January 2025 - ruby-core - ml.ruby-lang.org

[ruby-core:113153] [Ruby master Bug#19584] Crash in rb_gc_register_address
by peterzhu2118 (Peter Zhu) 13 Mar '25

13 Mar '25

Issue #19584 has been reported by peterzhu2118 (Peter Zhu). ---------------------------------------- Bug #19584: Crash in rb_gc_register_address https://bugs.ruby-lang.org/issues/19584 * Author: peterzhu2118 (Peter Zhu) * Status: Open * Priority: Normal * Backport: 3.0: REQUIRED, 3.1: REQUIRED, 3.2: REQUIRED ---------------------------------------- GitHub PR: https://github.com/ruby/ruby/pull/7670 Some C extensions pass a pointer to a global variable to rb_gc_register_address. However, if a GC is triggered inside of rb_gc_register_address, then the object could get swept since it does not exist on the stack. -- https://bugs.ruby-lang.org/

4 6

[ruby-core:115512] [Ruby master Bug#20025] Parsing identifiers/constants is case-folding dependent
by kddnewton (Kevin Newton) 13 Mar '25

13 Mar '25

Issue #20025 has been reported by kddnewton (Kevin Newton). ---------------------------------------- Bug #20025: Parsing identifiers/constants is case-folding dependent https://bugs.ruby-lang.org/issues/20025 * Author: kddnewton (Kevin Newton) * Status: Open * Priority: Normal * Backport: 3.0: UNKNOWN, 3.1: UNKNOWN, 3.2: UNKNOWN ---------------------------------------- When CRuby parses identifiers, it is encoding-dependent. Once the identifier is found, it determines if it starts with a uppercase or lowercase codepoint. This determines if the identifier is a constant or not. The function is charge of this is `rb_sym_constant_char_p`. For non-unicode encodings where the leading byte has the top-bit set, this relies on onigmo's `mbc_case_fold` to determine if it is a constant or not (as opposed to `is_code_ctype`). This works for almost every single codepoint in every encoding, but has one very weird edge case. In the Windows-1253 encoding for the 0xB5 byte, it's the micro sign. The micro sign, when case folded, becomes the uppercase mu character, and then the lowercase mu character, or 0xEC. This means that even though 0xB5 reports itself as being a lowercase codepoint, it gets parsed as a constant. This example might make this more clear: ``` ruby class Context < BasicObject def method_missing(name, *) = :identifier def self.const_missing(name) = :constant end encoding = Encoding::Windows_1253 character = 0xB5.chr(encoding) source = "# encoding: #{encoding.name}\n#{character}\n" result = Context.new.instance_eval(source) puts "#{encoding.name} encoding of 0x#{character.ord.to_s(16).upcase}" puts " [[:alpha:]] => #{character.match?(/[[:alpha:]]/)}" puts " [[:alnum:]] => #{character.match?(/[[:alnum:]]/)}" puts " [[:upper:]] => #{character.match?(/[[:upper:]]/)}" puts " [[:lower:]] => #{character.match?(/[[:lower:]]/)}" puts " parsed as #{result}" ``` this results in the output of: ``` Windows-1253 encoding of 0xB5 [[:alpha:]] => true [[:alnum:]] => true [[:upper:]] => false [[:lower:]] => true parsed as constant ``` To be clear, I don't think the case-folding is incorrect here (and @duerst confirms that it is correct). I believe instead that it is incorrect to use case-folding here to determine if a codepoint is uppercase or not. Note that this only impacts this one codepoint in this one encoding, so I don't believe this is actually a large-scale problem. But I found it surprising, and think we should change it. -- https://bugs.ruby-lang.org/

4 5

[ruby-core:112301] [Ruby master Bug#19426] Endless `Range#step` of object with `#succ` method does not work
by nobu (Nobuyoshi Nakada) 13 Mar '25

13 Mar '25

Issue #19426 has been reported by nobu (Nobuyoshi Nakada). ---------------------------------------- Bug #19426: Endless `Range#step` of object with `#succ` method does not work https://bugs.ruby-lang.org/issues/19426 * Author: nobu (Nobuyoshi Nakada) * Status: Open * Priority: Normal * Backport: 2.7: UNKNOWN, 3.0: UNKNOWN, 3.1: UNKNOWN, 3.2: UNKNOWN ---------------------------------------- Consider this `c` object which hash `#succ` method. ```ruby c = Struct.new(:i) do def succ; self.class.new(i+1); end def <=>(other) i <=> other.i;end end.new(0) ``` This `Range#step` works. ```ruby (c..c.succ).step(1) do |d| p d.i # 0, 1 end ``` But it fails on an endless range. ```ruby (c..).step(1) do |d| p d.i break if d.i > 3 end ``` ``` -:3:in `<=>': undefined method `i' for nil:NilClass (NoMethodError) from -:10:in `step' from -:10:in `<main>' ``` -- https://bugs.ruby-lang.org/

