DevMeeting-2026-02-12

https://bugs.ruby-lang.org/issues/21839

DateTime and location

2026/02/12 (Wed) 13:00-17:00 JST @ Online

Next Date

2026/03/17 (Tue) 13:00-17:00 JST @ Online

Announce

About release timeframe

Ordinary tickets

[Feature #19107] Allow trailing comma in method signature (k0kubun)

We provided use cases in response to Matz’s question, but no progress for 3 years. Can we clarify why we’re not adding this?

Discussion:

matz: I was persuaded.
mame: def foo(&block,) should be prohibited, right?
matz: Do not allow trailing comma after def foo(&block,) and def foo(...,)
nobu: carrige return is requred? def foo(a,) seems strange.
matz: no problem to allow it.

Conclusion:

matz: accepted. I replied

[Feature #21800] New API to efficiently scan directories efficiently (byroot)

It’s common for development tools to need to recursively scan the file system, but it’s slower than it should be in Ruby because the API impose N+1 stat syscalls.
This would benefit many popular gems such as Zeitwerk, rubocop, etc.
Previous discussion had concern about changing the behavior of Dir.each_child and other existing methods.
I propose:
Dir.scan(path) { |entry_name, entry_type| }
Dir.scan(path) # => [[entry_name, entry_type], ...]
The type is just a symbol, similar to File::Stat#ftype
In case of DT_UNKNOWN, Ruby issue a lstat to obtain the real type (important for portability).

Preliminary discussion:

File.stat("/etc/passwd").ftype #=> "file"

mame: I am not sure if it is consisitent between File::Stat#ftype and readdir’s struct dirent.d_type

Discussion:

matz: The name Dir.scan looks good
akr: it will yield "." and "..", or not? Excluding them looks useful.
matz: ok.
akr: automatically converting DT_UNKNOWN to File::Stat#ftype’s result looks good
akr: but it may bring extra overhead, so we need to provide a way to disable the automatic conversion: Dir.scan("dir", implicit_lstat: false) { ... } or Dir.scan("dir", lstat: false) { ... } or something
mame: What object should be yielded? Currently File::Stat#ftype returns a new string.
akr: I want to use symbols: :file, :directory, :blockSpecial, :link, etc.
akr: Or use method name convension: :symlink (not :link) from File::Stat#symlink?, :pipe from File::Stat#pipe?, :blockdev from File::Stat#blockdev?, etc.
mame: Or create a new name set? :file :directory :chardev :blockdev :fifo :symlink :socket :unknown
matz: blockSpecial, … is from tcl.
matz: I want not to introduce a new name convension, though I admit :symlink and :chardev are definitely better than :link and :characterSpecial…

Conclusion:

matz: Dir.scan is approved (including the name)
matz: automatic conversion by lstat is also approved
matz: disabling automatic conversion is postponed until real-world use case is shown
matz: it should yield a symbol version of File::Stat#ftype like :blockSpecial
matz: "." and ".." should be excluded

[Feature #21788] Promote Thread::Monitor to a core class (byroot)

Previous discussion ended on the discussion of whether a recursive Mutex would be enough
- Yes it would, but that’s what Monitor is used for in a ton of code, I don’t think it’s worth causing churn here.
Monitor is about as useful as Mutex and yet one is a core class and the other is a “stdlib” extension.
Bringing it into core allow to make is about as fast as mutex, while it is currently ~20% slower.
I don’t personally think MonitorMixin should be made core though. It can remain in monitor.rb.

Discussion:

ko1: Is it ok to define Thread::Monitor instead of ::Monitor?
ko1: most of Ruby script requires monitor.rb, so no name conflicts.

Conclusion:

matz: accept.
matz: keep the name monitor.rb. Do not move to lib/monitor_mixin.rb

[Feature #21863] Add `RBASIC_FLAGS()` and `RBASIC_SET_FLAGS()` (eregon)

Sounds OK to add? (I can make a PR, it’s trivial)
If not, what alternative? Deprecate RBasic->flags?

Preliminary discussion:

2007-05-30 /srv/gems/nodewrap-0.5.0/ext/nodewrap.c:  RBASIC(module)->flags = NUM2INT(flags);

Discussion:

nobu: why are RB_FL_SET, RB_FL_TEST, RB_FL_UNSET not enough? They can be used for bit flags in flags field. Other data in flags (such as object type) must not be tweaked.

