Using Android NDK R10E, I am trying to build a shared library for all supported ABI's and I am getting the following error for some but not all ABI's:
[armeabi] SharedLibrary : libMyLib.so /home/user/android-ndk-r10e/toolchains/arm-linux-androideabi-4.8/prebuilt/linux-x86_64/bin/../lib/gcc/arm-linux-androideabi/4.8/../../../../arm-linux-androideabi/bin/ld:
fatal error: /home/user/source/MyLib/obj/local/armeabi/libMyLib.so: Input/output error
The project successfully builds for arm64-v8a, mips and mips64 but fails with the above error for armeabi, armeabi-v7a, x86 and x86_64.
I have a static library project and another shared library project and they both build successfully for all 7 ABI's.
If I compare the contents of obj/local/ for an ABI that builds and one that does not, they both contain all the same files except for libMyLib.so.
The difference between those two sets of ABIs is that the failing ones link using ld.gold and the working ones use ld.bfd.
Two things to try:
Use the 4.9 toolchain. It hopefully has the bug fix.
If that doesn't work, you can add -fuse-ld=bfd to your ldflags to use bfd even on the architectures that default to gold.
Same issue happened to me in r15c.
The fix was to copy
android-ndk-r15c/toolchains/arm-linux-androideabi-4.9/prebuilt/linux-x86_64/arm-linux-androideabi/bin/ld.bfd over ld.
I had to copy it because I could not easily find a way to specify this flag to CMake to use it while detecting the compiler features.
Related
I'm attempting to cross-compile from Linux (NixOS) to Windows and encountering some frustrations.
There seem to be two parts that together are breaking the build:
Code in my Rust project requires multithreading, and as such requires a version of libpthread for Windows.
To build properly, I need a build.rs file. For some reason, Rust requires a version of libpthread for Linux for that.
What's the problem? Well, the build.rs has to be built with regular GCC and not MinGW because it needs to execute on my system. But for some reason, GCC is attempting to link to the Windows libpthread library instead of the system one, and as such is failing with an error about not supporting the library format.
(Failed) Alternatives
If I remove the build.rs, the project builds fine. Unfortunately, I need it for full functionality.
If I remove the Windows version of libpthread the build.rs builds and runs correctly, but MinGW fails with a missing library error when building the rest of the project.
Solution Paths?
Either I have to figure out why GCC's linking to the wrong version of libpthread, or I have to disable -lpthread entirely for the build.rs. I have no idea why it would need pthreads, considering for testing I stripped it down to only fn main() {}.
I have no idea where to start on either of these, and I've already spent a couple of days getting the problem down to this. I'd appreciate some help!
EDIT: I managed to built x86_64 version of libreactnativejni.so (see my answer). However it didn't resolved my problem.
To sum up what I learned here is:
You cannot test android components, which are using native libraries on Linux, because "libandroid.so" (core android lib), doesn't exists for Linux.
Original question:
My goal is to run tests on my Linux box using PowerMockito similar to RootViewTest.java.
Not sure how react-native team managed to run this test, but from my experience mocking Arguments.createArray is not possible, without loading jni library reactnativejni. Reason is that Arguments.createArray() calls WritableNativeArray which in static initializer ReactBridge.staticInit() calls SoLoader.loadLibrary("reactnativejni").
So I thought I've no other option, but compile this reactnativejni shared library for linux, using 64bit android-ndk-r10e
I managed to load this jni lib and test my code using:
ndkDir=$(pwd)/react-ndk/all/x86 \
JAVA_OPTS="-Djava.library.path=\".:$ndkDir\"" \
LD_LIBRARY_PATH="$ndkDir:$LD_LIBRARY_PATH" ./gradlew app:test
But test failed with:
libreactnativejni.so: wrong ELF class: ELFCLASS32 (Possible cause: architecture word width mismatch)
Checked platform with objdump:
objdump -s --section .comment libreactnativejni.so
libreactnativejni.so: file format elf32-i386
My question how is possible that 64 bit ndk produced 32 bit library and how to force to produce 64 bit.
Library build script is written as gradle script:
https://github.com/facebook/react-native/blob/26684cf3adf4094eb6c405d345a75bf8c7c0bf88/ReactAndroid/build.gradle#L199
ReactAndroid explicitly requests 32-bit targets. If you wish, you can replace this line to read
APP_ABI := all
I'm going to answer my own question (thank Alex for good APP_ABI lead). To compile x86_64 variant of reactnativejni follow this steps:
Update APP_ABI to APP_ABI := armeabi-v7a x86_64 x86 in file ReactAndroid/src/main/jni/Application.ml
The next problem is dependency jsc-android, which is not build for X86_64 target. There is alternative jsc-android-buildscripts which is also built for X86_64.
Extract folder X86_64 directly from aar and put it in ReactAndroid/build/third-party-ndk/jsc/jni/x86_64
However compilation will be failing on X86_64 with errors: lambda capture initializers only available with -std=c++1y or -std=gnu++1y
Fix that error, by replacing all occurrences of -std=c++1y with -std=c++11 in all Android.mk files.