3 4

[ruby-core:120637] [Ruby master Bug#21032] `Module#autoload?` is slow when `$LOAD_PATH` contains a relative path
by byroot (Jean Boussier) 13 Mar '25

13 Mar '25

Issue #21032 has been reported by byroot (Jean Boussier). ---------------------------------------- Bug #21032: `Module#autoload?` is slow when `$LOAD_PATH` contains a relative path https://bugs.ruby-lang.org/issues/21032 * Author: byroot (Jean Boussier) * Status: Open * Backport: 3.1: WONTFIX, 3.2: REQUIRED, 3.3: REQUIRED, 3.4: REQUIRED ---------------------------------------- Reproduction script: ```ruby require 'benchmark' $LOAD_PATH << 'relative-path' autoload :FOO, '/tmp/foo.rb' puts Benchmark.realtime { 500_000.times do Object.autoload?(:FOO) end } ``` The above takes 2.5 to 3 seconds on my machine, but just removing `$LOAD_PATH << 'relative-path'` make it complete in 50ms. It's such a stark difference that I think it is a bug, and it cause Zeitwerk, a very popular gem, to be way slower than it should when the load path contains relative paths. I have a patch for it, that passes all tests, but I'd appreciate some eyes on it: https://github.com/ruby/ruby/pull/12562 cc @fxn -- https://bugs.ruby-lang.org/

5 8

[ruby-core:120232] [Ruby master Misc#20951] Confusing handling of timezone object's `#utc_to_local` results
by andrykonchin (Andrew Konchin) 12 Mar '25

12 Mar '25

Issue #20951 has been reported by andrykonchin (Andrew Konchin). ---------------------------------------- Misc #20951: Confusing handling of timezone object's `#utc_to_local` results https://bugs.ruby-lang.org/issues/20951 * Author: andrykonchin (Andrew Konchin) * Status: Open ---------------------------------------- I am looking into the timezone object feature (that is supported by various Time class methods) now and I am confused by the current implementation. Specifically, how a time-like object **that is not inherited from the Time class** is handled. A time-like object is returned for instance from the timezone object's `#utc_to_local` method. The documentation states that: > A Time-like object is a container object capable of interfacing with timezone libraries for timezone conversion. Also > The zone value may be an object responding to certain timezone methods, an instance of Timezone and TZInfo for example. And indeed the `TZInfo::Timezone` class works as expected. But when I try to use for time-like objects a brand new class not inherited from Time - it works incorrectly. Let's consider an example with `TZInfo::Timezone`: ```ruby require 'tzinfo' zone = TZInfo::Timezone.get("Europe/Kiev") # UTC+2 time = Time.now.utc puts time.to_i # 1734107333 puts Time.now(in: zone) # 2024-12-13 18:28:53 +0200 puts zone.utc_to_local(time) # 2024-12-13 18:28:53 +0200 puts zone.utc_to_local(time).to_i # 1734107333 ``` And now an example with a brand new class. I make an assumption, that as far as `zone.utc_to_local(time).to_i` doesn't change Unix timestamp (it equals `time.to_i`, that's 1734107333), so in a new class also `#utc_to_local` should return not modified value too. ```ruby TimeObj = Struct.new(:year, :mon, :mday, :hour, :min, :sec, :isdst, :to_i) zone_obj = Object.new def zone_obj.utc_to_local(t) TimeObj.new(t.year, t.mon, t.mday, t.hour + 2, t.min, t.sec, t.isdst, t.to_i) # <=== adjust hours (`hours + 2`) to match "Europe/Kiev" timezone (that's UTC+2) end ``` Unfortunately it produces incorrect result: ```ruby puts Time.now(in: zone_obj) # 2024-12-13 18:28:53 +0000 <====== wrong UTC offset puts zone_obj.utc_to_local(time) # #<struct TimeObj year=2024, mon=12, mday=13, hour=18, min=28, sec=53, isdst=false, to_i=1734107333> puts zone_obj.utc_to_local(time).to_i # 1734107333 <===== the same Unix timestamp ``` So now result time object has wrong utc offset - `+0000` instead of `+0200`. Okey, so probably Unix timestamp should be adjusted as well. Let's check: ```ruby def zone_obj.utc_to_local(t) TimeObj.new(t.year, t.mon, t.mday, t.hour + 2, t.min, t.sec, t.isdst, t.to_i + 2 * 60 * 60) # <===== adjust #to_i as well so it returns timestamp + 2 hours end puts Time.now(in: zone_obj) # 2024-12-13 18:28:53 +0200 <======= correct UTC offset puts zone_obj.utc_to_local(time) # #<struct TimeObj year=2024, mon=12, mday=13, hour=18, min=28, sec=53, isdst=false, to_i=1734114533> puts zone_obj.utc_to_local(time).to_i # 1734114533 <====== different Unix timestamp ``` Now we have correct UTC offset `+0200` despite `zone_obj.utc_to_local(time).to_i` returns not original offset but an adjusted one. I assume the difference is caused by a special treatment of time-like object inherited from the Time class. So its `utc_offset` property is used only. But for all the other classes the `#to_i` is used instead. ```ruby zone.utc_to_local(time).class.ancestors # => [TZInfo::TimeWithOffset, TZInfo::WithOffset, Time, Comparable, Object, PP::ObjectMixin, Kernel, BasicObject] ``` This difference is confusing so I think it makes sense either to document it (I mean to document that `#to_i` should return adjusted value for non-related to Time classes) in case it's intentional or to change behaviour for non-related to Time classes and rely not on `#to_i` to calculate UTC offset but on difference in `sec`/`min`/`hours` values otherwise. -- https://bugs.ruby-lang.org/

