Re: include: Always add defines for the Interlocked* functions.

[Jacek Caban]

Hi Alexandre,
> -static FORCEINLINE LONG WINAPI InterlockedIncrement( LONG volatile 
> *dest )
> -{
> -    return _InterlockedIncrement( (long volatile *)dest );
> -}

What was wrong with those inlines? Note that the commit removes LONG* to 
long* casts and causes warnings on clang msvc target.


Jacek

include: Always add defines for the Interlocked* functions.

[Alexandre Julliard]

Jacek Caban <jacek(a)codeweavers.com> writes:
> Hi Alexandre,
> > -static FORCEINLINE LONG WINAPI InterlockedIncrement( LONG volatile
> >    *dest )
> > -{
> > -    return _InterlockedIncrement( (long volatile *)dest );
> > -}
>
> What was wrong with those inlines? Note that the commit removes LONG*
> to long* casts and causes warnings on clang msvc target.

They need to be defines, for instance libxml2 does this:

#ifdef InterlockedCompareExchangePointer
        InterlockedCompareExchangePointer((void **) &global_init_lock,
                                          cs, NULL);
#else /* Use older void* version */
        InterlockedCompareExchange((void **) &global_init_lock,
                                   (void *) cs, NULL);
#endif /* InterlockedCompareExchangePointer */

which will break on 64-bit. We could of course change only that one, but
they are all #defines in the MS headers.

I think that a better fix would be to use the correct types for LONG
etc. on PE targets.

-- 
Alexandre Julliard
julliard(a)winehq.org

include: Always add defines for the Interlocked* functions.

[Jacek Caban]

On 10/15/21 8:57 AM, Alexandre Julliard wrote:
> Jacek Caban <jacek(a)codeweavers.com> writes:
> > Hi Alexandre,
> > > -static FORCEINLINE LONG WINAPI InterlockedIncrement( LONG volatile
> > >     *dest )
> > > -{
> > > -    return _InterlockedIncrement( (long volatile *)dest );
> > > -}
> >
> > What was wrong with those inlines? Note that the commit removes LONG*
> > to long* casts and causes warnings on clang msvc target.
> > They need to be defines, for instance libxml2 does this:
>
> #ifdef InterlockedCompareExchangePointer
>          InterlockedCompareExchangePointer((void **) &global_init_lock,
>                                            cs, NULL);
> #else /* Use older void* version */
>          InterlockedCompareExchange((void **) &global_init_lock,
>                                     (void *) cs, NULL);
> #endif /* InterlockedCompareExchangePointer */
>
> which will break on 64-bit. We could of course change only that one, but
> they are all #defines in the MS headers.
>
> I think that a better fix would be to use the correct types for LONG
> etc. on PE targets.

That sounds indeed great for the final solution, those differences pop 
up every now and then.


It's a lot of changes (although Michael didn't run away when I mentioned 
such possibility at WineConf :) ) and a lot of warnings until then. 
Another possible intermediate solution would be having inlines paired 
with no-op macros like:

#define InterlockedCompareExchangePointer InterlockedCompareExchangePointer


Thanks,

Jacek