Segfault when calling QWidget::show (on Debian 9)

Christian Ehrlicher

@Bart_Vandewoestyne said in Segfault when calling QWidget::show (on Debian 9):

allowed to share any code.

Then good luck. We can't guess your code...

Apart from this you already shared code.

Bart_Vandewoestyne

@Christian-Ehrlicher said in Segfault when calling QWidget::show (on Debian 9):

I would guess this is either a nullptr or not initialized. Build your app with debug information, go to stack frame 7 and print out the value of mpMainWindow .

I've added the -g option to our release build and when I run the application in gdb it now segfaults with the following call stack:

user@debianvbox:~/SVN/PolarisRel/Apps$ gdb ./PolarisSlave 
GNU gdb (Debian 7.12-6) 7.12.0.20161007-git
Copyright (C) 2016 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "x86_64-linux-gnu".
Type "show configuration" for configuration details.
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>.
Find the GDB manual and other documentation resources online at:
<http://www.gnu.org/software/gdb/documentation/>.
For help, type "help".
Type "apropos word" to search for commands related to "word"...
Reading symbols from ./PolarisSlave...done.
(gdb) r
Starting program: /home/user/SVN/PolarisRel/Apps/PolarisSlave 
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".

Program received signal SIGSEGV, Segmentation fault.
strlen () at ../sysdeps/x86_64/strlen.S:106
106	../sysdeps/x86_64/strlen.S: No such file or directory.
(gdb) bt
#0  strlen () at ../sysdeps/x86_64/strlen.S:106
#1  0x00007ffff3e101ed in XSetCommand () from /usr/lib/x86_64-linux-gnu/libX11.so.6
#2  0x00007ffff3e147f0 in XSetWMProperties () from /usr/lib/x86_64-linux-gnu/libX11.so.6
#3  0x00007ffff659007d in QWidgetPrivate::create_sys(unsigned long, bool, bool) () from /home/user/SVN/PolarisRel/ThirdParty/Qt/qt-install/lib/libQtGui.so.4
#4  0x00007ffff6548769 in QWidget::create(unsigned long, bool, bool) () from /home/user/SVN/PolarisRel/ThirdParty/Qt/qt-install/lib/libQtGui.so.4
#5  0x00007ffff6550697 in QWidget::setVisible(bool) () from /home/user/SVN/PolarisRel/ThirdParty/Qt/qt-install/lib/libQtGui.so.4
#6  0x000055555594a1fd in QWidget::show (this=<optimized out>) at ../../ThirdParty/Qt/qt-install/include/QtGui/qwidget.h:497
#7  BSPPolarisSlave::mfRun (this=0x7fffffffe0e0, argc=<optimized out>, argv=<optimized out>, errormsg=...) at BSPPolarisSlave.cpp:443
#8  0x0000555555c010da in ICService::mfExec(int, char**, QString&, bool) ()
#9  0x0000555555bfd90e in ICService::mfParseArguments(int, char**, bool) ()
#10 0x000055555594dbb3 in BSPPolarisSlave::mfParseArguments (this=0x7fffffffe0e0, argc=2, argv=0x7fffffffe258) at BSPPolarisSlave.cpp:659
#11 0x000055555592a3ad in main (argc=1, argv=0x7fffffffe258) at BSPPolarisSlaveMain.cpp:71
(gdb) f 7
#7  BSPPolarisSlave::mfRun (this=0x7fffffffe0e0, argc=<optimized out>, argv=<optimized out>, errormsg=...) at BSPPolarisSlave.cpp:443
443			mpMainWindow->show();
(gdb) p mpMainWindow
$1 = (BSPPolarisSlaveMainWindow *) 0x555556627c40

Some things I noticed are:

mpMainWindow is not nullptr.
In the call to BSPPolarisSlave::mfRun argc and argv are marked as 'optimized out'... and similarly, in the call to QWidget::show, the this parameter is also 'optimized out'. I have not much experience with gdb (most of the time, I debug in the Visual Studio debugger)... but could this 'optimizing out' be the problem?

J.Hilk

@Bart_Vandewoestyne said in Segfault when calling QWidget::show (on Debian 9):

mpMainWindow is not nullptr

are you sure it is initialised then ? gdb, in contrast to its MSVC equivalent, does no null initialisations during debug runs. So an uninitialised pointer is very rarely a nullptr

Bart_Vandewoestyne

@J-Hilk said in Segfault when calling QWidget::show (on Debian 9):

are you sure it is initialised then ? gdb, in contrast to its MSVC equivalent, does no null initialisations during debug runs. So an uninitialised pointer is very rarely a nullptr

For as far as I can see yes, because right before the call to show(), the pointer is initialized:

mpMainWindow = new BSPPolarisSlaveMainWindow(this, windowsCaption, 0, true, Qt::Window | Qt::WindowTitleHint | Qt::WindowSystemMenuHint);
connect(mpApplication, SIGNAL(lastWindowClosed()), mpApplication, SLOT(quit()));
mpMainWindow->show();

Bart_Vandewoestyne

@Christian-Ehrlicher said in Segfault when calling QWidget::show (on Debian 9):

Then good luck. We can't guess your code...

