[question] How to robustly access tools like strip from cmake

[baldurk]

I use the android cmake toolchain file to build my project, and previously I was using CMAKE_STRIP to strip libraries in release builds. I recently built a new system and as part of that updated to the latest ndk, and found that CMAKE_STRIP is no longer exported by the android toolchain file.

This seems fine but I'm not 100% sure I have the right way of getting the proper way of getting the path to the strip utility. This is what I've updated to do:

    set(ANDROID_STRIP_TOOL "${ANDROID_TOOLCHAIN_ROOT}/bin/llvm-strip${ANDROID_TOOLCHAIN_SUFFIX}")
    if(NOT EXISTS "${ANDROID_STRIP_TOOL}")
        set(ANDROID_STRIP_TOOL "${CMAKE_STRIP}")
    endif()

    message(STATUS "libraries will be stripped with ${ANDROID_STRIP_TOOL}")

Which should work on windows/linux hosts for the new NDK, and fall back to CMAKE_STRIP on older NDKs (I support back to 14b for building).

Mainly I'm wondering if there's a better intended way to get the android strip tool, in case that hardcoded path changes in future, or if llvm-strip doesn't work for all ABIs (I'm unclear if I need a per-ABI strip tool, it seems to work for arm 32-bit and 64-bit).

NDK version: 23.0.7196353
ABI: armeabi-v7a & arm64-v8a
Build system: cmake (I'm using 3.20 but I support back to 2.8.12 in principle)

[hhb]

You can use CMAKE_STRIP directly. It will be set to the correct strip tool here.

[baldurk]

That is only true if you enable the legacy toolchain file, which I assume won't be available forever. By default that file is not invoked so CMAKE_STRIP is not set, otherwise I would never have noticed this problem in the first place.

[hhb]

For CMake >= 3.20, CMake should detect and set CMAKE_STRIP.

If this is really a problem we should revert this: https://android.googlesource.com/platform/ndk/+/e854ac5e0060b7c0061a06c1d32e39fcce9fbad2

[baldurk]

That doesn't happen for me with cmake 3.20. I made this little repro case (test.c is empty):

cmake_minimum_required(VERSION 2.8.12)

set(CMAKE_TOOLCHAIN_FILE "$ENV{ANDROID_NDK}/build/cmake/android.toolchain.cmake")
set(ANDROID_PLATFORM "android-21")
set(CMAKE_ANDROID_NDK_TOOLCHAIN_VERSION "clang")
set(ANDROID_ABI "armeabi-v7a")

project(test CXX C)
add_library(test SHARED test.c)

message(STATUS "Stripping with ${CMAKE_STRIP}")
add_custom_command(TARGET test POST_BUILD COMMAND ${CMAKE_STRIP} --strip-unneeded $<TARGET_FILE:test>)

In an msys shell it uses the host strip:

$ cmake -G "MSYS Makefiles" ..
-- Android: Targeting API '21' with architecture 'arm', ABI 'armeabi-v7a', and processor 'armv7-a'
-- Android: Selected unified Clang toolchain
-- The CXX compiler identification is Clang 11.0.5
-- The C compiler identification is Clang 11.0.5
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: T:/android/studiosdk/ndk/23.0.7196353/toolchains/llvm/prebuilt/windows-x86_64/bin/clang++.exe - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: T:/android/studiosdk/ndk/23.0.7196353/toolchains/llvm/prebuilt/windows-x86_64/bin/clang.exe - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Stripping with C:/msys/usr/bin/strip.exe
-- Configuring done
-- Generating done
-- Build files have been written to: T:/tmp/droid_cmake_test/build
$ cmake --build .
[ 50%] Building C object CMakeFiles/test.dir/test.c.o
[100%] Linking C shared library libtest.so
/C/msys/usr/bin/strip: Unable to recognise the format of the input file `T:/tmp/droid_cmake_test/build/libtest.so'
make[2]: *** [CMakeFiles/test.dir/build.make:97: libtest.so] Error 1
make[2]: *** Deleting file 'libtest.so'
make[1]: *** [CMakeFiles/Makefile2:83: CMakeFiles/test.dir/all] Error 2
make: *** [Makefile:91: all] Error 2
$ cmake --version
cmake version 3.20.0

CMake suite maintained and supported by Kitware (kitware.com/cmake).