Conclusion:

check why the proposed macros are really needed

[Bug #21855] Bundle `win32-registry` or implement it without `fiddle` (hsbt)

Is it okay to extract that at Ruby 4.1 without warning cycle?

Discussion:

Extract win32-registry at Ruby 4.2 and warn it at Ruby 4.1.
Extract win32-registry at Ruby 4.1 without warning <= chosen by naruse
Extract win32-registry at Ruby 4.1 and backport that warning to Ruby 4.0.x.

Conclusion:

hsbt: I will ask usa-san to Plan 2

[Feature #21835] Unbundle bunled gems like net-ftp (hsbt)

How about this?

Preliminary discussion:

mame: I want prime :-)

Discussion:

Conclusion:

hsbt: I will remove net-ftp and net-pop

[Misc #21630] Suggest @Earlopain for core contributor (earlopain)

Could this be reconsidered? Recently I’ve made changes/wanted to make changes that require small updates in ruby/ruby
- https://github.com/ruby/ruby/pull/15914
- https://github.com/ruby/ruby/pull/15999
- https://github.com/ruby/ruby/pull/16052
- https://github.com/ruby/ruby/pull/16085
I have to wait on others to fix CI/help me make the changes I want to do in prism
Sometimes prism CI doesn’t tell me that a change will break ruby CI (we run test-all only)
I would especially like to unbreak CI when I accidentally break it
It would also allow me to collaborate with parse.y maintainers in the same PR like https://github.com/ruby/ruby/pull/15498 (I think)

Discussion:

naruse: I understand earlopain’s trouble, but kddnewton (as a mentor) should help them
matz: Who can be a mentor? I am positive for the committer proposal itself but I also understand the mentoring concern

Conclusion:

discussed but no conclusion

[Feature #21871] Add `Module#undef_const` (jeremyevans0)

I’d like to add this method, which offers for constants the same behavior that undef_method offers for methods.
This allows hiding specific constants inside a namespace, to allow for encouraging safer coding patterns in some cases.
My expected use case for this is to assist in avoiding IDOR (insecure direct object reference) vulnerabilities.
I’ve implemented support for this in a pull request.
Is the feature itself OK?
If so, is the implementation in the pull request OK?

Preliminary discussion:

CONST = 1

class C
  undef_const :CONST
  begin
    p CONST
  rescue => e
    p e #=> #<NameError: uninitialized constant C::CONST>
  end

  p ::CONST #=> 1
end

https://ruby.github.io/play-ruby/?run=21814774663&code=CONST+%3D+1%0A%0Aclass+C%0A++undef_const+%3ACONST%0A++begin%0A++++p+CONST%0A++rescue+%3D%3E+e%0A++++p+e+%23%3D%3E+%23%3CNameError%3A+uninitialized+constant+C%3A%3ACONST%3E%0A++end%0A%0A++p+%3A%3ACONST+%23%3D%3E+1%0Aend%0A%0A

Discussion:

matz: I feel a gap between the use case and the proposal…

Conclusion:

matz: I will reply

[Misc #21872] `-S` with directory separator (nobu)

Its origin Perl’s -S doesn’t search the script with directory separator from $PATH.
sh seems similar to Perl.
Is this behavior intentional?

Conclusion:

nobu: I directly talked with matz and already merged it, discussion is no longer needed

[Feature #21785] Add LEB128 pack / unpack (tenderlovemaking)

Are we OK to merge now? I think yes, but want to check

Conclusion:

matz: go ahead

[Feature #21796] Add pack directive to return offset `^` (tenderlovemaking)

This would be useful with LEB128, as well as other directives that @nobu mentions in the ticket

Discussion:

bytes = "\x01\x02\x03"
offset = 0
leb128_value1, offset = bytes.unpack("R^", offset: offset) #=> 1
leb128_value2, offset = bytes.unpack("R^", offset: offset) #=> 2
leb128_value3, offset = bytes.unpack("R^", offset: offset) #=> 3

bytes = "\x01\x02\x03"
offset = 0
leb128_v1, leb128_v2, leb128_v3, offset = bytes.unpack("RRR^", offset: offset) #=> 1, 2, 3, (offset) 3

Conclusion:

matz: go ahead

[Feature #15330] Introduce `autoload_relative` (ioquatix)

autoload_relative :MyConstant, "my_constant.rb"

# Similar to:
autoload :MyConstant, File.expand_path("my_constant.rb", __dir__)

Preliminary discussion:

