procesing task3
    # Any class responding to #to_a, and any Enumerable can be used as a constructor:
    queue ='tasks.csv').each_line)
    # => #<Thread::Queue:0x00007f434de18998>


A new class representing low-level I/O abstraction. Internally, uses OS mechanisms to directly use memory-mapping of files and internal Ruby String representation.

  • Reason: The feature was requested for implementing non-blocking #io_read and #io_write for FiberScheduler, but probably will find other usages once matured.
  • Discussion: Feature #18020
  • Documentation: IO::Buffer
  • Code: This is a big new class, see class’ docs for detailed examples of usage, they are quite succinct.

File.dirname: optional level to go up the directory tree

  • Discussion: Feature #12194
  • Documentation: File.dirname
  • Code:
    path = "/home/zverok/projects/ruby/"
    # => "/home/zverok/projects/ruby"
    File.dirname(path, 1) # same as above
    # => "/home/zverok/projects/ruby"
    File.dirname(path, 2)
    # => "/home/zverok/projects"
    File.dirname(path, 3)
    # => "/home/zverok"
    File.dirname(path, 100)
    # => "/"
    File.dirname(path, 0) # no-op
    # => "/home/zverok/projects/ruby/"
    File.dirname(path, -1)
    # in `dirname': negative level: -1 (ArgumentError)
  • Notes: Unfortunately, Pathname#dirname is not updated to have a new argument.

Marshal.load accepts a freeze: option

If the freeze: true is passed, #load will return deeply frozen object graph.

  • Reason: Besides obvious immutability merits, constructing frozen graph may lead to lower memory consumption, because frozen strings are deduplicated.
  • Discussion: Feature #18148
  • Documentation: Marshal.load
  • Code:
    serialized = Marshal.dump(['value1', 'value2', 'value1', 'value2'])
    deserialized = Marshal.load(serialized)
    # => [false, false, false, false]
    # => [1023900, 1023920, 1023940, 1023960] -- 4 different objects
    deserialized = Marshal.load(serialized, freeze: true)
    # => [true, true, true, true]
    # => [1039360, 1039380, 1039360, 1039380] -- only 2 different objects, object_ids repeating
    # The only type of objects not frozen are class/module:
    MyType =, :b)
    serialized = Marshal.dump(MyType)
    deserialized = Marshal.load(serialized, freeze: true)
    deserialized.frozen? # => false
    # That's because Marshal.load doesn't "create" those objects, just finds the correspondence:
    Object.send(:remove_const, :MyType)
    # undefined class/module MyType (ArgumentError)
    # ...therefore, making that corresponding module frozen would be unexpected



Returns OS-specific thread ID.

  • Reason: The method might be useful for low-level debugging, and for system calls requiring “real” thread id.
  • Discussion: Feature #17853
  • Documentation: Thread#native_thread_id
  • Code:
    # On my Ubuntu:
    # => 2735439
    t = { sleep(1) }
    # => 2746533
    sleep(5) # -- allow t to finish (non-deterministic though!)
    # => nil
  • Notes: It is documented that native_thread_id might change through the thread lifetime in Ruby implementations that use M:N threading model (e.g. M Ruby threads is mapped on N OS threads, with the nature of mapping decided by the interpreter).


The new method is the process forking implementation which is guaranteed to be used by both Process.fork, Kernel#fork, and IO.popen.

  • Reason: The method is meant to be overridable by libraries that need to add some hook/additional processing (like freeing DB connections) on forking. Previously, this was inconvenient due to the fact that Process.fork and Kernel#fork were different methods, not relying on each other.
  • Discussion: Feature #17795
  • Documentation: Process.\_fork
  • Code:
    module ForkMonitor
      def _fork(...)
        puts "Before fork : in #{}"
        super.tap { puts "After fork  : in #{}" }
    puts "Main process: in #{}"
    pid = fork { puts "Forked body : in #{}" }
    puts "Wating for       #{pid}"
    Process.waitpid(pid) if pid

    This prints:

    Main process: in 2052974
    Before fork : in 2052974
    After fork  : in 2052974
    Wating for       2052975
    After fork  : in 2052975  -- note that AFTER code is performed both for parent and child
    Forked body : in 2052975
  • Notes:
    • The method should never be called directly; the only reason for it to exist is to be redefinable
    • The method should call super, and return what it had returned (it would become a return value of fork, e.g. child process’ pid)

Fiber Scheduler: New hooks

Continuing work on non-blocking Fiber and Fiber Scheduler concept introduced in 3.0, new optional hooks were introduced to support non-blocking DNS resolution (#address_resolve), IO (#io_read/#io_write)

Ractors can access module instance variables

Ractors, introduced in 3.0 to solve a problem of GVM (global virtual matchine lock), have a strict notion of “shareability”. Non-main Ractors can’t directly access any names defined outside of them except for classes/modules and frozen constants. One improvement was made to this model: allow to access module-wide instance variables.