Apart from this you already shared code.

In the past on this forum, I've had good answers leading to a solution even without sharing code. I do believe that's possible :-)

And you are right: I have shared some code snippets. That is indeed not consistent with what I wrote, but I am somehow assuming that I am allowed to share small, non meaningful snippets of code that do not reveal any company secrets, if that can help us get to a solution quicker. I hope no one in our company will blame me for that... Finding the right balance between what you can share in order to get to a solution quicker is not always easy, but I try to find that balance.

Bart_Vandewoestyne

Some more info on this problem:

It is only a release build on Debian 9 that segfaults. As mentioned earlier, the debug build on Debian 9 runs fine.
Release builds and debug builds on Debian 8 and Red Hat Enterprise Linux 8.5 run fine!

Bart_Vandewoestyne

@Christian-Ehrlicher said in Segfault when calling QWidget::show (on Debian 9):

wrt to your strange copy stuff (whyever you need to modify your command line - sounds like a strange hack for me):
std::vector<char*> newArgs;
newArgs.push_back(argv[0]);
newArgs.push_back(const_cast<char*>("-e"));
for (int i = 1; i < argc; ++i)
  newArgs.push_back(argv[i]);
argc += 1;
...mfParseArguments(argc, newArgs.data());

I totally agree that that modification of the command line is strange. Note that this was not my idea, but I inherited this legacy code from my predecessors :-(
I tried your suggestion using std::vector instead of using an array of char*, but that also didn't solve the segfault.

JonB

@Bart_Vandewoestyne
Just so you know. Your segfault emanates from this line: https://code.woboq.org/kde/qt4/src/gui/kernel/qwidget_x11.cpp.html#804

        XSetWMProperties(dpy, id, 0, 0,
                         qApp->d_func()->argv, qApp->d_func()->argc,
                         &size_hints, &wm_hints, &class_hint);

(Doubtless some sort of X set window manager properties on start up?) It's on a strlen() from there, so presumably some element in qApp->d_func()->argv is wrong. So you're still on the argv issue. Try to print out everything in the last argv you pass on.

J.Hilk

@Bart_Vandewoestyne new debian, huh. New/updated compiler then as well?

I assume you have tried the release build with -O0 ?

Christian Ehrlicher

@J-Hilk or run it with valgrind (compile with -O2 and -g)

J.Hilk

@Christian-Ehrlicher never used valgrind before, as I usually don't do linux stuff. But I trust your expertise :D

oh it also now supports macOS, maybe I should give it a try sometime soon than!

Bart_Vandewoestyne

@J-Hilk said in Segfault when calling QWidget::show (on Debian 9):

@Bart_Vandewoestyne new debian, huh. New/updated compiler then as well?

Yes, due to the switch from Debian 8 to Debian 9, a new compiler as well. Debian 8 (where everything works) has

dev@debian8:~$ g++ --version | head -1
g++ (Debian 4.9.2-10+deb8u2) 4.9.2

while Debian 9 (where the release build segfaults) has

user@debianvbox:~$ g++ --version | head -1
g++ (Debian 6.3.0-18+deb9u1) 6.3.0 20170516

I assume you have tried the release build with -O0 ?

I hadn't, but now I have ;-) And I have interesting news: when using -O0 the segfault is gone! From -O1 and further, we get the segfault.

Bart_Vandewoestyne

@Christian-Ehrlicher said in Segfault when calling QWidget::show (on Debian 9):

@J-Hilk or run it with valgrind (compile with -O2 and -g)

I have no experience with valgrind, but looks like a good suggestion so I will try and report back.

