Bug #21607: require 'concurrent-ruby' causes segfault with Ruby 3.4.6 on linux/i686 built with GCC 15

• Status: Open
• Assignee: ractor
• Target version: -
• ruby -v: ruby 3.4.6 (2025-09-16 revision dbd83256b1) +PRISM [i686-linux]
• Backport: 3.2: UNKNOWN, 3.3: UNKNOWN, 3.4: REQUIRED

[satadru (Satadru Pramanik)]

As mentioned at https://github.com/ruby/ruby/pull/14532#issuecomment-3288983469

chronos@buppie-i686:2.23 M58 /usr/local/lib/crew/packages (master >|SPARSE)$ ruby -e "require 'concurrent-ruby' ; puts 'success'"
Segmentation fault

This causes a segfault on i686 only on our Chromebrew/i686 systems. armv7l and x86_64 do not have this issue.

This may be linked to the update to prism 1.5.x?

Docker reproducer on our i686 ChromeOS Milestone 58 based container image:

docker run --init --platform linux/386 --rm --net=host -v $(pwd):/output -h $(hostname)-i686 --tmpfs /tmp --ulimit "nofile=$(ulimit -Sn):$(ulimit -Hn)" -it satmandu/crewbuild:alex-i686.m58

Note the lack of an issue with 3.4.5 with this command:

ruby -e "require 'concurrent-ruby' ; puts 'success'"

In that container, upgrade ruby to 3.4.6:

CREW_REPO=https://github.com/chromebrew/chromebrew.git CREW_BRANCH=updater-ruby_3.4.6 crew update
&& yes | crew upgrade

Then run the breaking command:

ruby -e "require 'concurrent-ruby' ; puts 'success'"

Note the segfault.

[satadru (Satadru Pramanik)]

Our downstream PR/Issue is here: https://github.com/chromebrew/chromebrew/pull/12806

[satadru (Satadru Pramanik)]

• File i686-ruby-github-action-build-job-log.txt i686-ruby-github-action-build-job-log.txt added

Our build run for this build, including make check is here:

https://github.com/chromebrew/chromebrew/actions/runs/17767133141/job/50493282824

I have attached this log with extraneous information edited out.

[k0kubun (Takashi Kokubun)]

• Assignee set to prism
• Backport changed from 3.2: UNKNOWN, 3.3: UNKNOWN, 3.4: UNKNOWN to 3.2: UNKNOWN, 3.3: UNKNOWN, 3.4: REQUIRED

[luke-gru (Luke Gruber)]

• Assignee changed from prism to ractor

There appear to be issues with Ractors. I'm going to reassign this to the ractor team.

[kddnewton (Kevin Newton)]

Thanks @luke-gru (Luke Gruber) I was struggling to figure out what that had to do with prism at all, seeing as all the failures were ractor related.

[luke-gru (Luke Gruber)]

@kddnewton (Kevin Newton) I'm not sure it is related to ractors because maybe those tests always failed on this platform. I'm also seeing compile warnings like:

2025-09-16T13:34:45.1561568Z In function ‘iseq_set_local_table’,
2025-09-16T13:34:45.1562672Z     inlined from ‘rb_iseq_compile_node’ at compile.c:875:9:
2025-09-16T13:34:45.1586403Z compile.c:2185:25: warning: argument 1 value ‘4294967293’ exceeds maximum object size 2147483647 [-Walloc-size-larger-than=]
2025-09-16T13:34:45.1587570Z  2185 |         ID *ids = (ID *)ALLOC_N(ID, size);

No ractor-related changes have made it into the 3.4.6 release AFAIK.

[satadru (Satadru Pramanik)]

Hello all, a build from earlier this summer's 3.4 branch using the current bundled gems also gave the same error when I built that two days ago.

The only other difference might be that we are using GCC 15 now.

But I'd also note that this builds fine and appears to work without issue on 32-bit armv7l as well as on x86_64, and those builds use GCC 15 too.

[luke-gru (Luke Gruber)]

It would help us a lot if you could get the output from the crashing process. It should report a backtrace to stderr after the segfault right before the process ends.

[satadru (Satadru Pramanik)]

It doesn't appear to get to a backtrace:

chronos@buppie-i686:2.23 M58 /usr/local/lib/crew/packages (master <>|SPARSE)$ irb
irb(main):001> require 'concurrent-ruby'
Segmentation fault         (core dumped) irb
chronos@buppie-i686:2.23 M58 /usr/local/lib/crew/packages (master <>|SPARSE)$ ruby -e "require 'concurrent-ruby' ; puts 'success'"
Segmentation fault         (core dumped) ruby -e "require 'concurrent-ruby' ; puts 'success'"

[Earlopain (Earlopain _)]

It does not seem to be related to prism (or ruby at all?):

$ ruby -e "Fiber.new {}.resume"
Segmentation fault         (core dumped) ruby -e "Fiber.new {}.resume"
$ ruby --disable-gems -e "Fiber.new {}.resume"
Inconsistency detected by ld.so: dl-runtime.c: 79: _dl_fixup: Assertion `ELFW(R_TYPE)(reloc->r_info) == ELF_MACHINE_JMP_SLOT' failed!

There is no further output, that's all you get. Does look like something that might happen with a compiler upgrade.

[satadru (Satadru Pramanik)]

Hmm. Ok, I will try rebuilding with GCC 14, in case this is possibly a regression in GCC.

[satadru (Satadru Pramanik)]

I can confirm that building with GCC 14 resolves the issue. So this seems like a problem with building ruby for i686 with GCC 15.

[satadru (Satadru Pramanik)]

• Subject changed from require 'concurrent-ruby' causes segfault with Ruby 3.4.6 on linux/i686 to require 'concurrent-ruby' causes segfault with Ruby 3.4.6 on linux/i686 built with GCC 15