chore(deps): update mimalloc to v2.5.2 and mimalloc3 to v3.5.2 + init fix

[Brooooooklyn]

Background

Upstream closed microsoft/mimalloc#1393 and took the fix in a different form: microsoft/mimalloc@439dc27b, shipped in v3.5.2. That commit works on the default TLS model (verified with the #1393 repro on Linux, debug shared build: e35400bd aborts, v3.5.2 prints OK).

But it moved mi_process_setup_auto_thread_done() into mi_process_init_once() before _mi_process_is_initialized = true. With MI_TLS_RECURSE_GUARD (auto-on for MI_TLS_MODEL_LOCAL on Apple, which is what build.rs forces for v3 on macOS) _mi_theap_default() returns _mi_theap_empty at that point:

mi_process_init_once()                          [v3.5.2 src/init.c:558-564]
  ├─ mi_thread_init()                           __mi_theap_default = &mi_process_theap_main
  ├─ _mi_tls_slots_init()
  ├─ _mi_thread_locals_init()
  ├─ mi_process_setup_auto_thread_done()
  │    └─ _mi_theap_default()  ──► RECURSE_GUARD: _mi_process_is_initialized == false
  │                                ──► returns &_mi_theap_empty
  │                                ──► debug: assert fires / release: no association
  └─ _mi_process_is_initialized = true          ← too late

So on plain v3.5.2, cargo test --features v3 on macOS aborts at process init in every debug binary:

mimalloc: assertion failed: at "src/init.c":450, mi_process_setup_auto_thread_done
  assertion: "mi_theap_is_initialized(theap)"

and release builds silently skip the association, i.e. the #1393 leak is back on macOS. Reproducible with upstream alone: cmake -DCMAKE_BUILD_TYPE=Debug -DMI_TLS_MODEL=LOCAL on macOS, ./mimalloc-test-api passes on v3.5.1 and aborts on v3.5.2. Same on Linux with -DMI_TLS_RECURSE_GUARD=ON.

Changes

• mimalloc (v2) — v2.5.1 → v2.5.2 (official tag, no patches).
• mimalloc3 — napi-rs fork 1e5d14ca (v3.5.1 + #1393) → napi-rs fork afa6c90d = upstream v3.5.2 + one-line fix: set _mi_process_is_initialized = true before mi_process_setup_auto_thread_done() (napi-rs/mimalloc@afa6c90d, branch fix/process-init-flag-before-thread-done). Same approach as fix(sys): use THREAD_LOCAL TLS for mimalloc v3 on Apple #71 / fix(sys): mimalloc3 v3.5.1 + fix for the process-init thread's thread-done hook #97.
• .gitmodules — mimalloc3 branch name updated.
• No change to build.rs or the Rust sources.

Verification

Upstream, macOS arm64, Debug, -DMI_TLS_MODEL=LOCAL:

tree	mimalloc-test-api
v3.5.1	50/50
v3.5.2	abort at process init (134)
v3.5.2 + afa6c90d	50/50

Upstream, Linux x64 (Docker gcc:14), Debug shared, #1393 repro:

tree	RECURSE_GUARD=OFF	RECURSE_GUARD=ON
v3.5.2	OK	abort (134)
v3.5.2 + afa6c90d	OK	OK

This repo, macOS arm64, every CI command: cargo test / --features secure / extended / v3 / extended,v3 all pass; libmimalloc-sys-test 258/258 (v2, secure, extended) and 251/251 (v3).

Pre-existing and untouched: upstream mimalloc-test-stress-heaps fails the refcount == 1 assert with MI_TLS_MODEL=LOCAL on macOS on v3.5.1 too; extended::tests::runtime_stable_option fails under cargo test --workspace on v2.5.1 too (not in CI).

🤖 Generated with Claude Code

https://claude.ai/code/session_016srh5svTkqfW21zGcU6ht9

---

Note

Medium Risk
Changes which mimalloc v3 C sources are built, affecting process/thread allocator initialization on macOS and other TLS recurse-guard configurations.

Overview
Points the mimalloc3 git submodule at the napi-rs fork branch fix/process-init-flag-before-thread-done instead of fix/first-thread-done, so v3 tracks upstream v3.5.2 plus a one-line init ordering fix: set _mi_process_is_initialized before mi_process_setup_auto_thread_done().

That avoids macOS MI_TLS_MODEL_LOCAL / recurse-guard failures where v3.5.2 alone aborts in debug or reintroduces the #1393 leak in release. No Rust or build.rs changes in this diff—only the submodule branch pin in .gitmodules.

^{Reviewed by Cursor Bugbot for commit 09fc0c6. 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_3d3890ae-67c0-4f8c-9c68-c873c1ecfb6e)

[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 2026-09-14T02:45:42.032641Z	09fc0c6	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.