Bart_Vandewoestyne

@Christian-Ehrlicher said in Segfault when calling QWidget::show (on Debian 9):

@J-Hilk or run it with valgrind (compile with -O2 and -g)

OK, so I compiled with -O2 and -g and ran my program through valgrind. This is what I got:

user@debianvbox:~/SVN/PolarisRel/Apps$ valgrind ./PolarisSlave
==5165== Memcheck, a memory error detector
==5165== Copyright (C) 2002-2015, and GNU GPL'd, by Julian Seward et al.
==5165== Using Valgrind-3.12.0.SVN and LibVEX; rerun with -h for copyright info
==5165== Command: ./PolarisSlave
==5165== 
==5165== Invalid read of size 8
==5165==    at 0x8B3B205: XSetCommand (in /usr/lib/x86_64-linux-gnu/libX11.so.6.3.0)
==5165==    by 0x8B3F7EF: XSetWMProperties (in /usr/lib/x86_64-linux-gnu/libX11.so.6.3.0)
==5165==    by 0x5EE707C: QWidgetPrivate::create_sys(unsigned long, bool, bool) (in /home/user/SVN/PolarisRel/ThirdParty/Qt/qt-install/lib/libQtGui.so.4.8.7)
==5165==    by 0x5E9F768: QWidget::create(unsigned long, bool, bool) (in /home/user/SVN/PolarisRel/ThirdParty/Qt/qt-install/lib/libQtGui.so.4.8.7)
==5165==    by 0x5EA7696: QWidget::setVisible(bool) (in /home/user/SVN/PolarisRel/ThirdParty/Qt/qt-install/lib/libQtGui.so.4.8.7)
==5165==    by 0x4FE1FC: show (qwidget.h:497)
==5165==    by 0x4FE1FC: BSPPolarisSlave::mfRun(int, char**, QString&) (BSPPolarisSlave.cpp:443)
==5165==    by 0x7B50D9: ICService::mfExec(int, char**, QString&, bool) (in /home/user/SVN/PolarisRel/Apps/PolarisSlave)
==5165==    by 0x7B190D: ICService::mfParseArguments(int, char**, bool) (in /home/user/SVN/PolarisRel/Apps/PolarisSlave)
==5165==    by 0x501BB2: BSPPolarisSlave::mfParseArguments(int, char**) (BSPPolarisSlave.cpp:659)
==5165==    by 0x4DE3AC: main (BSPPolarisSlaveMain.cpp:71)
==5165==  Address 0xbe71780 is 0 bytes after a block of size 16 alloc'd
==5165==    at 0x4C2C93F: operator new[](unsigned long) (vg_replace_malloc.c:423)
==5165==    by 0x501A27: BSPPolarisSlave::mfParseArguments(int, char**) (BSPPolarisSlave.cpp:637)
==5165==    by 0x4DE3AC: main (BSPPolarisSlaveMain.cpp:71)
==5165== 
==5165== Invalid read of size 1
==5165==    at 0x4C2EDA2: strlen (vg_replace_strmem.c:454)
==5165==    by 0x8B3B1EC: XSetCommand (in /usr/lib/x86_64-linux-gnu/libX11.so.6.3.0)
==5165==    by 0x8B3F7EF: XSetWMProperties (in /usr/lib/x86_64-linux-gnu/libX11.so.6.3.0)
==5165==    by 0x5EE707C: QWidgetPrivate::create_sys(unsigned long, bool, bool) (in /home/user/SVN/PolarisRel/ThirdParty/Qt/qt-install/lib/libQtGui.so.4.8.7)
==5165==    by 0x5E9F768: QWidget::create(unsigned long, bool, bool) (in /home/user/SVN/PolarisRel/ThirdParty/Qt/qt-install/lib/libQtGui.so.4.8.7)
==5165==    by 0x5EA7696: QWidget::setVisible(bool) (in /home/user/SVN/PolarisRel/ThirdParty/Qt/qt-install/lib/libQtGui.so.4.8.7)
==5165==    by 0x4FE1FC: show (qwidget.h:497)
==5165==    by 0x4FE1FC: BSPPolarisSlave::mfRun(int, char**, QString&) (BSPPolarisSlave.cpp:443)
