I recently ran into this problem and could not find any good solution on the Internet. So next comes a small summary of the problem with hopefully enough buzzwords, so Google can lead you here.

If you want to do C++ development on Android, you need the NDK for cross compilation. It comes by default with its own build system called ndk-build, which basically is a bunch of custom makefiles. But if you are sharing code between the Android Platform and lets say plain Linux, you have likely already a build system installed. For C/C++ CMake is quite popular as it supports different platforms and compilers. Fortunately there is already a project which adds Android support to CMake. I will not cover that – instead I assume you are using it already.

Unfortunately you cant use the ndk-gdb script supplied with the NDK to debug your application as it relies on the behaviour of ndk-build. But as said earlier, ndk-build is no wizardy, but just a bunch of scripts. So it is possible to emulate the behaviour using CMake, as following:

Add the following macro to your CMakeLists.txt file

macro(ndk_gdb_debuggable TARGET_NAME)
    get_property(TARGET_LOCATION TARGET ${TARGET_NAME} PROPERTY LOCATION)
    
    # create custom target that depends on the real target so it gets executed afterwards
    add_custom_target(NDK_GDB ALL) 
    add_dependencies(NDK_GDB ${TARGET_NAME})
    
    set(GDB_SOLIB_PATH ${PROJECT_SOURCE_DIR}/obj/local/${ANDROID_NDK_ABI_NAME}/)
    
    # 1. generate essential Android Makefiles
    file(WRITE ${PROJECT_SOURCE_DIR}/jni/Android.mk "APP_ABI := ${ANDROID_NDK_ABI_NAME}\n")
    file(WRITE ${PROJECT_SOURCE_DIR}/jni/Application.mk "APP_ABI := ${ANDROID_NDK_ABI_NAME}\n")

    # 2. generate gdb.setup
    get_directory_property(PROJECT_INCLUDES DIRECTORY ${PROJECT_SOURCE_DIR} INCLUDE_DIRECTORIES)
    string(REGEX REPLACE ";" " " PROJECT_INCLUDES "${PROJECT_INCLUDES}")
    file(WRITE ${PROJECT_SOURCE_DIR}/libs/${ANDROID_NDK_ABI_NAME}/gdb.setup "set solib-search-path ${GDB_SOLIB_PATH}\n")
    file(APPEND ${PROJECT_SOURCE_DIR}/libs/${ANDROID_NDK_ABI_NAME}/gdb.setup "directory ${PROJECT_INCLUDES}\n")

    # 3. copy gdbserver executable
    file(COPY ${ANDROID_NDK}/prebuilt/android-arm/gdbserver/gdbserver DESTINATION ${PROJECT_SOURCE_DIR}/libs/${ANDROID_NDK_ABI_NAME}/)

    # 4. copy lib to obj
    add_custom_command(TARGET NDK_GDB POST_BUILD COMMAND mkdir -p ${GDB_SOLIB_PATH})
    add_custom_command(TARGET NDK_GDB POST_BUILD COMMAND cp ${TARGET_LOCATION} ${GDB_SOLIB_PATH})

    # 5. strip symbols
    add_custom_command(TARGET NDK_GDB POST_BUILD COMMAND ${CMAKE_STRIP} ${TARGET_LOCATION})
endmacro()

Then use it like

add_library(YourTarget ...)
ndk_gdb_debuggable(YourTarget)

You should now be able to use ndk-gdb with CMake, just as if you would have used ndk-build.

Note that steps 4 and 5 are optional for debugging. They just reduce the size of the library that has to be transferred to the device. If you dont care, you can just leave them out. But then the solib search path from step 2 must be set to:

file(WRITE ./libs/${ANDROID_NDK_ABI_NAME}/gdb.setup "set solib-search-path ./libs/${ANDROID_NDK_ABI_NAME}\n")

Ideally someone should integrate that in the Android toolchain linked above.

Update Merged Upstream

When I first read about GNOME dropping support for BSD and Solaris, my impression was that this is a good idea to aiming to unify limit resources and get the work done. I was also excited about the idea of the GNOME OS. I think it is necessary to keep the big picture in mind when developing the different components. Previously Ubuntu was the only project that did this and it was also the reason why I started using Ubuntu. Because it made the different parts of Linux work together to achieve the big goal of a great overall system.

But then things started to go wrong. Instead of picking existing components and giving them the final polish like Ubuntu did before, the GNOME project started developing things from scratch without any apparent reason to do so. And even worse: incompatible to existing solutions. It started with the rejection of the appindicator specification implemented by Ubuntu and KDE. At that point it was not clear to me whether the specification was broken or whether the responsible people at GNOME were just ignorant.

Then came systemd. And it started to be apparent that unfortunately it was the latter. To my knowledge Ubuntu is the biggest deployment of GNOME and it is based around the Linux ecosystem. So dropping support for Ubuntu has nothing to do with unifying limited resources. Ubuntu is your target audience, so if you should try to collaborate with a project you should collaborate with Ubuntu. My opinion on that is that some Fedora developers were pissed that the Unity interface was exclusive for Ubuntu and instead of packaging it for Fedora they started making GNOME Shell exclusive for Fedora.

Next I read about the overlay scrollbars re-developed for GNOME. While the first reaction might be the developers simply do not want to use Ubuntu technology, I think the reason is different. The developer does not seem to have any antipathy towards Ubuntu and if we look at the project he developed the scrollbars for another explanation becomes visible.

But first lets take a step back. Lets take a look at the core of GNOME. By this I mean the programming language it is written in. It is C/GObject; plain C extended with naming conventions and libraries to allow modern paradigms such as object oriented programming and events/ observer pattern. From today’s perspective one might wonder why one should choose this over C++, which integrates most of the features at the language level. But back when the GNOME project started C++ was not mature yet which meant that your program might break with the next compiler update or even the next STL update.

Therefore basing your project on plain C was a good idea. But a few years back it became obvious that programming in C/GObject seriosly lacked behind more modern programming languages like C++, Java and C# for application development.

Unfortunately instead of moving the straightforward route from C to C++, which most of C developers took when C++ matured(that was about 10 years ago), Vala was born.

So instead of using a proven and mature foundation, a new layer of indirection was created to essentially provide the same feature set. Commonly this is referred to as the “not invented here” symptom. A more derogative phrase would be reinventing the wheel..

What is sad here is that being an open source project, GNOME disregards the biggest advantage of open source software, namely standing on the shoulders of giants. With open source software you can use take an existing solution and improve upon it. This way you get the base functionality as well as the bug fixes that went in it for free. If you would develop it from scratch, you most likely would have to fix the same bugs again yourself.

To sum up here is what GNOME is losing right now

• 30 years of language and library experience by using Vala instead of C++
• 5 years of deployment and bug fixing by using systemd instead of extending upstart
• 1 year of development testing and design if they reimplement overlay-scrollbars
• 8 years of foundation development that went into Eclipse, by developing Gnome Builder from scratch
• but most importantly: the synergy effects by collaborating with others

Do not get me wrong, I am not saying that the GNOME solutions could be replaced by existing solutions – I am saying that by extending existing solutions the GNOME project and the free software landscape would be better off as a whole.