Skip to content

Android 3.15t: mimalloc libc branch on Android (x86_64 seccomp SIGSYS) - #437

Merged
b-re-w merged 2 commits into
developfrom
feat/ft-android-seccomp
Nov 15, 2026
Merged

b-re-w merged 2 commits into
developfrom
feat/ft-android-seccomp

Conversation

@b-re-w

@b-re-w b-re-w commented Nov 15, 2026

Copy link
Copy Markdown
Member

Closes #436

The free-threaded x86_64 Android build died with SIGSYS at Py_Initialize (toolchain#145 run 43691526214): mimalloc's raw syscall(SYS_access) in _mi_prim_numa_node_count is blocked by the app seccomp filter. build-android.sh now patches mimalloc's raw-syscall guard with && !defined(ANDROID) (fails loudly if the guard line changes). arm64 already used the libc branch.

Verification needs build-ft-mobile.yml for x86_64-linux-android (rebuilds and re-uploads the asset), then toolchain #145 android-e2e.

…not killed by seccomp

CPython's vendored mimalloc calls syscall(SYS_access) and syscall(SYS_open) directly whenever those
numbers exist. Android's x86_64 app seccomp filter allows only faccessat/openat, so the free-threaded
x86_64 build died with SIGSYS at Py_Initialize (toolchain#145 android-e2e, run 43691526214, traced to
_mi_prim_numa_node_count). The build script now adds && !defined(__ANDROID__) to that guard and fails
loudly if the guard line changes. arm64 has no __NR_access and already took the libc branch.

Closes #436
The cpython-3.15.0t-mobile x86_64 asset was rebuilt with the mimalloc seccomp fix (build-ft-mobile
run 43697003385; the new libpython3.15t.so has no raw access/open syscall sites) and replaced on the
release with the owner's approval (2026-11-15). The old asset is kept under .cache. Refs #436.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant