Re: 'stack overflow' message for Darwin; host hooks

[Geoffrey Keating <geoffk at apple dot com>]

On Friday, February 7, 2003, at 11:03  PM, Fergus Henderson wrote:

> On 07-Feb-2003, Geoffrey Keating <geoffk@apple.com> wrote:
> > On Friday, February 7, 2003, at 08:01  AM, Fergus Henderson wrote:
> >
> > > If an application runs out of stack space,
> > > why would it *ever* be better to issue the error message "Bus Error"
> > > rather than "Stack Overflow"??
> > Right, but that's not this patch; that would require kernel changes to
> > change the signal sent.  If you look at where the 'bus error' message
> > comes from, it's because the signal sent is (reported as) SIGBUS.  To
> > get a different message, you'd need to report some other signal.  Then
> > you get into issues about backwards binary compatibility...
> ...
> > All this is interesting, but it's not really a GCC issue.  Feel free 
> > to
> > lobby the FreeBSD or Darwin folks for a SIGSTKOVERFLOW or whatever.
> Changing which signal is sent would be nice in the long term, but that
> could indeed cause backwards compatibility problems, so I'm not 
> suggesting
> that.  Instead, I'm suggesting just changing the default behaviour when
> a SIGBUS signal is received and the current signal handler for SIGBUS
> is SIG_DFL.  The default behaviour should IMHO be to print a reasonable
> error message to stderr before terminating the process.
This is such a bad idea it's hard to start explaining why it will never 
happen.  Consider, for instance, what would happen if the user program 
is really a database server that has closed stdin/stdout/stderr and 
happens to have fd 2 open on its database backing store.

> Incidentally, why does stack overflow result in SIGBUS on Darwin rather
> than SIGSEGV?  Posix mandates that stack overflow result in SIGSEGV.
I'm not sure exactly what happens, but I'm pretty sure it starts with 
SIGSEGV being raised, then the kernel attempts to create a signal frame 
on the already-full stack.  You'll notice that host-darwin.c does catch 
SIGSEGV.

--
Geoff Keating <geoffk@apple.com>