github
/
ruby
зеркало из https://github.com/github/ruby.git
1
0
Форкнуть 0
Код Задачи Пакеты Проекты Релизы Вики Активность
ruby/doc/NEWS/NEWS-3.2.0.md

30 KiB
Исходник Постоянная ссылка Ответственный История

NEWS for Ruby 3.2.0

This document is a list of user-visible feature changes since the 3.1.0 release, except for bug fixes.

Note that each entry is kept to a minimum, see links for details.

Language changes

  • Anonymous rest and keyword rest arguments can now be passed as arguments, instead of just used in method parameters. [Feature #18351]

    def foo(*)
  bar(*)
end
def baz(**)
  quux(**)
end

  • A proc that accepts a single positional argument and keywords will no longer autosplat. [Bug #18633]

    proc{|a, **k| a}.call([1, 2])
# Ruby 3.1 and before
# => 1
# Ruby 3.2 and after
# => [1, 2]

  • Constant assignment evaluation order for constants set on explicit objects has been made consistent with single attribute assignment evaluation order. With this code:

    foo::BAR = baz

    foo is now called before baz. Similarly, for multiple assignments to constants, left-to-right evaluation order is used. With this code:

    foo1::BAR1, foo2::BAR2 = baz1, baz2

    The following evaluation order is now used:

    1. foo1
    2. foo2
    3. baz1
    4. baz2

    [Bug #15928]

  • "Find pattern" is no longer experimental. [Feature #18585]

  • Methods taking a rest parameter (like *args) and wishing to delegate keyword arguments through foo(*args) must now be marked with ruby2_keywords (if not already the case). In other words, all methods wishing to delegate keyword arguments through *args must now be marked with ruby2_keywords, with no exception. This will make it easier to transition to other ways of delegation once a library can require Ruby 3+. Previously, the ruby2_keywords flag was kept if the receiving method took *args, but this was a bug and an inconsistency. A good technique to find the potentially-missing ruby2_keywords is to run the test suite, for where it fails find the last method which must receive keyword arguments, use puts nil, caller, nil there, and check each method/block on the call chain which must delegate keywords is correctly marked as ruby2_keywords. [Bug #18625] [Bug #16466]

    def target(**kw)
end

# Accidentally worked without ruby2_keywords in Ruby 2.7-3.1, ruby2_keywords
# needed in 3.2+. Just like (*args, **kwargs) or (...) would be needed on
# both #foo and #bar when migrating away from ruby2_keywords.
ruby2_keywords def bar(*args)
  target(*args)
end

ruby2_keywords def foo(*args)
  bar(*args)
end

foo(k: 1)

Core classes updates

Note: We're only listing outstanding class updates.

  • Fiber

    • Introduce Fiber.[] and Fiber.[]= for inheritable fiber storage. Introduce Fiber#storage and Fiber#storage= (experimental) for getting and resetting the current storage. Introduce Fiber.new(storage:) for setting the storage when creating a fiber. [Feature #19078]

      Existing Thread and Fiber local variables can be tricky to use. Thread-local variables are shared between all fibers, making it hard to isolate, while Fiber-local variables can be hard to share. It is often desirable to define unit of execution ("execution context") such that some state is shared between all fibers and threads created in that context. This is what Fiber storage provides.

      def log(message)
  puts "#{Fiber[:request_id]}: #{message}"
end

def handle_requests
  while request = read_request
    Fiber.schedule do
      Fiber[:request_id] = SecureRandom.uuid

      request.messages.each do |message|
        Fiber.schedule do
          log("Handling #{message}") # Log includes inherited request_id.
        end
      end
    end
  end
end

      You should generally consider Fiber storage for any state which you want to be shared implicitly between all fibers and threads created in a given context, e.g. a connection pool, a request id, a logger level, environment variables, configuration, etc.

