Feb 2025 Meeting

calendar_today February 13, 2025 confirmation_number 27 tickets discussed Source English

DevMeeting-2025-02-13

https://bugs.ruby-lang.org/issues/21019

DateTime and location

  • 2025/02/13 (Thu) 13:00-17:00 JST @ Online

Next Date

  • 2025/03/13 (Thu) 13:00-17:00 JST @ Online

Announce

About release timeframe

  • 3.4.2?
  • k0kubun will take care of 3.4 branch sooner.

Ordinary tickets

[Feature #21116] Extract RJIT as a third-party gem (k0kubun)

  • Are we okay with removing RJIT (--rjit) from the Ruby core?

Discussion:

  • k0kubun: I think I can separate RJIT from the core.
  • k0kubun: this should benefits core maintenability.
  • n0kada: how should RJIT be invoked, when it gets out of the core (there cannot be command line options?)
  • k0kubun: while enabling JIT at process bootstrap gets impossible, RJIT can also be enabled later, and that should be okay.
  • hsbt: what happens for ruby --rjit
  • k0kubun: let’s just delete. it’s still experimental so far.

Conclusion:

  • matz: no reason to object.

[Feature #21084] Declare objects have weak references (peterzhu2118)

  • The current way of marking weak references uses rb_gc_mark_weak(VALUE *ptr).
  • This presents challenges because Ruby’s GC is incremental, meaning that if the ptr changes (e.g. realloc’d or free’d), then we could have an invalid memory access.
  • This also overwrites *ptr = Qundef if *ptr is dead, which prevents any cleanup to be run (e.g. freeing memory or deleting entries from hash tables).
  • This ticket proposes rb_gc_declare_weak_references which declares that an object has weak references and calls a cleanup function after marking, allowing the object to clean up any memory for dead objects.

Preliminary discussion:

struct D {
  VALUE v1;
  int i;
  VALUE *values; // weak
}

weaks = []
rb_gc_mark_weak(VALUE *ptr)
{
  v = *ptr;
  weaks << ptr;
}

after_mark() {
  weaks.each{|ptr|
    unless *ptr is live?
      *ptr = Qundef
    end
  }
}

mark(D *d) {
  rb_gc_mark(d->v1);
  rb_gc_mark_weak(&d->values[0]);
  ...
  d->values = realloc(d->values);
}

Discussion:

  • peter: I found rb_gc_mark_weak() is not perfect. (1) the passed pointer can move during GC which is very difficuly to be done properly. (2) references to dead objects should be updated using Qundef, which prevents deallocation of such storages.
  • peter: I’m proposing an internal API that registers how to deallocate that reference at the time it is created.
  • mame: I heard ko1 wants some non-declaratibe API.
  • ko1: embedding number of weak references into flags is too much because there should be only one or very limited number of weak references in a process.
  • peter: also as a separate project I’m moving all those GC-related flags out of each objects.
  • ko1: how about declare this property in marking side.
  • peter: weak maps are T_DATA. marking function doesn’t know the self.
  • ko1: oh… I see… :confused:

Conclusion:

  • ko1: overall I understand the needs. Let myself and peter discuss about in-depth API design.

[Bug #20998] rb_str_locktmp() changes flags of frozen strings and string literals (eregon)

  • Current behavior on frozen strings is buggy (should not mutate the flags of possibly-shared frozen strings).
  • What should be the behavior for rb_str_locktmp() on frozen strings? No-op or raise error (e.g. FrozenError)?

Preliminary discussion:

  • locktmp
    • prohibit write
a = +"foo"
b = a.dup
b.freeze # a -> shared <-b
b.locktmp

s = ...
puts s

Discussion:

  • mame: rb_str_locktmp() is broken when applied to a frozen string.
  • akr: it is already problematic when multiple threads lock a string at once?
  • nobu: that doesn’t happen. rb_io_write() does not use this function now (it did before though).
  • shyouhei: do we need a fully reentrant reader-writer lock for locktmp? sounds overkill.
  • akr: frozen string need not be locked at the first place?
  • nobu: frozen strings also gets moved when compacted.
  • akr: that can be a different story here because GC doesn’t happen during locktemp is in effect.
  • ko1: what should happen when a string gets frozen, under locktmp is in effect?
  • mame: not possible?
  • nobu: rb_str_modifiable() already checks that situation (and raises).

Conclusion:

  • nobu: eregon is right here. rb_str_locktmp() need to raise when its receiver is frozen.
  • shyouhei: we acknowledge that there is currently no way for an extension library to reliably obtain a char* from a string (because objects are compacted, and rb_str_locktmp() is made unavailable conditionally due to this conclusion). We might need separate issues / proposals to fix the gap.