More useful than autoload itself as it would typically be used to load relative files within a single project.
- For example, all of these would be more correct if using autoload_relative: https://github.com/rack/rack/blob/550de048886a20c79ff33300556b10adcbd7426e/lib/rack.rb#L17-L63
- Rails folks specfically said: “Rails plans to implement its reload feature that today is based on const_missing using autoload, so we would prefer if that feature would not be removed, and instead improved.”
- Xavier (fxn, who created zeitwerk) also requested it: https://bugs.ruby-lang.org/issues/18841
This is a desirable feature requested by many respectable projects. Can we introduce it?
PR: https://github.com/ruby/ruby/pull/16148

Conclusion:

matz: I now accept. Go ahead

matz reminder

[Feature #6012] Proc#source_location also return the column
- mame: we need revert for now?
[Feature #21795] Methods for retrieving ASTs
- matz said “I’ll write my comments on issue, but not concluded yet.”

[Bug #21844] Inconsistent ArgumentError message for Data::define.new

C = Data.define(:a, :b)
o = C.new("a" => 1, "b" => 2) #=> #<data C a=1, b=2>

o = C.new(0 => 1); o          #=> missing keywords: :a, :b
o = C.new(0 => 1, -2 => 2); o #=> #<data C a=2, b=nil>

matz: looks a bug. Let’s prohibit Strings and Integers
ko1: How about Struct.new?
mame: It allows String keys and Integer keys.
nobu: should we prohibit String keys?
Matz: should be prohibited. I’ll reply.

[Feature #21768] Remove deprecated functions

https://github.com/ruby/ruby/pull/15447
https://github.com/nobu/ruby/tree/deprecate-rdata

[Feature #19979] Allow methods to declare that they don’t accept a block via &nil [Feature #21773] Support for setting encoding when a block is passed to Net::HTTPResponse.read_body

Please send a PR

[Feature #21781] Add fetch_values method on ENV

matz: go ahead

[Bug #21808] Inconsistency in support of additional newlines with boolean logical operators on new line

1

.times {} # syntax error found (SyntaxError)

matz: It is by design

[#21797] Make `Etc.nprocessors` cgroup-aware on Linux

[#21813] Add `[:forward, :...]` symbol tuple to indicate forwarding arguments when calling `Method#parameters`

def foo(*, **, &)
  p method(__method__).parameters
  #=> [[:rest, :*], [:keyrest, :**], [:block, :&]]

  binding.eval "print(#{(method(__method__).parameters.dig(0,1))})" 
  #== print(*)
end

def bar(...)
  p method(__method__).parameters
  #=> actual: [[:rest, :*], [:keyrest, :**], [:block, :&]]
  #=> expected: [[:forwarding, :...]]
  
  # use case
  binding.eval "print(#{(method(__method__).parameters.dig(0,1))})" 
  #== print(*) <= does not work
end

bar

matz: I will reject

[#21822] Expose Return Value in the ensure Block

similar ticket: https://bugs.ruby-lang.org/issues/18083

begin
  ...
ensure => ret, err
  ...
end

akr: ret == nil can distinguish between normal return with nil and non-local exit
akr: it can be written as follows.

begin
  ret = nil
  ret = begin
    ...
  end
ensure
  LOGGER.debug "return value: #{ret}"
end

matz: I will reject

[#21825] Status of the universal parser implementing the Prism API

[#21767] Consider procs which `self` is Ractor-shareable as Ractor shareable

matz: no objection.

[#21833] Switch default hash from SipHash13 to XXH3?

[#21851] Performance difference between / operator and fdiv

3 / 0x20_0000_0000_0001.to_f #=> 3.3306690738754696e-16
3.fdiv 0x20_0000_0000_0001   #=> 3.330669073875469e-16

0x20_0000_0000_0001.to_f == 0x20_0000_0000_0000.to_f #=> true

[#21858] `Kernel#Hash` considers `to_h` too

# current
Hash([1, 2]) #=> can't convert Array into Hash (TypeError)

# after the change
Hash([1, 2]) #=> wrong element type Integer at 0 (expected array) (TypeError)

S = Struct.new(:a, :b)
obj = S.new(1, 2)