  • Fiber::Scheduler

    • Introduce Fiber::Scheduler#io_select for non-blocking IO.select. [Feature #19060]

  • IO

    • Introduce IO#timeout= and IO#timeout which can cause IO::TimeoutError to be raised if a blocking operation exceeds the specified timeout. [Feature #18630]

      STDIN.timeout = 1
STDIN.read # => Blocking operation timed out! (IO::TimeoutError)

    • Introduce IO.new(..., path:) and promote File#path to IO#path. [Feature #19036]

  • Class

    • Class#attached_object, which returns the object for which the receiver is the singleton class. Raises TypeError if the receiver is not a singleton class. [Feature #12084]

      class Foo; end

Foo.singleton_class.attached_object        #=> Foo
Foo.new.singleton_class.attached_object    #=> #<Foo:0x000000010491a370>
Foo.attached_object                        #=> TypeError: `Foo' is not a singleton class
nil.singleton_class.attached_object        #=> TypeError: `NilClass' is not a singleton class

  • Data

    • New core class to represent simple immutable value object. The class is similar to Struct and partially shares an implementation, but has more lean and strict API. [Feature #16122]

      Measure = Data.define(:amount, :unit)
distance = Measure.new(100, 'km')            #=> #<data Measure amount=100, unit="km">
weight = Measure.new(amount: 50, unit: 'kg') #=> #<data Measure amount=50, unit="kg">
weight.with(amount: 40)                      #=> #<data Measure amount=40, unit="kg">
weight.amount                                #=> 50
weight.amount = 40                           #=> NoMethodError: undefined method `amount='

  • Encoding

    • Encoding#replicate has been deprecated and will be removed in 3.3. [Feature #18949]
    • The dummy Encoding::UTF_16 and Encoding::UTF_32 encodings no longer try to dynamically guess the endian based on a byte order mark. Use Encoding::UTF_16BE/UTF_16LE and Encoding::UTF_32BE/UTF_32LE instead. This change speeds up getting the encoding of a String. [Feature #18949]
    • Limit maximum encoding set size by 256. If exceeding maximum size, EncodingError will be raised. [Feature #18949]

  • Enumerator

    • Enumerator.product has been added. Enumerator::Product is the implementation. [Feature #18685]

  • Exception

    • Exception#detailed_message has been added. The default error printer calls this method on the Exception object instead of #message. [Feature #18564]

  • Hash

    • Hash#shift now always returns nil if the hash is empty, instead of returning the default value or calling the default proc. [Bug #16908]

  • Integer

  • Kernel

    • Kernel#binding raises RuntimeError if called from a non-Ruby frame (such as a method defined in C). [Bug #18487]

  • MatchData

  • Module

  • Proc

  • Process

    • Added RLIMIT_NPTS constant to FreeBSD platform

  • Regexp

    • The cache-based optimization is introduced. Many (but not all) Regexp matching is now in linear time, which will prevent regular expression denial of service (ReDoS) vulnerability. [Feature #19104]

    • Regexp.linear_time? is introduced. [Feature #19194]

    • Regexp.new now supports passing the regexp flags not only as an Integer, but also as a String. Unknown flags raise ArgumentError. Otherwise, anything other than true, false, nil or Integer will be warned. [Feature #18788]

    • Regexp.timeout= has been added. Also, Regexp.new new supports timeout keyword. See [Feature #17837]

  • Refinement

  • RubyVM::AbstractSyntaxTree

    • Add error_tolerant option for parse, parse_file and of. [Feature #19013] With this option

      1. SyntaxError is suppressed
      2. AST is returned for invalid input
      3. end is complemented when a parser reaches to the end of input but end is insufficient
      4. end is treated as keyword based on indent
      # Without error_tolerant option
root = RubyVM::AbstractSyntaxTree.parse(<<~RUBY)
def m
  a = 10
  if
end
RUBY
# => <internal:ast>:33:in `parse': syntax error, unexpected `end' (SyntaxError)

# With error_tolerant option
root = RubyVM::AbstractSyntaxTree.parse(<<~RUBY, error_tolerant: true)
def m
  a = 10
  if
end
RUBY
p root # => #<RubyVM::AbstractSyntaxTree::Node:SCOPE@1:0-4:3>

# `end` is treated as keyword based on indent
root = RubyVM::AbstractSyntaxTree.parse(<<~RUBY, error_tolerant: true)
module Z
  class Foo
    foo.
  end

  def bar
  end
end
RUBY
p root.children[-1].children[-1].children[-1].children[-2..-1]
# => [#<RubyVM::AbstractSyntaxTree::Node:CLASS@2:2-4:5>, #<RubyVM::AbstractSyntaxTree::Node:DEFN@6:2-7:5>]

    • Add keep_tokens option for parse, parse_file and of. Add #tokens and #all_tokens for RubyVM::AbstractSyntaxTree::Node [Feature #19070]

      root = RubyVM::AbstractSyntaxTree.parse("x = 1 + 2", keep_tokens: true)
root.tokens # => [[0, :tIDENTIFIER, "x", [1, 0, 1, 1]], [1, :tSP, " ", [1, 1, 1, 2]], ...]
root.tokens.map{_1[2]}.join # => "x = 1 + 2"

  • Set

    • Set is now available as a built-in class without the need for require "set". [Feature #16989] It is currently autoloaded via the Set constant or a call to Enumerable#to_set.

