We are pleased to announce the release of Ruby 3.2.0. Ruby 3.2 adds many features and performance improvements.
WASI based WebAssembly support
This is an initial port of WASI based WebAssembly support. This enables a CRuby binary to be available on a Web browser, a Serverless Edge environment, or other kinds of WebAssembly/WASI embedders. Currently this port passes basic and bootstrap test suites not using the Thread API.
Background
WebAssembly (Wasm) was originally introduced to run programs safely and fast in web browsers. But its objective - running programs efficiently with security on various environment - is long wanted not only for web but also by general applications.
WASI (The WebAssembly System Interface) is designed for such use cases. Though such applications need to communicate with operating systems, WebAssembly runs on a virtual machine which didn’t have a system interface. WASI standardizes it.
WebAssembly/WASI support in Ruby intends to leverage those projects. It enables Ruby developers to write applications which run on such promised platforms.
Use case
This support encourages developers to utilize CRuby in a WebAssembly environment. An example use case is TryRuby playground’s CRuby support. Now you can try original CRuby in your web browser.
Technical points
Today’s WASI and WebAssembly itself is missing some features to implement Fiber, exception, and GC because it’s still evolving, and also for security reasons. So CRuby fills the gap by using Asyncify, which is a binary transformation technique to control execution in userland.
In addition, we built a VFS on top of WASI so that we can easily pack Ruby apps into a single .wasm file. This makes distribution of Ruby apps a bit easier.
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 Apple M1/M2, AWS Graviton, Raspberry Pi 4 and more.
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 the ./configure script.
Please reach out to the YJIT team should you run into any issues.
The YJIT 3.2 release is faster than 3.1, and has about 1/3 as much memory overhead.
Overall YJIT is 41% faster (geometric mean) than the Ruby interpreter on yjit-bench.
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.
Regexp improvements against ReDoS
It is known that Regexp matching may take unexpectedly long. If your code attempts to match a possibly inefficient Regexp against an untrusted input, an attacker may exploit it for efficient Denial of Service (so-called Regular expression DoS, or ReDoS).
We have introduced two improvements that significantly mitigate ReDoS.
Improved Regexp matching algorithm
Since Ruby 3.2, Regexp’s matching algorithm has been greatly improved by using a memoization technique.
# This match takes 10 sec. in Ruby 3.1, and 0.003 sec. in Ruby 3.2
/^a*b?a*$/ =~ "a" * 50000 + "x"
The improved matching algorithm allows most Regexp matching (about 90% in our experiments) to be completed in linear time.
(For preview users: this optimization may consume memory proportional to the input length for each match. We expect no practical problems to arise because this memory allocation is usually delayed, and a normal Regexp match should consume at most 10 times as much memory as the input length. If you run out of memory when matching Regexps in a real-world application, please report it.)
The optimization above cannot be applied to some kind of regular expressions, such as those including advanced features (e.g., back-references or look-around), or with a huge fixed number of repetitions. As a fallback measure, a timeout feature for Regexp matches is also introduced.
Regexp.timeout=1.0/^a*b?a*()\1$/=~"a"*50000+"x"#=> Regexp::TimeoutError is raised in one second
Note that Regexp.timeout is a global configuration. If you want to use different timeout settings for some special Regexps, you may want to use the timeout keyword for Regexp.new.
Regexp.timeout=1.0# This regexp has no timeoutlong_time_re=Regexp.new('^a*b?a*()\1$',timeout: Float::INFINITY)long_time_re=~"a"*50000+"x"# never interrupted
The feature of syntax_suggest (formerly dead_end) is integrated into Ruby. This helps you find the position of errors such as missing or superfluous ends, to get you back on your way faster, such as in the following example:
Unmatched `end', missing keyword (`do', `def`, `if`, etc.) ?
1 class Dog
> 2 defbark
> 3 end
4 end
Now it points at the relevant argument(s) for TypeError and ArgumentError
test.rb:2:in `+': nil can't be coerced into Integer (TypeError)
sum = ary[0] + ary[1]
^^^^^^
Language
Anonymous rest and keyword rest arguments can now be passed as
arguments, instead of just used in method parameters.
[Feature #18351]
deffoo(*)bar(*)enddefbaz(**)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:
The 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 potentially missing ruby2_keywords
is to run the test suite, find the last method which must
receive keyword arguments for each place where the test suite fails, and use puts nil, caller, nil there. Then check that each
method/block on the call chain which must delegate keywords is correctly marked
with ruby2_keywords. [Bug #18625] [Bug #16466]
deftarget(**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_keywordsdefbar(*args)target(*args)endruby2_keywordsdeffoo(*args)bar(*args)endfoo(k: 1)
Performance improvements
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.
PubGrub is the next generation solving algorithm used by pub package manager for the Dart programming language.
You may get different resolution result after this change. Please report such cases to RubyGems/Bundler issues
RubyGems still uses Molinillo resolver in Ruby 3.2. We plan to replace it with PubGrub in the future.
Other notable changes since 3.1
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]
Proc#dup returns an instance of subclass. [Bug #17545]
Proc#parameters now accepts lambda keyword. [Feature #15357]
Refinement
Refinement#refined_class has been added. [Feature #12737]
RubyVM::AbstractSyntaxTree
Add error_tolerant option for parse, parse_file and of. [Feature #19013]
With this option
SyntaxError is suppressed
AST is returned for invalid input
end is complemented when a parser reaches to the end of input but end is insufficient
end is treated as keyword based on indent
# Without error_tolerant optionroot=RubyVM::AbstractSyntaxTree.parse(<<~RUBY)
def m
a = 10
if
end
RUBY# => <internal:ast>:33:in `parse': syntax error, unexpected `end' (SyntaxError)# With error_tolerant optionroot=RubyVM::AbstractSyntaxTree.parse(<<~RUBY,error_tolerant: true)
def m
a = 10
if
end
RUBYproot# => #<RubyVM::AbstractSyntaxTree::Node:SCOPE@1:0-4:3># `end` is treated as keyword based on indentroot=RubyVM::AbstractSyntaxTree.parse(<<~RUBY,error_tolerant: true)
module Z
class Foo
foo.
end
def bar
end
end
RUBYproot.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. [Feature #19070]
Set is now available as a builtin 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)
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">
We no longer bundle 3rd party sources like libyaml, libffi.
libyaml source has been removed from psych. You may need to install libyaml-dev with Ubuntu/Debian platform. The package name is different for each platform.
Bundled libffi source is also removed from fiddle
Psych and fiddle supported static builds with specific versions of libyaml and libffi sources. You can build psych with libyaml-0.2.5 like this:
rb_random_interface_t updated and versioned.
Extension libraries which use this interface and built for older versions.
Also init_int32 function needs to be defined.
Removed C APIs
The following deprecated APIs are removed.
rb_cData variable.
“taintedness” and “trustedness” functions. [Feature #16131]
Standard library updates
Bundler
Add –ext=rust support to bundle gem for creating simple gems with Rust extensions.
[GH-rubygems-6149]
Here is everything you need to know about this upgrade. Please take a good look at what changed and the test results before merging this pull request.
What changed?
Release Notes
3.2.2
3.2.1
3.2.0
3.1.4
All Depfu comment commands