On 10 Apr 2003, Tom Roeder wrote:
> I have a question to which I hope the answer is simple, but I haven't
> managed to find any way to do it short of hacking the code myself. I
> have a program (it happens to be ns, the network simulator) that I have
> been working on for a while. There's a memory leak, and it ends up
> getting a signal and dying on an out of memory error. The problem is
> that when it dies, valgrind gives up too! I would much rather have
> valgrind catch that signal and print out what it knows about the memory
> problems already. Similarly, it has occasionally happened that I have
> wanted to stop valgrind short in a run and tell it to print out what it
> knows. Unfortunately, at least when I did it, Ctrl-C just causes it to
> quit and not tell me anything. All of this could be solved if there
> were some sort of command-line option like
>
> "--catch-signal-and-report=<signals>"
>
> where I could give it a list of signals and it would report on memory
> problems before giving up when receiving them. I have a good knowledge
> of C and C++, but I don't have much time to try to understand the
> valgrind code and figure it out and add this option in a way that
> doesn't interfere with other valgrind signal functionality. If this has
> already been implemented, so much the better. If not, could anyone give
> me a pointer into where this might be done in the code, so I can try to
> add the signal handlers myself?
Can you catch the signal with the client program? If so, in the signal
handler, you could call the VALGRIND_DO_LEAK_CHECK macro (just #include
memcheck.h) to do a leak check at that point.
N
|
