I often get questions regarding segmentation faults and bus errors.
Sometimes, these occue when one calls library functions such as
gethostbyname(). Some students think this is some kind of
a networking bug. Well, it's often not. I will try to answer
this type of questions here once and for all.
Chances are that you have corrupted memory. This usually means that
you have corrupted memory a while back. It just happened
that when you call gethostbyname(), the corrupted memory
caused a bus error or the execution of an illegal instruction.
Bus errors and illegal instructions are basically the same thing
as segmentation faults.
How does one corrupt memory? You can write beyond an allocated
memory block. You can free the same object twice. You can
free an object that was not allocated. You can write to an
object that's already freed. These bugs are hard to find
because most of the them you only see that there is problem
long time after you have corrupted memory. If you have access
to a professional/expensive debugging tool, it may be helpful.
Otherwise, you just need to do binary search and see where the
bug(s) might be. There's no magical cures in debugging
memory corruption bugs.
One thing you might try is to temporarily turn off memory
deallocation (if you suspect that you have freed the same
object twice or freed an object that was not allocated).
You can do the following to define free() as a
no-op in a common header file when you are debugging:
#define DEBUGGING_MEMORY_CORRUPTION
/* comment out the above line when you are done debugging */
#ifdef DEBUGGING_MEMORY_CORRUPTION
#ifdef free
#undef free
#endif /* free */
#define free
#endif /* DEBUGGING_MEMORY_CORRUPTION */
As your code gets more and more complicated, and you are not
very careful in managing your resources, you may get more of
these. This is one reason why you want to keep your code
nice and clean.
On nunki, you can try efence. There is a copy
installed in:
~csci551b/public/efence/2.2/ElectricFence-2.2.2-15
Please read:
~csci551b/public/efence/2.2/README-solaris
I have not tried this library and I do not know if it's
suitable for our programming assignments. But I've heard good things
and bad things about it. Please try it at your own risk.
If you develop your code on a Linux, you can try
valgrind.
I have not tried this tool and I do not know if it's
suitable for our programming assignments. But I've heard good things
(e.g., easy to use) and bad things about it (e.g., it only runs on Linux).
Please try it at your own risk.
Also, if you want to try these, incorporate them in your programs
as early as possible. Don't start learning it when bugs
are happening and deadlines are approaching.
|