21 KiB

Исходник Ответственный История

NEWS for Ruby 3.3.0

This document is a list of user-visible feature changes since the 3.2.0 release, except for bug fixes.

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

Command line options

A new performance warning category was introduced. They are not displayed by default even in verbose mode. Turn them on with -W:performance or Warning[:performance] = true. [Feature #19538]
A new RUBY_CRASH_REPORT environment variable was introduced to allow redirecting Ruby crash reports to a file or sub command. See the BUG REPORT ENVIRONMENT section of the ruby manpage for further details. [Feature #19790]

Core classes updates

Note: We're only listing outstanding class updates.

Array
- Array#pack now raises ArgumentError for unknown directives. [Bug #19150]
Dir
- Dir.for_fd added for returning a Dir object for the directory specified by the provided directory file descriptor. [Feature #19347]
- Dir.fchdir added for changing the directory to the directory specified by the provided directory file descriptor. [Feature #19347]
- Dir#chdir added for changing the directory to the directory specified by the provided Dir object. [Feature #19347]
Encoding
- Encoding#replicate has been removed, it was already deprecated. [Feature #18949]

Fiber

Introduce Fiber#kill. [Bug #595]

fiber = Fiber.new do
  while true
    puts "Yielding..."
    Fiber.yield
  end
ensure
  puts "Exiting..."
end

fiber.resume
# Yielding...
fiber.kill
# Exiting...

MatchData
- MatchData#named_captures now accepts optional symbolize_names keyword. [Feature #19591]
Module
- Module#set_temporary_name added for setting a temporary name for a module. [Feature #19521]
ObjectSpace::WeakKeyMap
- New core class to build collections with weak references. The class use equality semantic to lookup keys like a regular hash, but it doesn't hold strong references on the keys. [Feature #18498]
ObjectSpace::WeakMap
- ObjectSpace::WeakMap#delete was added to eagerly clear weak map entries. [Feature #19561]
Proc
- Now Proc#dup and Proc#clone call #initialize_dup and #initialize_clone hooks respectively. [Feature #19362]
Process
- New Process.warmup method that notify the Ruby virtual machine that the boot sequence is finished, and that now is a good time to optimize the application. This is useful for long-running applications. The actual optimizations performed are entirely implementation-specific and may change in the future without notice. [Feature #18885]
Process::Status
- Process::Status#& and Process::Status#>> are deprecated. [Bug #19868]
Range
- Range#reverse_each can now process beginless ranges with an Integer endpoint. [Feature #18515]
- Range#reverse_each now raises TypeError for endless ranges. [Feature #18551]
- Range#overlap? added for checking if two ranges overlap. [Feature #19839]
Refinement
- Add Refinement#target as an alternative of Refinement#refined_class. Refinement#refined_class is deprecated and will be removed in Ruby 3.4. [Feature #19714]
Regexp
- The cache-based optimization now supports lookarounds and atomic groupings. That is, match for Regexp containing these extensions can now also be performed in linear time to the length of the input string. However, these cannot contain captures and cannot be nested. [Feature #19725]
String
- String#unpack now raises ArgumentError for unknown directives. [Bug #19150]
- String#bytesplice now accepts new arguments index/length or range of the source string to be copied. [Feature #19314]
Thread::Queue
- Thread::Queue#freeze now raises TypeError. [Bug #17146]
Thread::SizedQueue
- Thread::SizedQueue#freeze now raises TypeError. [Bug #17146]
Time
- Time.new with a string argument became stricter. [Bug #19293]
```
Time.new('2023-12-20')
#  no time information (ArgumentError)
```
TracePoint
- TracePoint supports rescue event. When the raised exception was rescued, the TracePoint will fire the hook. rescue event only supports Ruby-level rescue. [Feature #19572]

