Crash: Ksnip 1.2.0 binary segfaults when compiled in x86_64 with -fPIC in gcc-5.4.0 about ksnip HOT 23 CLOSED

No worries regarding the 1.1.0, it's just a variable that I forgot to change when I released the version, I've noticed it myself few minutes ago and changed it. See commit 8ce4b01

You can download from master and the install procedure again and it will be 1.2.0.

Regarding your crash, does it start only with gdb or also without gdb now?

from ksnip.

hmm, let's try something. Open up src/widgets/PaintArea.cpp, go to line 310 and comment following out (just add // in front of the line):

void PaintArea::setCursor()
{
    // delete mCursor;       <-- Like here
    mCursor = cursor();

    for ( int i = 0; i < views().length(); i++ ) {
        views().at( i )->setCursor( *mCursor );
    }
}

Save, delete all from build directory and run cmake and make again. Try to run ./ksnip again, see what happens.

from ksnip.

Great! Thanks to you for developing such a great tool!

from ksnip.

What distro are you using?

from ksnip.

Slackware-current, 64-bit series.
gcc 5.4.0

from ksnip.

ksnip_strace.txt

Attaching strace output (both stdin & stderr) of ksnip 1.2.0 invocation.

from ksnip.

I've just installed slackware 14.2 64bit in a virtual machine and ksnip was installed and started right away, without installing or changing anything else. So it seems for now not related to slackware rather something related to your system.

http://i.imgur.com/svYoS5Y.png

Can you try to run the ksnip in gdb, maybe we can see something there.
go back to the build directory and now run cmake but instead of cmake .. use cmake -DCMAKE_BUILD_TYPE=Debug ..

Then continue with make and gdb ./knsip
You should see the gdb prompt, type run and watch the output. If you get the prompt again without crash type next and hit enter, keep doing that until the crash happens or ksnip starts. Send me the output that you collect from there.

from ksnip.

I've just installed slackware 14.2 64bit in a virtual machine

Please be advised that Slackware 14.2 and Slackware-current differ significantly as of now (the latter has more up-to-date libraries and toolchain, etc.)
Will do what you suggest. Thanks!

from ksnip.

it's funny. I did what you said and now it runs correctly. Output:

(gdb) run
Starting program: /run/media/eduardo/Seagate Backup Plus Drive/build/slackbuilds/ksnip/ksnip-1.2.0/build/ksnip 
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".

ksnip run correctly. However it still identifies itself as 1.1.0.

from ksnip.

I found the issue!

I usually build ksnip as a Slackware package and for that we use bash scripts we call "SlackBuilds".

This particular SlackBuild was adapted from other software. There was a line where, if the CPU architecture was x86_64, it gave gcc as compiler flags the following; -O2 -fPIC.

Since there wasn't any such flags when compiling ksnip "by hand", I tried building with the -fPIC flag deleted. And it built without problem!

Ksnip now works correctly with the SlackBuild after deleting -fPIC from the compiler flags

The fact, however, remains: the -fPIC flag produced correct builds in 1.1.0 while in 1.2.0 gave segfaulting binaries.

For now, I deleted the -fPIC flag from my SlackBuild and Ksnip run correctly again. Thanks!!

from ksnip.

Updated bug title to better reflect the crash case.

from ksnip.

It really is strange, but I'm glad that you found the issue.
I'll check later regarding the -fPIC flag.

from ksnip.

I just built a SlackBuild with -fPIC enabled and with the debug flag. I then ran gdb on the ksnip executable (which promptly segfaulted). This is the relevant gdb output:

(gdb) run
Starting program: /bin/ksnip 
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff6df0000 in QCursor::~QCursor() () from /usr/lib64/libQtGui.so.4
(gdb) 

Thanks!

from ksnip.

I'm not familiar with the -fPIC flag but it looks like it should be used with shared libraries to make them more independent of where they're executed. So not sure if you should be using this flag in first place.

I'm also unable to reproduce the issue on my end. I have added set(CMAKE_POSITION_INDEPENDENT_CODE ON) to the CMakeLists.txt this is the equivalent of -fPIC, it compiles and runs without crash.

from ksnip.

Same on Slackware, no crash with PIC. Though I haven't tested it on -current which is a pre-release version for testing purposes.

from ksnip.

One more think we could check. Run gdb again, but when it crashes, run from the gdb prompt bt
Maybe the backtrace will show something.

from ksnip.

ok. First I forgot to give "next" when gdb showed a segfault. This is the full output:

(gdb) run
Starting program: /bin/ksnip 
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff6df0000 in QCursor::~QCursor() () from /usr/lib64/libQtGui.so.4
(gdb) next
Single stepping until exit from function _ZN7QCursorD2Ev,
which has no line number information.

Program terminated with signal SIGSEGV, Segmentation fault.
The program no longer exists.
(gdb) 

from ksnip.

Backtrace:

(gdb) run
Starting program: /bin/ksnip 
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib64/libthread_db.so.1".

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff6df0000 in QCursor::~QCursor() () from /usr/lib64/libQtGui.so.4
(gdb) bt
#0  0x00007ffff6df0000 in QCursor::~QCursor() () at /usr/lib64/libQtGui.so.4
#1  0x000000000042a980 in PaintArea::setCursor() ()
#2  0x000000000042c10a in MainWindow::loadSettings() ()
#3  0x000000000042e966 in MainWindow::MainWindow() ()
#4  0x000000000041a15c in main ()
(gdb) 

from ksnip.

After commenting out line 310 in PaintArea.cpp, Ksnip builds and run correctly even with the -fPIC flag enabled.
Thanks!!

from ksnip.

Interesting, so we know where the issue happens though I can't leave it like that, that would cause a memory leak.
I'll think about some solution then I will ask you again to test when I have a patch.

from ksnip.

Can you please download the latest version and try to build and run it?

https://github.com/DamirPorobic/ksnip/tree/v1.2.1

from ksnip.

I can confirm that 1.2.1 now builds and run correctly in x86_64 even with -O2 -fPIC enabled. Great!!

from ksnip.

Awesome! Thanks for testing and confirming. :)

from ksnip.