==5165==    by 0x7B50D9: ICService::mfExec(int, char**, QString&, bool) (in /home/user/SVN/PolarisRel/Apps/PolarisSlave)
==5165==    by 0x7B190D: ICService::mfParseArguments(int, char**, bool) (in /home/user/SVN/PolarisRel/Apps/PolarisSlave)
==5165==    by 0x501BB2: BSPPolarisSlave::mfParseArguments(int, char**) (BSPPolarisSlave.cpp:659)
==5165==    by 0x4DE3AC: main (BSPPolarisSlaveMain.cpp:71)
==5165==  Address 0x50 is not stack'd, malloc'd or (recently) free'd
==5165== 
==5165== 
==5165== Process terminating with default action of signal 11 (SIGSEGV)
==5165==  Access not within mapped region at address 0x50
==5165==    at 0x4C2EDA2: strlen (vg_replace_strmem.c:454)
==5165==    by 0x8B3B1EC: XSetCommand (in /usr/lib/x86_64-linux-gnu/libX11.so.6.3.0)
==5165==    by 0x8B3F7EF: XSetWMProperties (in /usr/lib/x86_64-linux-gnu/libX11.so.6.3.0)
==5165==    by 0x5EE707C: QWidgetPrivate::create_sys(unsigned long, bool, bool) (in /home/user/SVN/PolarisRel/ThirdParty/Qt/qt-install/lib/libQtGui.so.4.8.7)
==5165==    by 0x5E9F768: QWidget::create(unsigned long, bool, bool) (in /home/user/SVN/PolarisRel/ThirdParty/Qt/qt-install/lib/libQtGui.so.4.8.7)
==5165==    by 0x5EA7696: QWidget::setVisible(bool) (in /home/user/SVN/PolarisRel/ThirdParty/Qt/qt-install/lib/libQtGui.so.4.8.7)
==5165==    by 0x4FE1FC: show (qwidget.h:497)
==5165==    by 0x4FE1FC: BSPPolarisSlave::mfRun(int, char**, QString&) (BSPPolarisSlave.cpp:443)
==5165==    by 0x7B50D9: ICService::mfExec(int, char**, QString&, bool) (in /home/user/SVN/PolarisRel/Apps/PolarisSlave)
==5165==    by 0x7B190D: ICService::mfParseArguments(int, char**, bool) (in /home/user/SVN/PolarisRel/Apps/PolarisSlave)
==5165==    by 0x501BB2: BSPPolarisSlave::mfParseArguments(int, char**) (BSPPolarisSlave.cpp:659)
==5165==    by 0x4DE3AC: main (BSPPolarisSlaveMain.cpp:71)
==5165==  If you believe this happened as a result of a stack
==5165==  overflow in your program's main thread (unlikely but
==5165==  possible), you can try to increase the size of the
==5165==  main thread stack using the --main-stacksize= flag.
==5165==  The main thread stack size used in this run was 8388608.
==5165== 
==5165== HEAP SUMMARY:
==5165==     in use at exit: 1,121,308 bytes in 9,104 blocks
==5165==   total heap usage: 22,310 allocs, 13,206 frees, 4,039,124 bytes allocated
==5165== 
==5165== LEAK SUMMARY:
==5165==    definitely lost: 2,944 bytes in 6 blocks
==5165==    indirectly lost: 13,190 bytes in 537 blocks
==5165==      possibly lost: 54,718 bytes in 437 blocks
==5165==    still reachable: 1,050,456 bytes in 8,124 blocks
==5165==         suppressed: 0 bytes in 0 blocks
==5165== Rerun with --leak-check=full to see details of leaked memory
==5165== 
==5165== For counts of detected and suppressed errors, rerun with: -v
==5165== ERROR SUMMARY: 5 errors from 2 contexts (suppressed: 0 from 0)
Segmentation fault

