Guidelines and HOWTOs/Debugging/Debugging symbols

Debugging symbols allow you to debug your application better. Debugging symbols are added to your binary by the compiler. You have to decide during the cmake step if you want debugging symbols or not. To compile your application with debugging symbols, use

cmake . -DCMAKE_BUILD_TYPE=debugfull

to compile it without debugging symbols, use

cmake . -DCMAKE_BUILD_TYPE=release

Depending on your decision, output generated with the command kDebug will also be (debugfull) or not be (release) added to your application.

As an example, let's write an application that crashes:

main.cpp

#include <KApplication>
#include <KAboutData>
#include <KCmdLineArgs>
#include <KMessageBox>
#include <iostream>

using namespace std;

int main (int argc, char *argv[])
{
    KAboutData aboutData( "tutorial1", 0, ki18n("Tutorial 1"), "1.0",
                          ki18n("Displays a KMessageBox popup"),
                          KAboutData::License_GPL,
                          ki18n("(c) 2009"), ki18n("Some text..."),
                          "http://tutorial.com/",
                          "[email protected]");

    KCmdLineArgs::init( argc, argv, &aboutData );
    KApplication app;

    KMessageBox::questionYesNo( 0, i18n( "Hello World" ) );
    int* i;
    cout << "i is at " << i << " value " << *i << endl;
    i=(int*)0x0;
    cout << "i is at " << i << " value " << *i << endl;

    return 0;
}

CMakeLists.txt

project (tutorial1)
find_package(KDE4 REQUIRED)
include (KDE4Defaults)
include_directories(${KDE4_INCLUDES})
set(tutorial1_SRCS main.cpp)
kde4_add_executable(tutorial1 ${tutorial1_SRCS})
target_link_libraries(tutorial1 ${KDE4_KDEUI_LIBS})
install(TARGETS tutorial1  ${INSTALL_TARGETS_DEFAULT_ARGS})

Now let's compile this without debugging symbols:

cmake . -DCMAKE_BUILD_TYPE=release && make -j4

We see that the resulting file is small:

# ll tutorial1
-rwxr-xr-x 1 root root 18879 Jul 11 18:07 tutorial1

With debugging symbols, the file is bigger:

cmake . -DCMAKE_BUILD_TYPE=debugfull && make
# ll tutorial1
-rwxr-xr-x 1 root root 256622 Jul 11 18:09 tutorial1

Backtraces

Now let's start the application and look at the backtrace:

cmake . -DCMAKE_BUILD_TYPE=release && make -j4

Gives you the backtrace

Application: Tutorial 1 (tutorial1), signal SIGSEGV
�[?1034h[Thread debugging using libthread_db enabled]
0x00007f58abba4cb0 in nanosleep () from /lib64/libc.so.6
[Current thread is 1 (Thread 0x7f58b0cfd750 (LWP 21264))]

Thread 1 (Thread 0x7f58b0cfd750 (LWP 21264)):
[KCrash Handler]
#5  0x00000000004016aa in main ()

The debugging build

cmake . -DCMAKE_BUILD_TYPE=debugfull && make -j4

Gives you the backtrace

Application: Tutorial 1 (tutorial1), signal SIGSEGV
�[?1034h[Thread debugging using libthread_db enabled]
0x00007fd0b8161cb0 in nanosleep () from /lib64/libc.so.6
[Current thread is 1 (Thread 0x7fd0bd2ba750 (LWP 21327))]

Thread 1 (Thread 0x7fd0bd2ba750 (LWP 21327)):
[KCrash Handler]
#5  0x0000000000401625 in main (argc=1, argv=0x7fffc52f5138) at /root/kdehello/main.cpp:25

So you see: with debugging symbols, you see the line number where the crash occurred. Without, you do not see this.

Where are they?

Where are the debugging symbols stored? Use objdump -g to find out:

# objdump  -g tutorial1-release | wc -l
511
# objdump  -g tutorial1-debugfull | wc -l
40943

It is important to know that the code lines (in assembler) to be executed actually do not differ a lot. We see this when disassembling the code:

# objdump -d tutorial1-debugfull | wc -l
658
# objdump -d tutorial1-release | wc -l
697

This gives us hope that there will be no major speed difference between a debug- and a release-version of a binary.

Speed implications

We remove the lines that cause the crash and the messagebox. Then we execute the program 100 times:

cmake . -DCMAKE_BUILD_TYPE=debugfull && make -j4
time for i in $(seq 1 1 100); do ./tutorial1; done

real    0m6.201s
user    0m4.368s
sys     0m1.320s

Lasts 6 seconds. Now with the release version:

cmake . -DCMAKE_BUILD_TYPE=release && make -j4
time for i in $(seq 1 1 100); do ./tutorial1; done

real    0m6.259s
user    0m4.368s
sys     0m1.328s

Also lasts 6 seconds. So the main difference is in the binaries size.