[SDK Sentinel] Fix custom slot supplier GC teardown crash

[sdk-sentinel-bot]

Caution

This PR contains untrusted AI-generated code. Do not approve or run CI until a maintainer has reviewed the diff. SDK Sentinel verified that GitHub Actions remained approval-gated with zero executable jobs when this PR was opened.

Summary

• CI evidence: failed CI job · workflow run · last known pass · first known failure

Prevent a Ruby 4.0 FIPS worker-suite process crash seen in the [first captured target-branch failure](https://github.com/temporalio/sdk-ruby/actions/runs/33914370027), after [earlier equivalent successful CI](https://github.com/temporalio/sdk-ruby/actions/runs/33911088056). The following target-branch CI also passed without relevant bridge changes.

Root cause

Custom slot permits retain Ruby objects in BoxValue. Core can release the final permit reference while completing a local activity on a Tokio thread. Its destructor then calls Ruby's GC unregister API outside a Ruby thread, matching the native crash stack and an established repository failure mode.

Fix

Defer foreign-thread BoxValue destruction in a process-wide queue. The next Ruby-thread creation drains the queue, ensuring GC unregister operations run in a valid Ruby context and removing the thread-scheduling nondeterminism.

Validation

The unchanged baseline and candidate canonical validator passed. Ten standard and five FIPS focused stress attempts passed. The rare crash did not reproduce locally; final deferred values may remain retained until another boxed value is created or process exit.

Validation status: validated-without-reproduction

• Flake confidence: high
• Fix confidence: high
• Value: medium
• Patch scope: standard — 1 changed file; 2.9 KiB
• Local reproduction: not-reproduced — Six pre-change focused runs passed without reproducing the rare VM crash; one run showed a non-fatal worker-finalization timing warning.
• Regression coverage: pass-after-only — Existing custom slot supplier coverage passed in 10 standard and five FIPS post-change stress attempts.
• Unchanged baseline (RuboCop, Steep, Rust lint, and native compile) — passed
• Independent candidate (RuboCop, Steep, Rust lint, and native compile) — passed

Investigator-run checks

• bundle exec rake test TESTOPTS="--name=/custom_slot_supplier/" — passed; 6 attempt(s). Pre-change attempts passed; the rare crash was not reproduced naturally.
• for attempt in 1 2 3 4 5 6 7 8 9 10; do bundle exec rake test TESTOPTS="--name=/custom_slot_supplier/" \|\| exit 1; done — passed; 10 attempt(s). All 60 post-change standard-mode focused cases passed.
• for attempt in 1 2 3 4 5; do CARGO_NET_OFFLINE=true TEMPORALIO_FIPS=1 bundle exec rake test TESTOPTS="--name=/custom_slot_supplier/" \|\| exit 1; done — passed; 5 attempt(s). All 30 Ruby 4.0 FIPS-mode focused cases passed.
• .ci-flake-runtime/input/validate.sh — passed; 1 attempt(s). The trusted unchanged baseline passed RuboCop, Steep, Rust lint/format, and native compilation.
• .ci-flake-runtime/input/validate.sh — passed; 1 attempt(s). The candidate passed the same canonical validator.
• TEMPORALIO_FIPS=1 bundle exec rake compile — passed; 1 attempt(s). The exact FIPS native bridge configuration compiled successfully.

Required target CI

• Continuous Integration / build-lint-test (ubuntu-latest, 4.0, true)
• Continuous Integration / build-lint-test matrix for Ubuntu and macOS with Ruby 3.3 and 4.0

Residual risks

• Natural crash reproduction was not achieved, so validation relies on the native causal stack and stress coverage.
• Values deferred after the final Ruby operation remain retained until another SendSyncBoxValue is created or the process exits.

Automation provenance

• Investigator workflow
• Base SHA: d2dee2eebc252a1594f4c17d047be900e357d3b2

If this finding should not be fixed, apply the sdk-sentinel:false-positive label and close the PR. Sentinel will suppress the finding until its affected source changes.

SDK Sentinel recurrence history

• Unique failed CI run attempts: 1
• First occurrence: 2026-09-04 20:18 UTC
• Latest occurrence: 2026-09-04 20:18 UTC
• Recent occurrences:
  • 2026-09-04 20:18 UTC — failed CI job