Stdlib updates

RubyGems and Bundler warn if users do require the following gems without adding them to Gemfile or gemspec. This is because they will become the bundled gems in the future version of Ruby. This warning is suppressed if you use bootsnap gem. We recommend to run your application with DISABLE_BOOTSNAP=1 environmental variable at least once. This is limitation of this version. [Feature #19351] [Feature #19776] [Feature #19843]
- abbrev
- base64
- bigdecimal
- csv
- drb
- getoptlong
- mutex_m
- nkf
- observer
- racc
- resolv-replace
- rinda
- syslog
Socket#recv and Socket#recv_nonblock returns nil instead of an empty string on closed connections. Socket#recvmsg and Socket#recvmsg_nonblock returns nil instead of an empty packet on closed connections. [Bug #19012]
Name resolution such as Socket.getaddrinfo, Socket.getnameinfo, Addrinfo.getaddrinfo, etc. can now be interrupted. [Feature #19965]
Random::Formatter#alphanumeric is extended to accept optional chars keyword argument. [Feature #18183]

The following default gem is added.

prism 0.19.0

The following default gems are updated.

RubyGems 3.5.3
abbrev 0.1.2
base64 0.2.0
benchmark 0.3.0
bigdecimal 3.1.5
bundler 2.5.3
cgi 0.4.1
csv 3.2.8
date 3.3.4
delegate 0.3.1
drb 2.2.0
english 0.8.0
erb 4.0.3
error_highlight 0.6.0
etc 1.4.3
fcntl 1.1.0
fiddle 1.1.2
fileutils 1.7.2
find 0.2.0
getoptlong 0.2.1
io-console 0.7.1
io-nonblock 0.3.0
io-wait 0.3.1
ipaddr 1.2.6
irb 1.11.0
json 2.7.1
logger 1.6.0
mutex_m 0.2.0
net-http 0.4.0
net-protocol 0.2.2
nkf 0.1.3
observer 0.1.2
open-uri 0.4.1
open3 0.2.1
openssl 3.2.0
optparse 0.4.0
ostruct 0.6.0
pathname 0.3.0
pp 0.5.0
prettyprint 0.2.0
pstore 0.1.3
psych 5.1.2
rdoc 6.6.2
readline 0.0.4
reline 0.4.1
resolv 0.3.0
rinda 0.2.0
securerandom 0.3.1
set 1.1.0
shellwords 0.2.0
singleton 0.2.0
stringio 3.1.0
strscan 3.0.7
syntax_suggest 2.0.0
syslog 0.1.2
tempfile 0.2.1
time 0.3.0
timeout 0.4.1
tmpdir 0.2.0
tsort 0.2.0
un 0.3.0
uri 0.13.0
weakref 0.1.3
win32ole 1.8.10
yaml 0.3.0
zlib 3.1.0

The following bundled gem is promoted from default gems.

racc 1.7.3

The following bundled gems are updated.

minitest 5.20.0
rake 13.1.0
test-unit 3.6.1
rexml 3.2.6
rss 0.3.0
net-ftp 0.3.3
net-imap 0.4.9
net-smtp 0.4.0
rbs 3.4.0
typeprof 0.21.9
debug 1.9.1

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

Prism

Introduced the Prism parser as a default gem
- Prism is a portable, error tolerant, and maintainable recursive descent parser for the Ruby language
Prism is production ready and actively maintained, you can use it in place of Ripper
- There is extensive documentation on how to use Prism
- Prism is both a C library that will be used internally by CRuby and a Ruby gem that can be used by any tooling which needs to parse Ruby code
- Notable methods in the Prism API are:
  - Prism.parse(source) which returns the AST as part of a parse result object
  - Prism.parse_comments(source) which returns the comments
  - Prism.parse_success?(source) which returns true if there are no errors
You can make pull requests or issues directly on the Prism repository if you are interested in contributing
You can now use ruby --parser=prism or RUBYOPT="--parser=prism" to experiment with the Prism compiler. Please note that this flag is for debugging only.