[Bug #21102] Unexpected encoding when concatenating ASCII string with ASCII compatible string (toy)

  • Should ascii_string + utf8_string return an ASCII or UTF-8 string when both have 7bit coderange?

Conclusion:

  • naruse: We will not change this unless there is any good reason instead of “surprising”

net-http 205 Do not supply a default content type

  • Current net-http will use 'application/x-www-form-urlencoded' for default content type. AWS service have some issues with this behavior. Can we remove it from default content type?

Discussion:

  • mame: I guess AWS SDK is getting errors due to net/http’s behaviour.
  • akr: why is it difficult for the SDK to just add necessary headers?
  • shyouhei: in case of S3 the content type can untimaetly be unknown.
  • mame: maybe there is no way to force net/http preventing from adding this header.
  • akr: if we could ignore backward compatibility yes, changing this to octet-stream is the right thing to do.
  • shyouhei: why was x-www-form-urlencoded chosen to be the encoding right now?
  • knu: maybe at the time net/http was made there was no such thing like PUT; there were either GET or POST and POST was basically HTML forms back then.

Conclusion:

  • hsbt: OK, let’s delete suppy_default_content_type

[Bug #21049] Reconsider handling of the numbered parameters and “it” parameter in Binding#local_variables

Conclusion:

  • matz: yes, the proposed behaviour matches my expectation. accepted.
  • matz: Binding#numbered_parameters etc. need separate ticket before being accepted.

[#21082] Alias it to its

  • matz: Objection. I will reject

[#21029] Prism behavior for defined? (;x) differs

$ ruby --parser=prism -e 'p defined? ( ;x)'
nil => "expression"
$ ruby --parser=prism -e 'p defined? (();x)'
"expression"
$ ruby --parser=prism -e 'p defined? (nil;x)'
"expression"
$ ruby --parser=prism -e 'p defined? (x;x)'
"expression"
$ ruby --parser=prism -e 'def f(*);end; p defined? f(x)'
nil

[#21094] Module#set_temporary_name does not affect a name of a nested module

# similar code but it shows correct name
c = Class.new do |c|
  self.class_eval %q{
    class B
    end
  }
end

p c::B.name "#<Class:0x0000028038e52260>::B"
A = c
p A.name    #=> A
p A::B.name #=> A::B
  • matz: it should be fixed

[#21035] Clarify or redefine Module#autoload? and Module#const_defined?

# 1.rb
module M
  autoload :Foo, './2'

  constants            # => [:Foo] (as expected)
  const_defined?(:Foo) # => true (as expected)
  autoload?(:Foo)      # => 'foo' (as expected)
 
  Foo
end

# 2.rb
module M
  p constants            # => [:Foo] (as expected)
  p const_defined?(:Foo) # => false (NOT expected)
  p autoload?(:Foo)      # => nil (NOT expected)

  class Foo
  end
end

[#21105] Improve Ruby Stack Trace to Include Exact Error Position (Column Number)

[#20957] RangeError on Array#values_at with negative ranges

[0, 1, 2, 3].values_at(10)       #=> [nil]
[0, 1, 2, 3].values_at(10..10)   #=> [nil]
[0, 1, 2, 3].values_at(-10)      #=> [nil]
[0, 1, 2, 3].values_at(-10..-10) #=> 'Array#values_at': -10..-10 out of range (RangeError)
[0, 1, 2, 3].values_at(-5..-5)   #=> RangeError
[0, 1, 2, 3].values_at(-5..-4)   #=> RangeError
[0, 1, 2, 3].values_at(-5...-4)  #=> RangeError
[0, 1, 2, 3].slice(-5..-4)       #=> nil (why?)
[0, 1, 2, 3].values_at(-4..-4)   #=> [0]
[0, 1, 2, 3].values_at(-4...-4)  #=> []

