On Thu, Aug 30, 2001 at 07:34:57AM +0200, Lourens Veen wrote:
> Stephen Robert Norris wrote:
> > I'd be interested to know how to avoid that. I'm pretty sure I can
> > construct a scenario (with multiple threads and memory mapping,
> > for example) where it's impossible to tell until you get the SEGV. For
> > instance, I memory map a file, pass a pointer into the mapped
> > region into the library and then unmap it some time later from another
> > thread.
> > Even if the library were checking (and I'm not sure how it could) that
> > the pointer points to valid address space, there will be a time gap
> > between the check and the use, and my unmapping can get in there.
> > Having the library install its' own signal handler is not an acceptable
> > solution, either.
> Well, call me stupid, but isn't that what mutexes are for? Thread 1 sets
> the mutex, then calls the library with a pointer to some part of the
> shared memory. Make sure thread 2 checks the mutex before unmapping and
> there's no problem at all.
> Thing is, how is the library going to know whether the pointer is valid
> or not? All the standard C functions that expect pointers will happily
> write wherever you point them to, even if it causes a segfault. I don't
> see how this is a problem with the library. If I divide by zero (which
> is essentially calling the divide function with illegal values) I get an
> exception as well.
Yes, this is a way the application can avoid the problem; it's not a way
the library can.
My point was that it's impossible with modern OS's to avoid the possibility
of the library crashing.
Stephen Norris [EMAIL PROTECTED]
Farrow Norris Pty Ltd +61 417 243 239
Gimp-developer mailing list