  • String

    • String#byteindex and String#byterindex have been added. [Feature #13110]
    • Update Unicode to Version 15.0.0 and Emoji Version 15.0. [Feature #18639] (also applies to Regexp)
    • String#bytesplice has been added. [Feature #18598]
    • String#dedup has been added as an alias to String#-@. [Feature #18595]

  • Struct

    • A Struct class can also be initialized with keyword arguments without keyword_init: true on Struct.new [Feature #16806]

      Post = Struct.new(:id, :name)
Post.new(1, "hello") #=> #<struct Post id=1, name="hello">
# From Ruby 3.2, the following code also works without keyword_init: true.
Post.new(id: 1, name: "hello") #=> #<struct Post id=1, name="hello">

  • Thread

  • Thread::Queue

  • Thread::SizedQueue

  • Time

    • Time#deconstruct_keys is added, allowing to use Time instances in pattern-matching expressions [Feature #19071]

    • Time.new now can parse a string like generated by Time#inspect and return a Time instance based on the given argument. [Feature #18033]

  • SyntaxError

  • TracePoint

    • TracePoint#binding now returns nil for c_call/c_return TracePoints. [Bug #18487]
    • TracePoint#enable target_thread keyword argument now defaults to the current thread if a block is given and target and target_line keyword arguments are not passed. [Bug #16889]

  • UnboundMethod

    • UnboundMethod#== returns true if the actual method is same. For example, String.instance_method(:object_id) == Array.instance_method(:object_id) returns true. [Feature #18798]

    • UnboundMethod#inspect does not show the receiver of instance_method. For example String.instance_method(:object_id).inspect returns "#<UnboundMethod: Kernel#object_id()>" (was "#<UnboundMethod: String(Kernel)#object_id()>").

  • GC

    • Expose need_major_gc via GC.latest_gc_info. GH-6791

  • ObjectSpace

    • ObjectSpace.dump_all dump shapes as well. GH-6868

Stdlib updates

  • Bundler

  • RubyGems

  • CGI

    • CGI.escapeURIComponent and CGI.unescapeURIComponent are added. [Feature #18822]

  • Coverage

    • Coverage.setup now accepts eval: true. By this, eval and related methods are able to generate code coverage. [Feature #19008]

    • Coverage.supported?(mode) enables detection of what coverage modes are supported. [Feature #19026]

  • Date

    • Added Date#deconstruct_keys and DateTime#deconstruct_keys same as [Feature #19071]

  • ERB

    • ERB::Util.html_escape is made faster than CGI.escapeHTML.
      • It no longer allocates a String object when no character needs to be escaped.
      • It skips calling #to_s method when an argument is already a String.
      • ERB::Escape.html_escape is added as an alias to ERB::Util.html_escape, which has not been monkey-patched by Rails.
    • ERB::Util.url_encode is made faster using CGI.escapeURIComponent.
    • -S option is removed from erb command.

  • FileUtils

    • Add FileUtils.ln_sr method and relative: option to FileUtils.ln_s. [Feature #18925]

  • IRB

    • debug.gem integration commands have been added: debug, break, catch, next, delete, step, continue, finish, backtrace, info
    • More Pry-like commands and features have been added.
      • edit and show_cmds (like Pry's help) are added.
      • ls takes -g or -G option to filter out outputs.
      • show_source is aliased from $ and accepts unquoted inputs.
      • whereami is aliased from @.