% ruby -e 'a = [0,1,2,3]
-6.upto(6) {|i|
  -6.upto(6) {|j|
    x = begin a.values_at(i..j) rescue :err end
    print x.inspect.gsub(/\s/, ""); print "\t"
}
puts
}
'|lineup 
:err :err :err :err  :err    :err      :err :err  :err    :err      :err          :err              :err                 
:err :err :err :err  :err    :err      :err :err  :err    :err      :err          :err              :err                 
[]   []   [0]  [0,1] [0,1,2] [0,1,2,3] [0]  [0,1] [0,1,2] [0,1,2,3] [0,1,2,3,nil] [0,1,2,3,nil,nil] [0,1,2,3,nil,nil,nil]
[]   []   []   [1]   [1,2]   [1,2,3]   []   [1]   [1,2]   [1,2,3]   [1,2,3,nil]   [1,2,3,nil,nil]   [1,2,3,nil,nil,nil]  
[]   []   []   []    [2]     [2,3]     []   []    [2]     [2,3]     [2,3,nil]     [2,3,nil,nil]     [2,3,nil,nil,nil]    
[]   []   []   []    []      [3]       []   []    []      [3]       [3,nil]       [3,nil,nil]       [3,nil,nil,nil]      
[]   []   [0]  [0,1] [0,1,2] [0,1,2,3] [0]  [0,1] [0,1,2] [0,1,2,3] [0,1,2,3,nil] [0,1,2,3,nil,nil] [0,1,2,3,nil,nil,nil]
[]   []   []   [1]   [1,2]   [1,2,3]   []   [1]   [1,2]   [1,2,3]   [1,2,3,nil]   [1,2,3,nil,nil]   [1,2,3,nil,nil,nil]  
[]   []   []   []    [2]     [2,3]     []   []    [2]     [2,3]     [2,3,nil]     [2,3,nil,nil]     [2,3,nil,nil,nil]    
[]   []   []   []    []      [3]       []   []    []      [3]       [3,nil]       [3,nil,nil]       [3,nil,nil,nil]      
[]   []   []   []    []      []        []   []    []      []        [nil]         [nil,nil]         [nil,nil,nil]        
[]   []   []   []    []      []        []   []    []      []        []            [nil]             [nil,nil]            
[]   []   []   []    []      []        []   []    []      []        []            []                [nil]

[#20953] Array#fetch_values vs #values_at protocols

  • matz: if beginless range or endless range is passed, fetch_values should raise a RangeError consistently
[1, 2, 3].values_at(0..)     #=> [1, 2, 3]
                  # ~~~ -> values_at(0..ary.length-1)
[1, 2, 3].values_at(0..4)    #=> [1, 2, 3, nil, nil]
[1, 2, 3].fetch_values(0..)  #=> IndexError
[1, 2, 3].fetch_values(4..)  #=> IndexError
[1, 2, 3].fetch_values(4..5) #=> IndexError
[1, 2, 3].fetch_values(4..) { true } #=> IndexError
[1, 2, 3].fetch_values(-2)   #=> [2]
[1, 2, 3].fetch_values(-2..) #=> IndexError
[1, 2, 3].fetch_values(-3..) #=> IndexError
[1, 2, 3].fetch_values(-4..) #=> IndexError
[1, 2, 3].fetch_values(-4..) { true } #=> IndexError
[1, 2, 3].fetch_values(..1)  #=> IndexError
[1, 2, 3].fetch_values(..-2) #=> IndexError
[1, 2, 3].fetch_values(..-4) #=> IndexError
[1, 2, 3].fetch_values(..-4) { true } #=> IndexError
[1, 2, 3].fetch_values(..3)  #=> IndexError
[1, 2, 3].fetch_values(..3) { true } #=> IndexError
[1, 2, 3].fetch_values(nil..nil)  #=> IndexError

[1, 2, 3].fetch(-4) #=> IndexError
[1, 2, 3].fetch(-3) #=> 1
[1, 2, 3].fetch(2)  #=> 3
[1, 2, 3].fetch(3)  #=> IndexError

[#21015] Add in a -g flag, like -s but with a few more quality of life features

s = "str"
Marshal.old_dump([s, 1.0, 1.0, s]) #=> ["str", 1.0, 1.0, "otherstr", object-link 4]
"\x04\b[\tI\"\bstr\x06:\x06ETf\x061f\x061@\x06"

Marshal.new_load(s) #=> 0:[], 1:"str", 2:1.0, 3:1.0, 4:"otherstr", object-link 4

[#19555] Allow passing default options to Data.define

[#21097] x = a rescue b in c and def f = a rescue b in c parsed differently between parse.y and prism

[#20682] Slave PTY output is lost after a child process exits in macOS

[#21033] Allow lambdas that don’t access self to be Ractor shareable

x = 1
1.instance_eval("binding").local_variables #=> [:x]
Kernel.binding.eval("self")
1.send(:eval, "binding")

# other explicit method?
Proc.new{ ... }.isolate

matz: I prefer the idea but it may have escape hatches

[#21028] Method for finding why an object isn’t Ractor shareable

[#20971] Deprecate rb_path_check

[#20996] Embed Ruby 3.4 Failure

[#21111] RbConfig::CONFIG[‘CXX’] quietly set to “false” when Ruby cannot build C++ programs

katei: I’ll check other languages.

[#21104] Net::HTTP connections failing in Ruby >= 3.4.0 on macOS with Happy Eyeballs enabled

good luck shioi-san

[#21119] Programs containing Dir.glob with a thread executing a CPU-heavy task run very slowly.

[#20919] IO#seek and IO#pos= do not clear the character buffer in some cases while transcoding

ask akr, nobu to review the PR.

[#21121] Ractor channels

ko1: I also want it.

[#21100] DevMeeting before or after RubyKaigi2025