4 6

[ruby-core:120817] [Ruby master Bug#21096] Ruby hangs up when compiling for bytecode on AArch64 emulated by QEMU
by midnight (Sarun R) 11 Mar '25

11 Mar '25

Issue #21096 has been reported by midnight (Sarun R). ---------------------------------------- Bug #21096: Ruby hangs up when compiling for bytecode on AArch64 emulated by QEMU https://bugs.ruby-lang.org/issues/21096 * Author: midnight (Sarun R) * Status: Open * ruby -v: ruby 3.4.1 (2024-12-25 revision 48d4efcb85) +PRISM [x86_64-linux] * Backport: 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN, 3.4: UNKNOWN ---------------------------------------- Hello, We are experiencing issues when using [Bootsnap](https://github.com/Shopify/bootsnap) for production container image building, specifically when running `bundle exec bootsnap precompile --gemfile` on an emulated ARM64 environment on AMD64 hosts. Here are more details: * Bootsnap is a Ruby gem for bytecode caching, which speeds up loading times. It achieves this by calling `RubyVM::InstructionSequence.compile_file` and `RubyVM::InstructionSequence.load_from_binary`. * When running `bundle exec bootsnap precompile --gemfile` in the environment described (using QEMU to emulate the AArch64 instruction set), the process can compile and generate some bytecode but eventually hangs. * The hang seems random, occurring at different points during the process. * According to [a Rails GitHub issue](https://github.com/rails/rails/issues/54039), the issue affects both `Docker` and `containerd`, as well as native Linux and macOS environments. * From [a Bootsnap GitHub issue](https://github.com/Shopify/bootsnap/issues/495), the problem likely doesn't occur in Ruby `3.2.6` but appears in later versions. * So far, the common factors observed are: * Emulated ARM64 on AMD64 CPUs (likely using QEMU) * Newer Ruby versions (e.g., 3.3 and 3.4) As a user, I don't have the expertise to debug the issue but am willing to gather more information if provided with sufficient instructions. -- https://bugs.ruby-lang.org/

3 5

[ruby-core:120811] [Ruby master Bug#21095] Prefer `uname -n` over `hostname` in tests.
by ioquatix (Samuel Williams) 08 Mar '25

08 Mar '25

Issue #21095 has been reported by ioquatix (Samuel Williams). ---------------------------------------- Bug #21095: Prefer `uname -n` over `hostname` in tests. https://bugs.ruby-lang.org/issues/21095 * Author: ioquatix (Samuel Williams) * Status: Closed * Backport: 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: REQUIRED, 3.4: REQUIRED ---------------------------------------- It turns out that `hostname`, while a defacto standard, is not actually a standard in any official sense. On Linux, it's distributed as part of the `inettools` package, and while generally available on other platforms (BSD, Windows, MacOS), it isn't actually part of any standard. The `name(1)` system call and `uname(2)` command ARE standardised by POSIX and the Open Group, and are included in most base systems without the need to install extra packages (e.g. `inettools` on Linux). As such, I was requested by the Arch Linux Ruby maintainer, to prefer using `uname -n` as they would like to drop the dependency on `inettools`. See <https://gitlab.archlinux.org/archlinux/packaging/packages/inetutils/-/issue…> for more context. I've been asked if this can be back ported to 3.3 and 3.4 and while it's not strictly a bug, it will reduce friction in the distribution channels, so I'd like to propose that we backport to 3.4 and if possible 3.3 too. -- https://bugs.ruby-lang.org/

