Test with android NDK r27

[dg0yt]

New LTS Version

• offering support for new arm64 page size
  https://android.googlesource.com/platform/ndk/+/master/docs/BuildSystemMaintainers.md#Page-sizes
  Test with android NDK r27 #40339 (comment)

Problems:

• breaking compatibility with CMake projects which don't ask for CMake 3.3 or set CMP0057 to NEW
  [FR] push/pop required policies in cmake toolchain file android/ndk#2032
• gdal:arm-neon-android: clang++ crash
• ncnn:x64-android: clang++ crash, [BUG] Build error of ncnn for Android 'x86_64' ABI with NDK r27  android/ndk#2054
• several ports: 'ALooper_pollAll' is unavailable: obsoleted in Android 1 - ALooper_pollAll may ignore wakes. Use ALooper_pollOnce instead.
• some ports with compiler flags issues (minor probably)

[Neumann-A]

breaking compatibility with CMake projects which don't ask for CMake 3.5.

Why? Just set the required policy in the toolchain?

[dg0yt]

Why?

Sounds aggressive. Ask Google.

Just set the required policy in the toolchain?

My plan is to do just that, in scripts/toolchains/android.cmake.

[bwrsandman]

CMake projects which don't ask for CMake 3.5.

Looked into the versions. It looks like 3.3 (not 3.5) is the minimum necessary

[dg0yt]

https://github.com/android/ndk/wiki/Changelog-r27:

Android V will allow OEMs to ship arm64-v8a devices with 16KiB page sizes. Devices that use this configuration will not be able to run existing apps that use native code. To be compatible with these devices, applications will need to rebuild all their native code to be 16KiB aligned, and rewrite any code which assumes a specific page size.

Also https://android.googlesource.com/platform/ndk/+/master/docs/BuildSystemMaintainers.md#Page-sizes

Q: Shall 16KiB be the default for NDK r27 arm64-android?
Q: Shall there be triplet variable?

OTOH it isn't more urgent than raising the API level.

[Neumann-A]

Why?

Sounds aggressive. Ask Google.

Reads like: Words, has no sound.... ¯\(ツ)/¯ Don't mind me asking stupid questions.
But you already expressed my opinion upstream before I even had one :) .

Q: Shall there be a special arm64-v8a triplet?

[dg0yt]

Initial success:

	Success	Fail	Cascade
arm-neon-android	1120	252	230
arm64-android	1139	255 (due to IN_LIST: 251)	257
x64-android	1135	263	278

[dg0yt]

I have difficulties to get CMP0057 from the toolchain scope (android.cmake, called via CMakeDetermineSystem.cmake) to the failing scope (flags.cmake, called via CMakeSystem.cmake). Even when I put it directly into vcpkg.cmake.

Testing the proposed fix to the NDK instead.

[dg0yt]

This works:

diff --git a/scripts/toolchains/android.cmake b/scripts/toolchains/android.cmake
index d675d06a73..4564f29e7e 100644
--- a/scripts/toolchains/android.cmake
+++ b/scripts/toolchains/android.cmake
@@ -31,6 +31,11 @@ endif()
 
 include("${ANDROID_NDK_HOME}/build/cmake/android.toolchain.cmake")
 
+if(ANDROID_NDK_MAJOR STREQUAL "27")
+    # https://github.com/android/ndk/issues/2032#issuecomment-2274923977
+    string(APPEND CMAKE_SYSTEM_CUSTOM_CODE "cmake_policy(SET CMP0057 NEW)\n")
+endif()
+
 if(NOT _VCPKG_ANDROID_TOOLCHAIN)
     set(_VCPKG_ANDROID_TOOLCHAIN 1)
 

But it might be a somewhat tight binding to both CMake and NDK script code.

[dg0yt]

Success with patch to NDK's flags.cmake:

	Success	Fail	Cascade
arm-neon-android	n/a*	6	20
arm64-android	n/a*	8 (due to IN_LIST: 0	28
x64-android	n/a*	9	77

*) successfuild builds from the first run were cached and not tested again