fix(sys): mimalloc3 v3.5.1 + fix for the process-init thread's thread-done hook

[Brooooooklyn]

Background

mimalloc v3.5.0 (shipped in libmimalloc-sys2 0.1.61 / mimalloc-safe 0.1.65 by #92) never associates the process-init thread's theap with the pthread thread-done key. If the first thread that touches the allocator exits — a Node.js worker_threads worker that loads the addon first, then exits — _mi_thread_done never runs for it, its theap stays registered with its thread id, and the next thread that reuses that pthread TCB aborts in debug builds:

mimalloc: assertion failed: at "src/init.c":327, _mi_thread_init_with_heap
  assertion: "theap==NULL"

Release builds skip the assert but leak the dead thread's theap and pages. Linux only in practice (macOS enables MI_TLS_RECURSE_GUARD, which makes the new thread adopt the stale theap). Upstream culprit: microsoft/mimalloc@5d9cb381 ("remove static main theap and tld"); v3.4.5 was the last good tag. Reported upstream with the fix: microsoft/mimalloc#1393. Found via rolldown/rolldown#10268 CI (watch debounce timers survive a worker-first registrant that exited), which pinned mimalloc-safe = "=0.1.64" as a stopgap.

Changes

• .gitmodules — mimalloc3 url microsoft/mimalloc → napi-rs/mimalloc, branch fix/first-thread-done (same approach as fix(sys): use THREAD_LOCAL TLS for mimalloc v3 on Apple #71).
• mimalloc3 — 18b08671 (v3.5.0) → 1e5d14ca = upstream v3.5.1 + the 11-line src/init.c fix (associate the already-initialized default theap right after _mi_prim_thread_init_auto_done()). v3.5.1 additionally carries the upstream fix for realloc() on new thread incorrectly asserts on 3.5.0 on Linux microsoft/mimalloc#1371 (realloc assert on a new thread).
• No change to build.rs, the v2 submodule, or the Rust sources.

Verification (Docker linux/arm64, node 22, debug + release builds of example/ with --features v3)

tree	Node worker-first flow (1 / 64 / 64+4 concurrent workers)	Rust std threads reusing the dead TCB	release	cargo test --features v3	sys-test
v3.5.0 / v3.5.1 unfixed	assertion + SIGABRT	assertion + SIGABRT	OK (leaks)	—	—
1e5d14ca	OK	OK	OK	6/6	251/251

Bisect over v3.4.5..v3.5.0 (7 steps) isolates 5d9cb381. The v2 line (c_src/mimalloc v2.5.0) does not have the gap and passes the same flow.

Once released, rolldown can drop its =0.1.64 pin.

🤖 Generated with Claude Code

https://claude.ai/code/session_01T34fMrYCf2V13Mg8BKmnda

---

Note

High Risk
Touches process-wide allocator and per-thread teardown in native code; wrong behavior can cause crashes or leaks in multi-threaded hosts (e.g. Node worker_threads), though scope is limited to the v3 submodule.

Overview
Repoints the mimalloc3 git submodule from microsoft/mimalloc (dev3) to napi-rs/mimalloc on branch fix/first-thread-done, matching the prior fork-based workflow used for targeted allocator fixes.

That branch carries mimalloc v3.5.1 plus a small src/init.c change so the process-init thread’s default heap is registered with the pthread thread-done hook. Without it, a worker thread that is the first allocator user and then exits can leave a stale heap on reused pthread TCBs (debug abort in _mi_thread_init_with_heap, leaks in release on Linux).

No Rust build.rs or crate source changes in this diff; consumers get the fix via the updated vendored mimalloc3 tree.

^{Reviewed by Cursor Bugbot for commit dd53da2. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_fceb089e-55a8-4416-8113-a18f533d39a9)

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-04T15:15:48.078684Z	dd53da2	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.