Compatibility issues

Subprocess creation/forking via the following file open methods is deprecated. [Feature #19630]
- Kernel#open
- URI.open
- IO.binread
- IO.foreach
- IO.readlines
- IO.read
- IO.write
When given a non-lambda, non-literal block, Kernel#lambda with now raises ArgumentError instead of returning it unmodified. These usages have been issuing warnings under the Warning[:deprecated] category since Ruby 3.0.0. [Feature #19777]
The RUBY_GC_HEAP_INIT_SLOTS environment variable has been deprecated and removed. Environment variables RUBY_GC_HEAP_%d_INIT_SLOTS should be used instead. [Feature #19785]
it calls without arguments in a block with no ordinary parameters are deprecated. it will be a reference to the first block parameter in Ruby 3.4. [Feature #18980]
Error message for NoMethodError have changed to not use the target object's #inspect for efficiency, and says "instance of ClassName" instead. [Feature #18285]
```
([1] * 100).nonexisting
# undefined method `nonexisting' for an instance of Array (NoMethodError)
```
Now anonymous parameters forwarding is disallowed inside a block that uses anonymous parameters. [Feature #19370]

Stdlib compatibility issues

racc is promoted to bundled gems.
- You need to add racc to your Gemfile if you use racc under bundler environment.
ext/readline is retired
- We have reline that is pure Ruby implementation compatible with ext/readline API. We rely on reline in the future. If you need to use ext/readline, you can install ext/readline via rubygems.org with gem install readline-ext.
- We no longer need to install libraries like libreadline or libedit.

C API updates

rb_postponed_job updates
- New APIs and deprecated APIs (see comments for details)
  - added: rb_postponed_job_preregister()
  - added: rb_postponed_job_trigger()
  - deprecated: rb_postponed_job_register() (and semantic change. see below)
  - deprecated: rb_postponed_job_register_one()
- The postponed job APIs have been changed to address some rare crashes. To solve the issue, we introduced new two APIs and deprecated current APIs. The semantics of these functions have also changed slightly; rb_postponed_job_register now behaves like the once variant in that multiple calls with the same func might be coalesced into a single execution of the func [Feature #20057]
Some updates for internal thread event hook APIs
- rb_internal_thread_event_data_t with a target Ruby thread (VALUE) and callback functions (rb_internal_thread_event_callback) receive it. https://github.com/ruby/ruby/pull/8885
- The following functions are introduced to manipulate Ruby thread local data from internal thread event hook APIs (they are introduced since Ruby 3.2). https://github.com/ruby/ruby/pull/8936
  - rb_internal_thread_specific_key_create()
  - rb_internal_thread_specific_get()
  - rb_internal_thread_specific_set()
rb_profile_thread_frames() is introduced to get a frames from a specific thread. [Feature #10602]
rb_data_define() is introduced to define Data. [Feature #19757]
rb_ext_resolve_symbol() is introduced to search a function from extension libraries. [Feature #20005]
IO related updates:
- The details of rb_io_t will be hidden and deprecated attributes are added for each members. [Feature #19057]
- rb_io_path(VALUE io) is introduced to get a path of io.
- rb_io_closed_p(VALUE io) to get opening or closing of io.
- rb_io_mode(VALUE io) to get the mode of io.
- rb_io_open_descriptor() is introduced to make an IO object from a file descriptor.

Implementation improvements

Parser

Replace Bison with Lrama LALR parser generator. No need to install Bison to build Ruby from source code anymore. We will no longer suffer bison compatibility issues and we can use new features by just implementing it to Lrama. [Feature #19637]
- See The future vision of Ruby Parser for detail.
- Lrama internal parser is a LR parser generated by Racc for maintainability.
- Parameterizing Rules (?, *, +) are supported, it will be used in Ruby parse.y.

GC / Memory management

Major performance improvements over Ruby 3.2
- Young objects referenced by old objects are no longer immediately promoted to the old generation. This significantly reduces the frequency of major GC collections. [Feature #19678]
- A new REMEMBERED_WB_UNPROTECTED_OBJECTS_LIMIT_RATIO tuning variable was introduced to control the number of unprotected objects cause a major GC collection to trigger. The default is set to 0.01 (1%). This significantly reduces the frequency of major GC collection. [Feature #19571]
- Write Barriers were implemented for many core types that were missing them, notably Time, Enumerator, MatchData, Method, File::Stat, BigDecimal and several others. This significantly reduces minor GC collection time and major GC collection frequency.
- Most core classes are now using Variable Width Allocation, notably Hash, Time, Thread::Backtrace, Thread::Backtrace::Location, File::Stat, Method. This makes these classes faster to allocate and free, use less memory and reduce heap fragmentation.
defined?(@ivar) is optimized with Object Shapes.