And in regular cmd it can't find strip at all:

>cmake -G Ninja ..
-- Android: Targeting API '21' with architecture 'arm', ABI 'armeabi-v7a', and processor 'armv7-a'
-- Android: Selected unified Clang toolchain
-- The CXX compiler identification is Clang 11.0.5
-- The C compiler identification is Clang 11.0.5
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: T:/android/studiosdk/ndk/23.0.7196353/toolchains/llvm/prebuilt/windows-x86_64/bin/clang++.exe - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: T:/android/studiosdk/ndk/23.0.7196353/toolchains/llvm/prebuilt/windows-x86_64/bin/clang.exe - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Stripping with CMAKE_STRIP-NOTFOUND
-- Configuring done
-- Generating done
-- Build files have been written to: T:/tmp/droid_cmake_test/build
>cmake --build .
[2/2] Linking C shared library libtest.so
FAILED: libtest.so
cmd.exe /C "cd . && T:\android\studiosdk\ndk\23.0.7196353\toolchains\llvm\prebuilt\windows-x86_64\bin\clang.exe --target=armv7-none-linux-androideabi21 --sysroot=T:/android/studiosdk/ndk/23.0.7196353/toolchains/llvm/prebuilt/windows-x86_64/sysroot -fPIC -DANDROID -fdata-sections -ffunction-sections -funwind-tables -fstack-protector-strong -no-canonical-prefixes -D_FORTIFY_SOURCE=2 -march=armv7-a -mthumb -Wformat -Werror=format-security -fexceptions -O2 -g -DNDEBUG  -Wl,--exclude-libs,libgcc.a -Wl,--exclude-libs,libgcc_real.a -Wl,--exclude-libs,libatomic.a -Wl,--build-id=sha1 -Wl,--no-rosegment -Wl,--fatal-warnings -Qunused-arguments -Wl,--no-undefined -shared -Wl,-soname,libtest.so -o libtest.so CMakeFiles/test.dir/test.c.o  -latomic -lm && cmd.exe /C "cd /D T:\tmp\droid_cmake_test\build && CMAKE_STRIP-NOTFOUND --strip-unneeded T:/tmp/droid_cmake_test/build/libtest.so""
'CMAKE_STRIP-NOTFOUND' is not recognized as an internal or external command,
operable program or batch file.
ninja: build stopped: subcommand failed.
>cmake --version
cmake version 3.20.0

CMake suite maintained and supported by Kitware (kitware.com/cmake).

[hhb]

Sent a fix upstream here.

Once that is in we can figure out what should we do in NDK side.

[hhb]

@DanAlbert Can you help change this to 3.20.1 and cherry-pick it to r23? I have some mysterious permission issue to upload a change...

[DanAlbert]

Done (and thanks!)

I'm a little unsure how the testing works here. I see we have https://cs.android.com/android/platform/superproject/+/master-ndk:ndk/ndk/test/types.py;l=421-424, but prebuilts/cmake only has CMake 3.18. Is the case here that our new toolchain file more or less works with 3.18 but 3.20 (and now 3.21) is the better default since some features weren't ready?

[hhb]

Done (and thanks!)

The fix will be in CMake 3.20.1. Not 3.21. But surely you can set it to 3.21 to be safe.

I'm a little unsure how the testing works here.

I cherry-picked necessary changes into that Cmake before 3.20 is available. 😀

[DanAlbert]

The fix will be in CMake 3.20.1. Not 3.21. But surely you can set it to 3.21 to be safe.

I had misread, but I think I'm going to stick with it. Potential behavior changes shouldn't be introduced because the user took a bugfix release of CMake IMO. Sound reasonable to you?

[enh-google]

(i know that question wasn't addressed to me, but from hhb's comment, your comment, and my usual cowardice about such things, it looks like we're all in agreement that 3.21 is the safe choice :-) )

[hhb]

Sounds good. The alternative is to revert this for 3.20.0. But I guess that's too much trouble.

[baldurk]

Just to clarify then it sounds like my original approach should be fine, if that executable moves location in a future NDK then the $CMAKE_STRIP variable will be pointing to the new right place?

[DanAlbert]

I believe that's the intent, yes. @hhb is the expert here so he'll correct me if I'm wrong :)