Re: [AD] double signal hang fix

[Peter Wang <tjaden@xxxxxxxxxx>]

On 2003-06-23, Vincent <Vincent.Penquerch@xxxxxxxxxx> wrote:
> > I haven't looked at it properly, but I think it's the wrong way to fix
> > it.  It should be possible to stop signals from arriving 
> > while we're in
> > the middle of a signal handler, using sigaction() to install 
> > the handler
> 
> Ah, I wasn't aware of that difference.
> 
> > instead of signal().  I'd also prefer to know why two signals 
> > are being
> > raised in the first place.
> 
> In my case, I got a SIGSEGV, the handler is called, then calls
> allegro_exit, which raises a SIGSEGV as well (memory corruption).

I see.  To protect against this, I think the signal handler should not
trust allegro_exit() to do what it is supposed to.  The old signal
handlers (i.e. the ones that were present when allegro_init() was
called) should be restored by the signal handler itself, before it calls
allegro_exit().

PS. I don't know why signals are arriving during the execution of the
signal handler.  The man page says that the raised signal is blocked for
the duration, and I confirmed it on a small test case.  I also tried
blocking _all_ signals during the signal handler, by installing the
handler with `sigaction'.  But a second signal still arrives while the
first signal is being handled.

PPS. A test case for this is to simply make allegro_exit() segfault
somewhere, e.g. in a _add_exit_func() callback.  No need to hack the
Allegro source, as I was doing :-)

-- 
王浩禎