4 9

[ruby-core:120733] [Ruby master Bug#21046] Backport: TLS fix for ARM64
by nobu (Nobuyoshi Nakada) 08 Mar '25

08 Mar '25

Issue #21046 has been reported by nobu (Nobuyoshi Nakada). ---------------------------------------- Bug #21046: Backport: TLS fix for ARM64 https://bugs.ruby-lang.org/issues/21046 * Author: nobu (Nobuyoshi Nakada) * Status: Closed * Backport: 3.1: REQUIRED, 3.2: REQUIRED, 3.3: REQUIRED, 3.4: REQUIRED ---------------------------------------- commit:f7059af50a31a4d27a04ace0beadb60616f3f971 Use no-inline version `rb_current_ec` on Arm64 The TLS across .so issue seems related to Arm64, but not Darwin. -- https://bugs.ruby-lang.org/

5 5

[ruby-core:116369] [Ruby master Feature#20202] Memoized endless method definitions
by matheusrich (Matheus Richard) 27 Feb '25

27 Feb '25

Issue #20202 has been reported by matheusrich (Matheus Richard). ---------------------------------------- Feature #20202: Memoized endless method definitions https://bugs.ruby-lang.org/issues/20202 * Author: matheusrich (Matheus Richard) * Status: Open * Priority: Normal ---------------------------------------- I propose introducing a shorthand for memoized endless method definitions: ```rb class Foo def bar ||= :memoized_value # It should behave like def bar = (@bar ||= :memoized_value) end ``` Not only is this shorter and (IMO) a natural follow-up for endless method definitions, but it's also a common pattern on Ruby codebases. It's very useful to decompose responsibilities into several objects: ```rb class User def notifications_enabled? = settings.notifications? def enable_notifications = (settings.notifications = true) def disable_notifications = (settings.notifications = false) private def settings = @settings ||= Settings.new(self) end class User::Settings attr_writer :notifications def initialize(user) @user = user @notifications = false end def notifications? = @notifications end u = User.new u.notifications_enabled? # => false u.enable_notifications u.notifications_enabled? # => true ``` In the example, the `settings` method could be rewritten as ```rb def settings ||= Settings.new(self) ``` -- https://bugs.ruby-lang.org/

4 5

[ruby-core:120250] [Ruby master Feature#20953] Array#fetch_values vs #values_at protocols
by zverok (Victor Shepelev) 26 Feb '25

26 Feb '25

Issue #20953 has been reported by zverok (Victor Shepelev). ---------------------------------------- Feature #20953: Array#fetch_values vs #values_at protocols https://bugs.ruby-lang.org/issues/20953 * Author: zverok (Victor Shepelev) * Status: Open ---------------------------------------- I believe that the user might expect `#fetch_values` to be a stricter version of `#values_at`, confirming to the same protocol for arguments. But the current implementation for `#fetch_values` is simpler: ```ruby [1, 2, 3, 4, 5].values_at(0, 3..4) #=> [1, 4, 5] [1, 2, 3, 4, 5].fetch_values(0, 3..4) # TypeError: in 'Array#fetch': no implicit conversion of Range into Integer ``` I believe aligning the implementations would lessen confusion (even if it makes `#fetch_values` implementation somewhat less trivial). The practical example of usefulness: ```ruby HEADERS = %w[Name Department] def table_headers(rows) HEADERS.fetch_values(...rows.map(&:size).max) { '<unknown'> } # Or, alternatively: # HEADERS.fetch_values(...rows.map(&:size).max) { raise ArgumentError, "No header defined for column #{it + 1}" } end table_headers([ ['John'], ['Jane'], ]) #=> ["Name"] table_headers([ ['John'], ['Jane', 'Engineering'], ]) #=> ["Name", "Department"] table_headers([ ['John', 'Accounting', 'Feb 24'], ['Jane', 'Engineering'], ]) #=> ["Name", "Department", "<unknown>"] # or ArgumentError No header defined for column 3 ``` (Obviously, we can use `fetch_values(*(0...max_row_size))` as an argument, but it feels like an unjustified extra work when `values_at` already can do this.) -- https://bugs.ruby-lang.org/

3 10