  • Net::Protocol

  • Pathname

  • Socket

    • Added the following constants for supported platforms.
      • SO_INCOMING_CPU
      • SO_INCOMING_NAPI_ID
      • SO_RTABLE
      • SO_SETFIB
      • SO_USER_COOKIE
      • TCP_KEEPALIVE
      • TCP_CONNECTION_INFO

  • SyntaxSuggest

    • The feature of syntax_suggest formerly dead_end is integrated in Ruby. [Feature #18159]

  • UNIXSocket

    • Add support for UNIXSocket on Windows. Emulate anonymous sockets. Add support for File.socket? and File::Stat#socket? where possible. [Feature #19135]

  • The following default gems are updated.

    • RubyGems 3.4.1
    • abbrev 0.1.1
    • benchmark 0.2.1
    • bigdecimal 3.1.3
    • bundler 2.4.1
    • cgi 0.3.6
    • csv 3.2.6
    • date 3.3.3
    • delegate 0.3.0
    • did_you_mean 1.6.3
    • digest 3.1.1
    • drb 2.1.1
    • english 0.7.2
    • erb 4.0.2
    • error_highlight 0.5.1
    • etc 1.4.2
    • fcntl 1.0.2
    • fiddle 1.1.1
    • fileutils 1.7.0
    • forwardable 1.3.3
    • getoptlong 0.2.0
    • io-console 0.6.0
    • io-nonblock 0.2.0
    • io-wait 0.3.0
    • ipaddr 1.2.5
    • irb 1.6.2
    • json 2.6.3
    • logger 1.5.3
    • mutex_m 0.1.2
    • net-http 0.3.2
    • net-protocol 0.2.1
    • nkf 0.1.2
    • open-uri 0.3.0
    • open3 0.1.2
    • openssl 3.1.0
    • optparse 0.3.1
    • ostruct 0.5.5
    • pathname 0.2.1
    • pp 0.4.0
    • pstore 0.1.2
    • psych 5.0.1
    • racc 1.6.2
    • rdoc 6.5.0
    • readline-ext 0.1.5
    • reline 0.3.2
    • resolv 0.2.2
    • resolv-replace 0.1.1
    • securerandom 0.2.2
    • set 1.0.3
    • stringio 3.0.4
    • strscan 3.0.5
    • syntax_suggest 1.0.2
    • syslog 0.1.1
    • tempfile 0.1.3
    • time 0.2.1
    • timeout 0.3.1
    • tmpdir 0.1.3
    • tsort 0.1.1
    • un 0.2.1
    • uri 0.12.0
    • weakref 0.1.2
    • win32ole 1.8.9
    • yaml 0.2.1
    • zlib 3.0.0

  • The following bundled gems are updated.

    • minitest 5.16.3
    • power_assert 2.0.3
    • test-unit 3.5.7
    • net-ftp 0.2.0
    • net-imap 0.3.4
    • net-pop 0.1.2
    • net-smtp 0.3.3
    • rbs 2.8.2
    • typeprof 0.21.3
    • debug 1.7.1

See GitHub releases like GitHub Releases of Logger or changelog for details of the default gems or bundled gems.

Supported platforms

Compatibility issues

  • String#to_c currently treat a sequence of underscores as an end of Complex string. [Bug #19087]

  • Now ENV.clone raises TypeError as well as ENV.dup [Bug #17767]

Removed constants

The following deprecated constants are removed.

Removed methods

The following deprecated methods are removed.

Source code incompatibility of extension libraries

  • Extension libraries provide PRNG, subclasses of Random, need updates. See [PRNG update] below for more information. [Bug #19100]

Error printer

  • Ruby no longer escapes control characters and backslashes in an error message. [Feature #18367]

Constant lookup when defining a class/module

  • When defining a class/module directly under the Object class by class/module statement, if there is already a class/module defined by Module#include with the same name, the statement was handled as "open class" in Ruby 3.1 or before. Since Ruby 3.2, a new class is defined instead. [Feature #18832]

Stdlib compatibility issues

  • Psych no longer bundles libyaml sources. And also Fiddle no longer bundles libffi sources. Users need to install the libyaml/libffi library themselves via the package manager like apt, yum, brew, etc.

    Psych and fiddle supported the static build with specific version of libyaml and libffi sources. You can build psych with libyaml-0.2.5 like this.

    $ ./configure --with-libyaml-source-dir=/path/to/libyaml-0.2.5

    And you can build fiddle with libffi-3.4.4 like this.