YJIT

Major performance improvements over Ruby 3.2
- Support for splat and rest arguments has been improved.
- Registers are allocated for stack operations of the virtual machine.
- More calls with optional arguments are compiled. Exception handlers are also compiled.
- Unsupported call types and megamorphic call sites no longer exit to the interpreter.
- Basic methods like Rails #blank? and specialized #present? are inlined.
- Integer#*, Integer#!=, String#!=, String#getbyte, Kernel#block_given?, Kernel#is_a?, Kernel#instance_of?, and Module#=== are specially optimized.
- Compilation speed is now slightly faster than Ruby 3.2.
- Now more than 3x faster than the interpreter on Optcarrot!
Significantly improved memory usage over Ruby 3.2
- Metadata for compiled code uses a lot less memory.
- --yjit-call-threshold is automatically raised from 30 to 120 when the application has more than 40,000 ISEQs.
- --yjit-cold-threshold is added to skip compiling cold ISEQs.
- More compact code is generated on Arm64.
Code GC is now disabled by default
- --yjit-exec-mem-size is treated as a hard limit where compilation of new code stops.
- No sudden drops in performance due to code GC. Better copy-on-write behavior on servers reforking with Pitchfork.
- You can still enable code GC if desired with --yjit-code-gc
Add RubyVM::YJIT.enable that can enable YJIT at run-time
- You can start YJIT without modifying command-line arguments or environment variables. Rails 7.2 will enable YJIT by default using this method.
- This can also be used to enable YJIT only once your application is done booting. --yjit-disable can be used if you want to use other YJIT options while disabling YJIT at boot.
More YJIT stats are available by default
- yjit_alloc_size and several more metadata-related stats are now available by default.
- ratio_in_yjit stat produced by --yjit-stats is now available in release builds, a special stats or dev build is no longer required to access most stats.
Add more profiling capabilities
- --yjit-perf is added to facilitate profiling with Linux perf.
- --yjit-trace-exits now supports sampling with --yjit-trace-exits-sample-rate=N
More thorough testing and multiple bug fixes
--yjit-stats=quiet is added to avoid printing stats on exit.

MJIT

MJIT is removed.
- --disable-jit-support is removed. Consider using --disable-yjit --disable-rjit instead.

RJIT

Introduced a pure-Ruby JIT compiler RJIT.
- RJIT supports only x86_64 architecture on Unix platforms.
- Unlike MJIT, it doesn't require a C compiler at runtime.
RJIT exists only for experimental purposes.
- You should keep using YJIT in production.

M:N Thread scheduler

M:N Thread scheduler is introduced. [Feature #19842]
- Background: Ruby 1.8 and before, M:1 thread scheduler (M Ruby threads with 1 native thread. Called as User level threads or Green threads) is used. Ruby 1.9 and later, 1:1 thread scheduler (1 Ruby thread with 1 native thread). M:1 threads takes lower resources compare with 1:1 threads because it needs only 1 native threads. However it is difficult to support context switching for all of blocking operation so 1:1 threads are employed from Ruby 1.9. M:N thread scheduler uses N native threads for M Ruby threads (N is small number in general). It doesn't need same number of native threads as Ruby threads (similar to the M:1 thread scheduler). Also our M:N threads supports blocking operations well same as 1:1 threads. See the ticket for more details. Our M:N thread scheduler refers on the goroutine scheduler in the Go language.
- In a ractor, only 1 thread can run in a same time because of implementation. Therefore, applications that use only one Ractor (most applications) M:N thread scheduler works as M:1 thread scheduler with further extension from Ruby 1.8.
- M:N thread scheduler can introduce incompatibility for C-extensions, so it is disabled by default on the main Ractors. RUBY_MN_THREADS=1 environment variable will enable it. On non-main Ractors, M:N thread scheduler is enabled (and can not disable it now).
- N (the number of native threads) can be specified with RUBY_MAX_CPU environment variable. The default is 8. Note that more than N native threads are used to support many kind of blocking operations.

21 KiB Исходник Ответственный История