[kleisauke]

No description provided.

[kleisauke]

A trivial patch update. All existing patches are reapplied as-is (see e.g. commit 724728a were these patches were removed and commit affa429 were they were reapplied). However, some of these changes have since been upstreamed (or backported):

• Fix -Wattribute-alias warning with malloc_usable_size() override (microsoft/mimalloc@3ce047e).
• Prefer Emscripten's prim.c implementation over WASI (microsoft/mimalloc@2d8aa7e, backported via b6f35aa).
• Avoid an extra newline in mimalloc's stderr output (microsoft/mimalloc@21a9e8e, backported via 3877c73).

I also had to revert the mi_prim_heap_has_room() logic introduced in commit microsoft/mimalloc@0fec3da, as we should forward users to build with -sABORTING_MALLOC=0 instead (which is already done in the test_wrap_malloc_mimalloc test and when building with -sALLOW_MEMORY_GROWTH). As possible follow-up, we could make -sABORTING_MALLOC=0 the default when building with -sMALLOC=mimalloc.

Note that upstream now always uses a 2-level page map (see commit microsoft/mimalloc@cbd6dce), which seems to work OK with our default INITIAL_HEAP of 16 MiB. So, perhaps my analysis in #27982 (comment) was incorrect after all.

Tested with:

$ ./test/runner {core2,wasmfs,wasm64_4gb}.test_wrap_malloc_mimalloc other.test_malloc_size_mimalloc* other.test_malloc_multithreading_mimalloc* other.test_mimalloc_options
Running 8 tests
Using 8 parallel test processes
[8/8] test_malloc_size_mimalloc_O2 (test_other.other.test_malloc_size_mimalloc_O2) ... ok

----------------------------------------------------------------------
Ran 8 tests in 1.142s

OK
Total core time: 3.706s. Wallclock time: 1.142s. Parallelization: 3.25x.

[gemini-code-assist]

Code Review

This pull request updates mimalloc to version 3.6.1, introducing several key architectural, security, and compatibility enhancements. Notably, it consolidates almost all static and global variables into a single, guarded mi_process_t structure (_mi_process) to protect allocator metadata. It also transitions to always using a 2-level page map for simplicity and security, adds support for C23 free_sized and free_aligned_sized overrides, and introduces experimental support for CHERI platforms. Additionally, it improves fork support (particularly on macOS), optimizes the STL allocator trait, and enhances WebAssembly/Emscripten/WASI compatibility. There are no review comments to address, and I have no further feedback to provide.

[sbc100]

Looks like system/lib/mimalloc/README.emscripten is missing from this change?

Otherwise lgtm

[kleisauke]

Looks like system/lib/mimalloc/README.emscripten is missing from this change?

Ah, my bad. Forgot to update that file, fixed with commit 1b8015b.