Now you can start build with: ./gradlew buildReactNdkLib
Optionally you might need to clean build folder with ./gradlew cleanReactNdkLib
Built reactnativejni - libreactnativejni.so will be present in ReactAndroid/build/react-ndk/all/x86_64.
But unfortunately libreactnativejni.so will not work on Linux, as it has dependency on libandroid.so, which doesn't exists for Linux.
I am following this site to compile OpenCV for ARM.
It could not find my compiler so i hardcoded it into cmake file
find_program(CMAKE_C_COMPILER NAMES arm-linux-gnueabi-gcc-4.7)
find_program(CMAKE_CXX_COMPILER NAMES arm-linux-gnueabi-g++-4.7)
set(ARM_LINUX_SYSROOT /usr/arm-linux-gnueabi CACHE PATH "ARM cross compilation system root")
It compiles to aproximately 50% and then throws the following error:
Linking CXX shared library ../../lib/libopencv_viz.so
/usr/lib/libvtkCharts.so.5.8.0: file not recognized: File format not recognized
collect2: error: ld returned 1 exit status
I am not every experienced in cross compilation (or straight compilation for the matter). How do i fix this?
I think it is a mismatch between libopencv_viz and libvtkCharts. Some of your 3rdparty libs are built for another platform. These libraries themselves must be recompiled from source to match the details (ABI, dynamic system library dependencies, etc) of the system on which they are intended to be used.
Compiling OpenCV 2.4.10 worked for me. I did not have any preferred version. If you want to compile v3.0 see #Kornel's answer, that suggests to leave viz library out of compilation.
Use this command to checkout v2.4.10
git checkout 2.4.10
I am using Ubuntu 13.04 and installed mingw-w64 using apt-get install mingw-w64. I can compile and link a working 64-bit version of my program with the following command:
x86_64-w64-mingw32-g++ code.cpp -o app.exe
Which generates a 64-bit app.exe file.
What binary or command line flags do I use to generate a 32-bit version of app.exe?
That depends on which variant of toolchain you're currently using. Both DWARF and SEH variants (which come starting from GCC 4.8.0) are only single-target. You can see it yourself by inspecting the directory structure of their distributions, i.e. they contain only the libraries with either 64- or 32-bit addressing, but not both. On the other hand, plain old SJLJ distributions are indeed dual-target, and in order to build 32-bit target, just supply -m32 flag. If that doesn't work, then just build with i686-w64-mingw32-g++.
BONUS
By the way, the three corresponding dynamic-link libraries (DLLs) implementing each GCC exception model are
libgcc_s_dw2-1.dll (DWARF);
libgcc_s_seh-1.dll (SEH);
libgcc_s_sjlj-1.dll (SJLJ).
Hence, to find out what exception model does your current MinGW-w64 distribution exactly provide, you can either
inspect directory and file structure of MinGW-w64 installation in hope to locate one of those DLLs (typically in bin); or
build some real or test C++ code involving exception handling to force linkage with one of those DLLs and then see on which one of those DLLs does the built target depend (for example, can be seen with Dependency Walker on Windows); or
take brute force approach and compile some test code to assembly (instead of machine code) and look for presence of references like ___gxx_personality_v* (DWARF), ___gxx_personality_seh* (SEH), ___gxx_personality_sj* (SJLJ); see Obtaining current GCC exception model.
Good day
Currently, I'm working on an embedded device based on arm-linux. I want to build GCC for my target architecture with Glibc. GCC builds successful, but I have trouble with Glibc build.
I use the latest version of Glibc (ftp.gnu.org/gnu/glibc/glibc-2.12.1.tar.gz) and port for them (ftp.gnu.org/gnu/glibc/glibc-ports-2.12.1.tar.gz)
my configuration line:
../../glibc-2.12.1/configure --host=arm-none-linux-gnueabi --prefix=/home/anatoly/Desktop/ARM/build/glibc-build --enable-add-ons --with-binutils=/home/anatoly/Desctop/ARM/toolchain/arm/bin/
configuration script work fine, but i get some compile error:
...
/home/anatoly/Desktop/ARM/src/glibc-2.12.1/malloc/libmemusage_pic.a(memusage.os): In function me':
/home/anatoly/Desktop/ARM/src/glibc-2.12.1/malloc/lmemusage.c:253: undefined reference to__eabi+read_tp'
...
I also tried using the old version (2.11, 2.10) but have the same error.
Does anybody know the solution for this problem?
Use a precompiled toolchain, like those provided by code sourcery.
If you want to make your own, optimised (premature optimization is the root of all evil), use crosstool-NG, which is a tool dedicated to cross-compilation toolchain building.
If you are not convinced, and want to do everything with your own hands, ask your question on the crosstool-NG mailing list.
Try substituting arm-linux-gnueabi for arm-none-linux-gnueabi. Check that a compiler, loader etc. with the prefix you used for "host" exist on your path.