Bug #22261: ruby-head builds have been failing on aarch64-linux CI in ruby/ruby-dev-builder

• Status: Closed
• Assignee: -
• Target version: -
• Backport: 3.3: UNKNOWN, 3.4: UNKNOWN, 4.0: UNKNOWN

[tikkss (Tsutomu Katsube)]

The daily "CRuby Dev Builds" workflow in ruby/ruby-dev-builder has failed every day since 2026-08-07:

• https://github.com/ruby/ruby-dev-builder/actions/workflows/build.yml

Because of this, ruby-version: head currently installs a snapshot from 2026-08-05:

• https://github.com/ruby/ruby-dev-builder/releases

Only the build (ubuntu-26.04-arm, head) job fails, during make test-all TESTS="-j4". All other platforms build and test successfully, but since one matrix job fails, the "Publish Release" step is skipped and no new release is published.

This looks like a ruby master issue on aarch64-linux rather than a issue in ruby-dev-builder. For what it is worth, it does not seem to reproduce on the rubyci.org Ubuntu 26.04 aarch64 machines, which have been green for the same period:

• https://logs.rubyci.org/ubuntu-arm/ruby-master/recent.html

The most frequent failure (2026-08-13, 08-16, 08-22, 08-23, 08-24) is a crash in TestEnumerator#test_lazy_chain_under_gc_compact_stress:

test/ruby/test_enumerator.rb:869: [BUG] try to mark T_NONE object (obj: out-of-heap:0x0000e8d1b901d640, parent: 0x0000e8d1b937ff80 T_IMEMO/<fields>
(embed) len:2 capa:20x0000e8d1d3eb0040 proc_entry/Object )
ruby 4.1.0dev (2026-08-24T18:44:40Z :detached: c4fdf07d44) +PRISM [aarch64-linux]

Full crash dumps are in the logs:

• https://github.com/ruby/ruby-dev-builder/actions/runs/32767004006/job/97558786838 (2026-08-24, c4fdf07d44)
  • job-logsc4fdf07d44.txt
• https://github.com/ruby/ruby-dev-builder/actions/runs/32592837360/job/97079257443 (2026-08-22, 774cc95fd4)
  • job-logs774cc95fd4.txt

On other days (2026-08-07, 08-08, 08-21), TestObjectSpace#test_finalizer_with_super failed instead: the child process exits successfully but prints garbage bytes (e.g. "\xC5w<") instead of :ok:

• https://github.com/ruby/ruby-dev-builder/actions/runs/31211225131/job/92974386628 (2026-08-07, 3156d5adb4)
  • job-logs3156d5adb4.txt

[nobu (Nobuyoshi Nakada)]

Does GH-18495 fix it?

[Eregon (Benoit Daloze)]

nobu (Nobuyoshi Nakada) wrote in #note-1:

Does GH-18495 fix it?

Thank you for creating a fix.
The easiest way to know would be to merge that PR, and ruby-dev-builder runs once per day.
An alternative would be to add a CI job running make test-all on ubuntu-26.04-arm in ruby/ruby CI, that would be even better and prevent future regressions on that platform.

[nobu (Nobuyoshi Nakada)]

• Status changed from Open to Closed

Applied in changeset git|e77a77111224426ca18eaeff55d7ad1618cbacc9.

---

[Bug #22261] Keep GC root-marking frame alive

Prevent a tail call from discarding saved machine-register roots
before rb_gc_mark_roots scans the machine stack.

[tikkss (Tsutomu Katsube)]

The daily "CRuby Dev Builds" workflow in ruby/ruby-dev-builder has been successful. Thanks for your quick fix!