[7nohe]

Problem

The sqlite-smoke job (Bun 1.4.2) died twice on main in one day with Segmentation fault at address 0x70 and test:bun exiting 139, and the build-and-test (1.4.2) lane earlier ended with 0 fail, 1 error (Unhandled error between tests, a readonly-property TypeError inside aws-cdk-lib). Bun 1.3.14 never hit either.

Both are one mechanism. JSC queues a FinalizationRegistry cleanup as a DeferredWorkTimer ticket and Bun 1.4.0-1.4.2 run it with no realm check (Bun::runPendingWork), so it lands after the test file that created the registry is retired under --isolate:

• in a retired realm the cleanup surfaces as Unhandled error between tests (fixed upstream in bun test --isolate: drop a retired file's deferred work oven-sh/bun#42110, not in any release yet);
• once the retired global has been collected it reads a freed registry, takeDeadHoldingsValue on a vector whose buffer is null, the segfault at 0x70 (bun.report stack from the crash; Let JSC's DeferredWorkTimer own the deferred-work tickets so a collection cancels a dead realm's work (WebKit bump) oven-sh/bun#39994 with DeferredWorkTimer: keep the embedder's tickets in m_pendingTickets so the end of a collection cancels them oven-sh/WebKit#487, still open).

Any registry does it. One crash came right after the plugin-lambda CDK test, where aws-cdk-lib's bundled @aws/cloudformation-validate (wasm-bindgen) creates four registries per synth; the other came while loading a CLI test file, where the registry is one of Bun's own builtins (fs.promises, util, diagnostics_channel, v8, worker_threads each hold one).

Fix

A third test:bun preload, scripts/test-finalization-registry-guard.ts, replaces the global FinalizationRegistry with one that registers nothing on Bun 1.4.x, so no cleanup is ever queued. --preload re-runs in each isolated file's fresh global before its imports, so aws-cdk-lib's wasm bindings and Bun's builtins both capture the inert one (verified through the runner with a probe test). A registered cleanup only ever released memory the test process gives back at exit. Bun 1.3.14 is left untouched; the header says when to delete the file.

--isolate itself stays: it is what keeps a mock.module() swap in one file from leaking into the next (#114).

Verification

• bun test scripts/test-finalization-registry-guard.test.ts on Bun 1.3.14 and 1.4.2
• full bun run test:bun on Bun 1.4.2 with the guard: 9973 pass, 157 skip, 0 fail, no segfault, no unhandled error
• oxlint on the three touched files

Fixes #969