I'll try to decipher this myself, but if in the meanwhile someone more experienced with valgrind can point me in the right direction, that would be nice :-)

J.Hilk

from my point of view, I would say the issue is with your strange string manipulation stuff.

You have to be very careful, when working with stringliterals, it is super easy to run into undefined behaviour, when you try to modify them.

Christian Ehrlicher

@Bart_Vandewoestyne said in Segfault when calling QWidget::show (on Debian 9):

by 0x501A27: BSPPolarisSlave::mfParseArguments(int, char**) (BSPPolarisSlave.cpp:637)

This is where you have to take a look on. You do something wrong there for an argument.

kshegunov

Please provide the the line that @Christian-Ehrlicher mentioned; it's the call to mfExec.

Bart_Vandewoestyne

OK, I think we're getting there... In ICBlackBoxBase::mfInitialize we call the ICBlackBoxBaseApplication constructor which calls the QApplication constructor with certain argc and argv arguments:

ICBlackBoxBaseApplication::ICBlackBoxBaseApplication(int &argc, char** argv, ICBlackBoxBase* apApp)
: QApplication(argc,argv), mpApp(apApp)
{
}

Now let's see what argc and argv we are passing there. I've set a breakpoint right before the location where we call this constructor, and this is the call stack:

(gdb) bt
#0  ICBlackBoxBase::mfInitialize (this=0x7fffffffe0e0, argc=2, argv=0x5555565cd3d0, errormsg=...) at ICBlackBoxBase.cpp:101
#1  0x0000555555c0101d in ICService::mfExec(int, char**, QString&, bool) ()
#2  0x0000555555bfd86e in ICService::mfParseArguments(int, char**, bool) ()
#3  0x000055555594db13 in BSPPolarisSlave::mfParseArguments (this=0x7fffffffe0e0, argc=2, argv=0x7fffffffe258) at BSPPolarisSlave.cpp:654
#4  0x000055555592a3bd in main (argc=1, argv=0x7fffffffe258) at BSPPolarisSlaveMain.cpp:71

As you can see, in main we have that argc is 1, but in ICBlackBoxBase::mfInitialize (the function from which we call the ICBlackBoxBaseApplication constructor, and thus also the QApplication constructor) we have that argc is 2 (since an extra -e argument was added). Now let's look at argv in both main and ICBlackBoxBase::mfInitialize. In main we have:

(gdb) f 4
#4  0x000055555592a3bd in main (argc=1, argv=0x7fffffffe258) at BSPPolarisSlaveMain.cpp:71
71		return (polarisSlave.mfParseArguments(argc, argv));	
(gdb) p argc
$12 = 1
(gdb) p argv[0]
$13 = 0x7fffffffe53a "/home/user/SVN/PolarisRel/Apps/PolarisSlave"
(gdb) p argv[argc]
$14 = 0x0

but in ICBlackBoxBase::mfInitialize we have:

(gdb) f 0
#0  ICBlackBoxBase::mfInitialize (this=0x7fffffffe0e0, argc=2, argv=0x5555565cd3d0, errormsg=...) at ICBlackBoxBase.cpp:101
101	{
(gdb) p argc
$15 = 2
(gdb) p argv[0]
$16 = 0x5555565cf830 "/home/user/SVN/PolarisRel/Apps/PolarisSlave"
(gdb) p argv[1]
$17 = 0x5555565cf940 "-e"
(gdb) p argv[argc]
$18 = 0x20 <error: Cannot access memory at address 0x20>

so there argv[argc] is not null! And now I have to find out why :-)

JonB

@Bart_Vandewoestyne said in Segfault when calling QWidget::show (on Debian 9):

so there argv[argc] is not null! And now I have to find out why :-)

Earlier I wrote:

I shall be surprised if it is this, but....

I think your code is not 100% technically correct. You do not NULL terminate your new vector. Technically you should find your original argv had an extra element at the end: argv[argc] == NULL. You do not copy this or NULL terminate your new newArgvs. E.g. https://stackoverflow.com/questions/16418932/is-argvargc-equal-to-null-pointer
— argv[argc] shall be a null pointer.

It is not clear whether this matters or not. If code only uses argc to index up to argv[argc - 1] then it does not. If code does do something about looking at argv[argc] to check for nullptr then it does matter. If you have the source code where it goes wrong you may be able to delermine.

JonB

@Bart_Vandewoestyne said in Segfault when calling QWidget::show (on Debian 9):

ICBlackBoxBaseApplication::ICBlackBoxBaseApplication(int &argc, char** argv, ICBlackBoxBase* apApp)
: QApplication(argc,argv), mpApp(apApp)
{
}

I have an alternative theory. Work with me on this one, please!

I believe the BSPPolarisSlave & ICService classes are your own(?). This ICBlackBoxBaseApplication constructor calls QApplication(argc,argv) with argc, and especially argv, *with the values they have on entry to the constructor. However (I believe you are saying) something it then does alters argc, and especially argv, to add further arguments, right? Note in your traceback how argv has a different (pointer) value from BSPPolarisSlave::mfParseArguments() down to ICBlackBoxBase::mfInitialize() --- it has been changed.

Now, I am unclear whether the original argv, with its strings, is being used from the original QApplication(argc,argv), and just possibly the value/strings array is no longer valid. But it may be what is used the very first time QWidget::show() is called (to do some X initialisations from any (potential) X-type command-line arguments passed to your Qt app. Remember that https://doc.qt.io/qt-5/qcoreapplication.html#QCoreApplication says

Warning: The data referred to by argc and argv must stay valid for the entire lifetime of the QCoreApplication object. In addition, argc must be greater than zero and argv must contain at least one valid character string.

I think you should find that QWidget::show() which is in BSPPolarisSlave::mfRun(). On the line above it do something about printing out Qt's idea of the command line arguments which were used during the earlier QApplication(argc,argv). I believe static QStringList QCoreApplication::arguments() accesses these. So qDebug() those before QWidget::show(), do they look good or garbled?

Finally, the question is why/what are these classes altering argc/argvfor, apparently to add additional arguments? If these new arguments are intended to be seen by Qt then it looks like you should not be callingQApplication(argc,argv)` until after they have been modified to hold the new command line?