    $ ./configure --with-libffi-source-dir=/path/to/libffi-3.4.4

    [Feature #18571]

  • Check cookie name/path/domain characters in CGI::Cookie. [CVE-2021-33621]

  • URI.parse return empty string in host instead of nil. [sec-156615]

C API updates

Updated C APIs

The following APIs are updated.

  • PRNG update

    rb_random_interface_t in ruby/random.h updated and versioned. Extension libraries which use this interface and built for older versions need to rebuild with adding init_int32 function.

Added C APIs

  • VALUE rb_hash_new_capa(long capa) was added to created hashes with the desired capacity.
  • rb_internal_thread_add_event_hook and rb_internal_thread_add_event_hook were added to instrument threads scheduling. The following events are available:
    • RUBY_INTERNAL_THREAD_EVENT_STARTED
    • RUBY_INTERNAL_THREAD_EVENT_READY
    • RUBY_INTERNAL_THREAD_EVENT_RESUMED
    • RUBY_INTERNAL_THREAD_EVENT_SUSPENDED
    • RUBY_INTERNAL_THREAD_EVENT_EXITED
  • rb_debug_inspector_current_depth and rb_debug_inspector_frame_depth are added for debuggers.

Removed C APIs

The following deprecated APIs are removed.

  • rb_cData variable.
  • "taintedness" and "trustedness" functions. [Feature #16131]

Implementation improvements

  • Fixed several race conditions in Kernel#autoload. [Bug #18782]
  • Cache invalidation for expressions referencing constants is now more fine-grained. RubyVM.stat(:global_constant_state) was removed because it was closely tied to the previous caching scheme where setting any constant invalidates all caches in the system. New keys, :constant_cache_invalidations and :constant_cache_misses, were introduced to help with use cases for :global_constant_state. [Feature #18589]
  • The cache-based optimization for Regexp matching is introduced. [Feature #19104]
  • Variable Width Allocation is now enabled by default. [Feature #18239]
  • Added a new instance variable caching mechanism, called object shapes, which improves inline cache hits for most objects and allows us to generate very efficient JIT code. Objects whose instance variables are defined in a consistent order will see the most performance benefits. [Feature #18776]
  • Speed up marking instruction sequences by using a bitmap to find "markable" objects. This change results in faster major collections. [Feature #18875]

JIT

YJIT

  • YJIT is no longer experimental
    • Has been tested on production workloads for over a year and proven to be quite stable.
  • YJIT now supports both x86-64 and arm64/aarch64 CPUs on Linux, MacOS, BSD and other UNIX platforms.
    • This release brings support for Mac M1/M2, AWS Graviton and Raspberry Pi 4.
  • Building YJIT now requires Rust 1.58.0+. [Feature #18481]
    • In order to ensure that CRuby is built with YJIT, please install rustc >= 1.58.0 before running ./configure
    • Please reach out to the YJIT team should you run into any issues.
  • Physical memory for JIT code is lazily allocated. Unlike Ruby 3.1, the RSS of a Ruby process is minimized because virtual memory pages allocated by --yjit-exec-mem-size will not be mapped to physical memory pages until actually utilized by JIT code.
  • Introduce Code GC that frees all code pages when the memory consumption by JIT code reaches --yjit-exec-mem-size.
    • RubyVM::YJIT.runtime_stats returns Code GC metrics in addition to existing inline_code_size and outlined_code_size keys: code_gc_count, live_page_count, freed_page_count, and freed_code_size.
  • Most of the statistics produced by RubyVM::YJIT.runtime_stats are now available in release builds.
    • Simply run ruby with --yjit-stats to compute and dump stats (incurs some run-time overhead).
  • YJIT is now optimized to take advantage of object shapes. [Feature #18776]
  • Take advantage of finer-grained constant invalidation to invalidate less code when defining new constants. [Feature #18589]
  • The default --yjit-exec-mem-size is changed to 64 (MiB).
  • The default --yjit-call-threshold is changed to 30.

MJIT

  • The MJIT compiler is re-implemented in Ruby as ruby_vm/mjit/compiler.
  • MJIT compiler is executed under a forked Ruby process instead of doing it in a native thread called MJIT worker. [Feature #18968]
    • As a result, Microsoft Visual Studio (MSWIN) is no longer supported.
  • MinGW is no longer supported. [Feature #18824]
  • Rename --mjit-min-calls to --mjit-call-threshold.
  • Change default --mjit-max-cache back from 10000 to 100.