Application disappears without any crash dump

[Rahul]

Hi All,

We have a VC++ application (Native) developed in VSTS 2008. Sometimes
the application just vanishes without showing any crash dialog (The
default Windows Error Handler dialog) or generating any crash dump.
This happens in many systems which have identical installation,
Windows XP (SP3) with Windbg installed as the default debugger which
catches the exceptions and shows stack trace 99% of the time. But
sometimes the application just vanishes as if it had executed exit(0)
(Nothing in the TaskManager also)

Are there some exceptions which even WinDbg can not catch and hence
the program just terminates, What could be the other reasons for this.

Thanks in advance
Rahul

 

[Alf P. Steinbach]

* Rahul:

We have a VC++ application (Native) developed in VSTS 2008. Sometimes
the application just vanishes without showing any crash dialog (The
default Windows Error Handler dialog) or generating any crash dump.
This happens in many systems which have identical installation,
Windows XP (SP3) with Windbg installed as the default debugger which
catches the exceptions and shows stack trace 99% of the time. But
sometimes the application just vanishes as if it had executed exit(0)
(Nothing in the TaskManager also)

Are there some exceptions which even WinDbg can not catch and hence
the program just terminates, What could be the other reasons for this.

At least with g++, which uses the old msvcrt.dll MS runtime library, a stack
overflow can generate your observed behavior.

This is typically caused by an infinite recursion.

With novice programmers (so prevalent in the industry, many of them with 5+
years experience) it can conceivably also be caused by large raw arrays as
locals, including use of Microsoft's alloca-based Unicode/char conversion.

In either case it might help to turn on stack probe checking and try to
reproduce the faults. In the very last case, stack based string conversions, a
fix might be to make the app Unicode only. I.e., simply avoiding conversions.

But it sounds like your application is prone to crashing.

This might be caused by bad use of raw pointers, which in turn is caused by
complexity, which is a euphemism for spaghetti.

If that's the case then it's much more difficult to track down, because memory,
including the stack, might be corrupted in a way "just so" to foil your
detection attempt (this is a consequence of Murphy's law).


Cheers & hth.,

- Alf

 

[Goran]

+1 for Alf, stack exhaustion is the culprit no. 1.

Goran.

 

[Rahul]

Thanks Alf,

So infinite recursion and large stack based array's seems to be the
problem. But why does the default debugger not catch these crashes.
I even tried on the system where Visual Studio was installed, It also
failed to catch those exceptions (unless we run the program the inside
debugger itself)

Is there any way to catch these crashes without running the program in
the debugger, and why are they not caught by the debugger by default
(just for understanding the technical difficulty involved in this).

Thanks
Rahul

 

[Jochen Kalmbach]

Hi Rahul!

Is there any way to catch these crashes without running the program in
the debugger, and why are they not caught by the debugger by default
(just for understanding the technical difficulty involved in this).

The WER/Default-Debugger is called from within the application. If the
application is not able to catch the unhandled exception, then WER will
not be called.
In Vista / Windows 7 there was some improvement in the OS to better
handle such situations.

Greetings
Jochen

 

[Charlie Gibbs]

Thanks Alf,

So infinite recursion and large stack based array's seems to be the
problem. But why does the default debugger not catch these crashes.
I even tried on the system where Visual Studio was installed, It also
failed to catch those exceptions (unless we run the program the inside
debugger itself)

Is there any way to catch these crashes without running the program in
the debugger, and why are they not caught by the debugger by default
(just for understanding the technical difficulty involved in this).

My favourite way of dealing with these things is to define a buffer
on either site of the local variables:

void myfunc ()
{
char buffer1[512];
... other local variables ...
char buffer2[512];

memset ((void *) buffer1, 0, sizeof (buffer1));
memset ((void *) buffer2, 0, sizeof (buffer2));
... code ...
}

Often that's enough to stop the mysterious disappearances. You can
then check "buffer1" and "buffer2" at various points in the code to
see whether they suddenly become nonzero. That should catch simple
overflows, but a truly wild pointer could clobber the stack far
enough away that nothing bad happens until the program exits.
Still, it's a start...

 

[Charlie Gibbs]

Adding some way to monitor progress of the program can help you find
the part of your source code that causes a crash. I've used calls to
MessageBox. I add at least one to code I'm pretty sure runs before the
crash and at least one to code I'm pretty sure runs after the crash.
And after each crash, I narrow the gap between calls. This is very low
tech, and seems painful. But often enough, I can rather quickly narrow
the gap enough for me to see the likely problem, blanket the code with
ASSERTs, and/or step through with a debugger.

If you narrow it down using a kind of binary search, the process can
indeed be fairly fast.

A variation of this is to #ifdef out chunks of your code, if necessary
replacing them with a dummy routine that inserts needed values. Once
you get the program to stop crashing, start re-enabling sections of
code until it resumes crashing. Again, a binary search technique can
speed up the process. A possible fly in the ointment (which applies
to any technique that adds or removes code) is that the clobbered
memory location might move to someplace non-critical, giving the
illusion that you've found the bug when it's really just gone into
hiding.