Bug #20264: Segfault in Enumerator#next while installing RMagick on M1 Mac

• Status: Feedback
• Assignee: -
• Target version: -
• ruby -v: 3.3
• Backport: 3.0: UNKNOWN, 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN

[[e-mail] (Andy Jeffries)]

I raised a bug at RMagick, they said to go to pkg-config. I raised a bug there, and they said to raise it here. Hopefully this is the right place...

https://github.com/ruby-gnome/pkg-config/issues/24

Simplest reproduction steps (although a full backtrace of a Rails app trying to do it is at the link above):

$ docker run -it --platform=linux/arm64 registry.docker.com/library/ruby:3.3-slim bash
root@163711a82570:/# apt-get update
root@163711a82570:/# apt-get install --no-install-recommends -y build-essential pkg-config libmagickwand-dev
root@163711a82570:/# gem install rmagick

[byroot (Jean Boussier)]

The interesting parts from the upstream ticket:

Segmentation fault at 0x0067ffffb37006c0
ruby 3.3.0 (2023-12-25 revision 5124f9ac75) [aarch64-linux]

-- Control frame information -----------------------------------------------
c:0012 p:---- s:0068 E:002060 CFUNC  :next
c:0011 p:0005 s:0064 e:000063 BLOCK
/usr/local/bundle/ruby/3.3.0/gems/pkg-config-1.5.6/lib/pkg-config.rb:439
c:0010 p:0018 s:0060 e:000059 METHOD <internal:kernel>:187
c:0009 p:0014 s:0055 e:000054 METHOD
/usr/local/bundle/ruby/3.3.0/gems/pkg-config-1.5.6/lib/pkg-config.rb:438
c:0008 p:0035 s:0048 e:000047 METHOD
/usr/local/bundle/ruby/3.3.0/gems/pkg-config-1.5.6/lib/pkg-config.rb:403
c:0007 p:0003 s:0039 e:000038 METHOD
/usr/local/bundle/ruby/3.3.0/gems/pkg-config-1.5.6/lib/pkg-config.rb:297
c:0006 p:0007 s:0033 e:000032 METHOD
/usr/local/bundle/ruby/3.3.0/gems/pkg-config-1.5.6/lib/pkg-config.rb:603
c:0005 p:0108 s:0028 e:000025 METHOD extconf.rb:111
c:0004 p:0018 s:0016 e:000015 METHOD extconf.rb:53 [FINISH]
c:0003 p:---- s:0012 e:000011 CFUNC  :new
c:0002 p:0067 s:0008 E:000f60 EVAL   extconf.rb:449 [FINISH]
c:0001 p:0000 s:0003 E:000520 DUMMY  [FINISH]

Unfortunately, no C-level backtrace included.

So it's crashing in Enumerator#next on this line https://github.com/ruby-gnome/pkg-config/blob/eff47c9f57f8411b5e3c535cae0d8e4c9a368c27/lib/pkg-config.rb#L439

  def normalize_cflags(cflags)
    normalized_cflags = []
    enumerator = cflags.to_enum
    begin
      loop do
        cflag = enumerator.next # CRASH!
        normalized_cflags << cflag
        case cflag
        when "-I"
          normalized_cflags << enumerator.next
        end
      end
    rescue StopIteration
    end
    normalized_cflags
  end

I haven't tried to repro yet.

[byroot (Jean Boussier)]

• Subject changed from Segfault installing RMagick on M1 Mac to Segfault in Enumerator#next while installing RMagick on M1 Mac

[byroot (Jean Boussier)]

Coincidentally, I'm pretty sure that code is useless: https://github.com/ruby-gnome/pkg-config/pull/25.

[sorah (Sorah Fukumori)]

seems like

https://github.com/ruby/ruby/pull/9385
https://bugs.ruby-lang.org/issues/20085

[hsbt (Hiroshi SHIBATA)]

• Related to Bug #20085: Fiber.new{ }.resume causes Segmentation fault for Ruby 3.3.0 on aarch64-linux added

[hsbt (Hiroshi SHIBATA)]

I could install rmagick with linux/amd64.

$ docker run -it --platform=linux/amd64 registry.docker.com/library/ruby:3.3-slim bash

This may be fixed with #20085 with @sorah (Sorah Fukumori) 's suggestion.

[[e-mail] (Andy Jeffries)]

hsbt (Hiroshi SHIBATA) wrote in #note-6:

I could install rmagick with linux/amd64.

$ docker run -it --platform=linux/amd64 registry.docker.com/library/ruby:3.3-slim bash

This may be fixed with #20085 with @sorah (Sorah Fukumori) 's suggestion.

This is definitely an arm64 issue, our production containers use amd64 and I agree it works fine on that platform.

[sorah (Sorah Fukumori)]

• Related to deleted (Bug #20085: Fiber.new{ }.resume causes Segmentation fault for Ruby 3.3.0 on aarch64-linux)

[sorah (Sorah Fukumori)]

• Is duplicate of Bug #20085: Fiber.new{ }.resume causes Segmentation fault for Ruby 3.3.0 on aarch64-linux added

[sorah (Sorah Fukumori)]

• Status changed from Open to Feedback

there's no workaround right now (AFAIK affected platform is Linux VM on Apple Silicon), need to wait 3.3.1 to be shipped (any estimates?)