# current
Hash(obj) #=> can't convert S into Hash

# after the change
Hash(obj) #=> {a: 1, b: 2}

matz: I will reply

[#21272] Class.new doesn’t trigger :class TracePoint

$ ruby -e 'TracePoint.new { p it }.enable { 1 + 2 }'
#<TracePoint:b_call -e:1>
#<TracePoint:line -e:1>
#<TracePoint:c_call '+' -e:1>
#<TracePoint:c_return '+' -e:1>
#<TracePoint:b_return -e:1>

[#21870] Regexp: Warnings when using slightly overlapping \p{…} classes

$VERBOSE = true
/[\p{Word}||\p{S}]/ #=> warning: character class has duplicated range: /[\p{Word}||\p{S}]/

p /[a||b]/ =~ "|" #=> 0
#=> warning: character class has duplicated range: /[a||b]/

p /[A-z_]/
p /[\p{Word}A-Z]/ # tompng: I want to notice that this regexp has something wrong
p /[A-Z\p{Word}]/
p /[#asdavasada#]/

Warning[:regexp_lint] = true

regparse.c:    onig_syntax_warn(env, "character class has '%s' without escape", c);
regparse.c:    onig_syntax_warn(env, "regular expression has '%s' without escape", c);
regparse.c:    onig_syntax_warn(env, "character class has duplicated range: %04x-%04x", from, to);
regparse.c:    onig_syntax_warn(env, "character class has duplicated range");
regparse.c:  onig_syntax_warn(env, "Unknown escape \\%c is ignored", c);
regparse.c:        onig_syntax_warn(env, "invalid Unicode Property \\%c", c);
regparse.c:          onig_syntax_warn(env, "invalid back reference");
regparse.c:          onig_syntax_warn(env, "invalid subexp call");
regparse.c:        onig_syntax_warn(env, "invalid Unicode Property \\%c", c);
regparse.c:            onig_syntax_warn(env, "regular expression has redundant nested repeat operator '%s'",
regparse.c:            onig_syntax_warn(env, "nested repeat operator '%s' and '%s' was replaced with '%s' in regular expression",

DevMeeting-2026-02-12

DateTime and location

Next Date

Announce

About release timeframe

Ordinary tickets

[Feature #19107] Allow trailing comma in method signature (k0kubun)

[Feature #21800] New API to efficiently scan directories efficiently (byroot)

[Feature #21788] Promote Thread::Monitor to a core class (byroot)

[Feature #21863] Add RBASIC_FLAGS() and RBASIC_SET_FLAGS() (eregon)

[Bug #21855] Bundle win32-registry or implement it without fiddle (hsbt)

[Feature #21835] Unbundle bunled gems like net-ftp (hsbt)

[Misc #21630] Suggest @Earlopain for core contributor (earlopain)

[Feature #21871] Add Module#undef_const (jeremyevans0)

[Misc #21872] -S with directory separator (nobu)

[Feature #21785] Add LEB128 pack / unpack (tenderlovemaking)

[Feature #21796] Add pack directive to return offset ^ (tenderlovemaking)

[Feature #15330] Introduce autoload_relative (ioquatix)

[#21797] Make Etc.nprocessors cgroup-aware on Linux

[#21813] Add [:forward, :...] symbol tuple to indicate forwarding arguments when calling Method#parameters

[#21822] Expose Return Value in the ensure Block

[#21825] Status of the universal parser implementing the Prism API

[#21767] Consider procs which self is Ractor-shareable as Ractor shareable

[#21833] Switch default hash from SipHash13 to XXH3?

[#21851] Performance difference between / operator and fdiv

[#21858] Kernel#Hash considers to_h too

[#21272] Class.new doesn’t trigger :class TracePoint

[#21870] Regexp: Warnings when using slightly overlapping \p{…} classes

[Feature #21863] Add `RBASIC_FLAGS()` and `RBASIC_SET_FLAGS()` (eregon)

[Bug #21855] Bundle `win32-registry` or implement it without `fiddle` (hsbt)

[Feature #21871] Add `Module#undef_const` (jeremyevans0)

[Misc #21872] `-S` with directory separator (nobu)

[Feature #21796] Add pack directive to return offset `^` (tenderlovemaking)

[Feature #15330] Introduce `autoload_relative` (ioquatix)

[#21797] Make `Etc.nprocessors` cgroup-aware on Linux

[#21813] Add `[:forward, :...]` symbol tuple to indicate forwarding arguments when calling `Method#parameters`

[#21767] Consider procs which `self` is Ractor-shareable as Ractor shareable

[#21858] `Kernel#Hash` considers `to_h` too