Re: [AD] (still) a race condition in mixer.c

[Peter Wang <tjaden@xxxxxxxxxx>]

On 2003-09-23, Eric <ebotcazou@xxxxxxxxxx> wrote:
> > Now what to do on the 4.0 branch? Backporting the internal API means
> > modifying the system_driver object, which is part of the ABI through
> > several inline functions.
> 
> I think that modifying system_driver is too risky. So I can think of two 
> options:
> 
> (1) adding a system_driver clone, which would only have the 4 methods of the 
> synchronization API (with a kludge in order not to be exported by the DLL),
> 
> (2) applying Sam's patch as is, on the ground that all ports in the 4.0.x 
> series use pthreads except the Windows port, but the default sound driver of 
> that port doesn't use the mixer.
> 
> Which one seems to be the most sensible?

I don't like (2) at all.  (1) seems fine.

Tacking on methods to the back of system_driver seems like it could be
ok (though I agree that it's not the safest route).  What problems do
you anticipate, exactly?  That the compiler might reorder the system
driver structure?

-- 
王浩禎