agra 4ddee931b5 ffi 1.29: retire the C sx_android_query_safe_insets body ...

Closes the Phase 1D migration for the safe-insets JNI chain. The C
function and its `#foreign` declaration in `android.sx` are gone;
all dispatch now goes through the sx-side `#jni_call` machinery
plus the JavaVM helpers landed in 1.26.

What's gone from `library/vendors/sx_android_jni/sx_android_jni.c`:
- `#include <android/native_activity.h>` and `<jni.h>` (no longer
  needed without the JNI body).
- `sx_android_query_safe_insets` — 55 lines of `(*env)->Foo` chain
  with manual `goto done` early-exit. Migrated to
  `library/modules/platform/android.sx::sx_query_safe_insets_jni`
  in 1.25 (15 lines of `#jni_call`).

What stays:
- `sx_android_install_input_handler` — non-JNI; struct-field
  assignment against `struct android_app`'s `onInputEvent` slot.
  No sx equivalent yet (would need to either land a `#android_app`-
  style intrinsic or hand-roll the offset, neither of which is
  Phase 1 scope).
- `<android/input.h>` and the `struct sx_android_app_min` mirror
  needed by the input-handler installer.

Net diff: -55 lines in the .c file, -1 line `#foreign` decl in
android.sx. Phase 2 (declarative JNI imports) will revisit whether
the .c file can be deleted entirely (the input-handler hop may
move into a different shape).

Verification:
- zig build + zig test + run_examples + cross_compile all green.
  Notable: the previously-failing `ffi-objc-call-12-rect-u64-returns`
  also passes now — looks like the working-tree `#import c` work
  was tidied up alongside.
- chess Android APK rebuilt + reinstalled + launched on Pixel
  device; safe-insets behavior unchanged (board top edge sits below
  the status bar correctly, all pieces in starting positions, no
  status-bar overlap).

2026-05-19 23:13:51 +03:00

25 KiB

Raw Blame History

View Raw