  • Reason: Calling SomeClass.some_method from inside Ractor is generally OK, unless the method changes the class’/module’s internal variables (for example maintains a global cache of all objects constructed)—in this case you’ll see an IsolationError. So far, it corresponds to expectations of Ractor’s data sharing model. But this expectation is broken if some of the module’s methods uses immutable data stored in instance variable: intuitively it doesn’t break isolation, but still didn’t work. Now this limitation is relaxed.
  • Discussion: Feature #17592
  • Documentation: Ractor#Shareable and unshareable objects (docs aren’t updated yet)
  • Code:
    class APIClient <, :use_https, keyword_init: true)
      def self.setup!(**defaults)
        @defaults = defaults.freeze
      def self.make
    APIClient.setup!(version: 1, use_https: true) # sets immutable instance variable @defaults {
      p APIClient.make # uses that variable, should be safe
      # 3.0: in `make': can not access instance variables of classes/modules from non-main Ractors (Ractor::IsolationError)
      # 3.1: => #<struct APIClient version=1, use_https=true>



Returns value set by --backtrace-limit command-line option (which is used in Exception#full_message to format backtrace).

  • Reason: The ability to read the parameter might be useful for custom error formatters.
  • Discussion: Feature #17479
  • Documentation: Thread::Backtrace.limit
  • Code: (in shell)
    $ ruby -r net/http -e "p Thread::Backtrace.limit; Net::HTTP.get(URI('http://wrong.address'))"
    .../lib/ruby/3.1.0/socket.rb:227:in `getaddrinfo': Failed to open TCP connection to wrong.address:80 (getaddrinfo: Name or service not known) (SocketError)
        from .../lib/ruby/3.1.0/socket.rb:227:in `foreach'
        from .../lib/ruby/3.1.0/socket.rb:632:in `tcp'
        from .../lib/ruby/3.1.0/net/http.rb:998:in `connect'
        from .../lib/ruby/3.1.0/net/http.rb:976:in `do_start'
        from .../lib/ruby/3.1.0/net/http.rb:965:in `start'
        from .../lib/ruby/3.1.0/net/http.rb:627:in `start'
        from .../lib/ruby/3.1.0/net/http.rb:503:in `get_response'
        from .../lib/ruby/3.1.0/net/http.rb:474:in `get'
    .../lib/ruby/3.1.0/socket.rb:227:in `getaddrinfo': getaddrinfo: Name or service not known (SocketError)
        from .../lib/ruby/3.1.0/socket.rb:227:in `foreach'
        from .../lib/ruby/3.1.0/socket.rb:632:in `tcp'
        from .../lib/ruby/3.1.0/net/http.rb:998:in `connect'
        from .../lib/ruby/3.1.0/net/http.rb:976:in `do_start'
        from .../lib/ruby/3.1.0/net/http.rb:965:in `start'
        from .../lib/ruby/3.1.0/net/http.rb:627:in `start'
        from .../lib/ruby/3.1.0/net/http.rb:503:in `get_response'
        from .../lib/ruby/3.1.0/net/http.rb:474:in `get'
        from -e:1:in `<main>'
    $ ruby --backtrace-limit 2 -r net/http -e "p Thread::Backtrace.limit; Net::HTTP.get(URI('http://wrong.address'))"
    .../lib/ruby/3.1.0/socket.rb:227:in `getaddrinfo': Failed to open TCP connection to wrong.address:80 (getaddrinfo: Name or service not known) (SocketError)
        from .../lib/ruby/3.1.0/socket.rb:227:in `foreach'
        from .../lib/ruby/3.1.0/socket.rb:632:in `tcp'
         ... 7 levels...
    .../lib/ruby/3.1.0/socket.rb:227:in `getaddrinfo': getaddrinfo: Name or service not known (SocketError)
        from .../lib/ruby/3.1.0/socket.rb:227:in `foreach'
        from .../lib/ruby/3.1.0/socket.rb:632:in `tcp'
         ... 7 levels...
    $ ruby --backtrace-limit 0 -r net/http -e "p Thread::Backtrace.limit; Net::HTTP.get(URI('http://wrong.address'))"
    .../lib/ruby/3.1.0/socket.rb:227:in `getaddrinfo': Failed to open TCP connection to wrong.address:80 (getaddrinfo: Name or service not known) (SocketError)
         ... 9 levels...
    .../lib/ruby/3.1.0/socket.rb:227:in `getaddrinfo': getaddrinfo: Name or service not known (SocketError)
         ... 9 levels...

GC: measuring total time

New methods added to measure time spent in garbage collection.

  • Discussion: Feature #10917
  • Documentation: GC.measure_total_time, GC.measure_total_time=, GC.stat, GC.total_time
  • Code:
    GC.total_time  # => 10903796 -- nanoseconds at script start
    GC.stat[:time] # => 10 -- milliseconds at script start { }
    # After some GCing
    GC.total_time  # => 20483486 -- nanoseconds
    GC.stat[:time] # => 20 -- milliseconds
    GC.measure_total_time= false { }
    # Measuring was disabled, same values as above
    GC.total_time  # => 20483486
    GC.stat[:time] # => 20
  • Notes: It is stated that measurement can introduce overhead, but it is enabled by default.


Normally, during handling of TracePoint event, all other active TracePoint handlers wouldn’t be called. Since Ruby 3.1, the handler can wrap some of its code into TracePoint.allow_reentry { ... } and inside that block, other handlers (including the current one) will be invoked.

  • Reason: The issue was raised by author of the byebug debugger. Debugging sessions are implemented with TracePoint API, and when the program is inside a debugging session, it is actually inside one of the handlers. Thus, all other libraries relying on TracePoint API would not work during the session—and one of them is zeitwerk, Rails code reloader.
  • Discussion: Feature #15912
  • Documentation: TracePoint.allow_reentry
  • Code:
    # Without reentry
    # ---------------
    line_handler = do |tp| # invoked for each statement
      next if tp.path != __FILE__ # only work in this file
      puts "Line handler"
      binding.eval("class C; end")
    class_handler = do |tp| # invoked for each class definition
      puts "Class handler"
    class B
    # This script will print "Class handler" only once: when inside :line
    # handler, all other handlers are ignored, and `class C; end` is NOT invoking the class_handler
    # With reentry
    # ------------
    line_handler = do |tp| # invoked for each statement
      next if tp.path != __FILE__ # only work in this file
      next if (__LINE__..__LINE__+3).cover?(tp.lineno) # don't be invoked from itself
      puts "Line handler"
      TracePoint.allow_reentry { binding.eval("class C; end") }
    class_handler = do |tp| # invoked for each class definition
      puts "Class handler"
    class B
    # This will print "Class handler" twice: inside allow_reentry block in :line
    # handler, other handlers are enabled.
  • Notes: The default behavior (no reentry) is necessary to protect from infinite recursion; when allowing reentry, extreme precautions should be taken (note the second if inside :line handler: without it, every line inside the handler would invoke itself).

$LOAD_PATH.resolve_feature_path does not raise

Since being added in 2.6 (then named RubyVM.resolve_feature_path), the method have raised when the requested name wouldn’t be found. It was fixed in 3.1 to just return nil.

  • Reason: It was argued that raising LoadError makes method looks like it has a side effect instead of just being a query method, and makes it inconvenient to use.
  • Discussion: Feature #16043
  • Documentation: doc/globals.rdoc
  • Code:
    # 3.0: LoadError (cannot load such file -- garbage111)
    # 3.1: => nil

Standard library

By this release, most of the standard library is extracted to either default or bundled gems; their development happens in separate repositories, and changelogs are either maintained there, or absent altogether. Either way, their changes aren’t mentioned in the combined Ruby changelog, and I’ll not be trying to follow all of them. project has a nice explanations of default and bundled gems concepts, as well as a list of currently gemified libraries and links to their docs.

“For the rest of us” this means libraries development extracted into separate GitHub repositories, and they are just packaged with main Ruby before release. It means you can do issue/PR to any of them independently, without going through more tough development process of the core Ruby.

A few changes to mention, though:

  • Code from securerandom/formatter.rb (services to convert random numbers into various representations) is moved to lib/random/formatter.rb. It was under the securerandom for historical reasons, but defined Random::Formatter and included it into Random instances.
    • Discussion: Feature #18190
    • Docs: Random::Formatter
    • Code:
      Random.hex # NoMethodError: the method is not available in core Random
      # In Ruby 3.0, it required this to work, which was weird:
      require 'securerandom'
      # => "f3a3839a10c84eef96bee3e1c3d0d60a"
      # Since Ruby 3.1:
      require 'random/formatter'
      # => "d270131fead4ca327f64ef6bbcfd7f01"
      # Still works with SecureRandom, too:
      # => "f1945af8640074d186acd1e1c9aa84fc"
  • Coverage#suspend and #resume in built-in coverage measurement library:
  • Psych library (which implements YAML parsing and is required on require 'yaml) switched to safe_load as a default behavior of load. See Bug #17866.
  • IRB console received a large update, and now provides modern experience with autocomplete and hints:
  • Ripper standard parser got updated too (without changing the version) to match new syntaxes, see this blog post for details.

Version updates

Default gems

Bundled gems

Standard library content changes

  • lib/debug.rb is replaced with debug gem
  • new gem error_highlight was introduced and is enabled by default, allowing more fine-grained error messages:
    data = {body: {items: []}}
    data.dig(:body, :items, 0)[:title]
    # Ruby 3.0 prints:
    #   in `<main>': undefined method `[]' for nil:NilClass (NoMethodError)
    # Ruby 3.1 prints:
    #   undefined method `[]' for nil:NilClass (NoMethodError)
    #   data.dig(:body, :items, 0)[:title]
    #                             ^^^^^^^^

Default gems that became bundled

Note: In bundler-enabled environment, some code using this libraries might require adding gem "<gemname>" into your Gemfile, otherwise they wouldn’t be available.