[PATCH v2] dsound: Make GlobalHandle() for the sound buffer failed.

[Akihiro Sagawa]

In v2, remove an unreliable conformance test. It crashes on x64 Windows.

I think GlobalHandle() does the right thing here because it should succeed
if the buffer is allocated by GlobalAlloc(GMEM_FIXED) or HeapAlloc().
Therefore, dsound needs to be fixed.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=48635
Signed-off-by: Akihiro Sagawa <sagawa.aki(a)gmail.com>
---
 dlls/dsound/buffer.c | 16 +++++-----------
 1 file changed, 5 insertions(+), 11 deletions(-)

[Andrew Eikum]

On Wed, Feb 26, 2020 at 09:43:03PM +0900, Akihiro Sagawa wrote:
> In v2, remove an unreliable conformance test. It crashes on x64 Windows.

Would it be possible to add the test for 32-bit only? Without a test I
worry we could lose this change in a future patch.
> diff --git a/dlls/dsound/buffer.c b/dlls/dsound/buffer.c
> index 8d4a3b9..b019b0d 100644
> --- a/dlls/dsound/buffer.c
> +++ b/dlls/dsound/buffer.c
> @@ -1054,19 +1054,15 @@ HRESULT secondarybuffer_create(DirectSoundDevice *device, const DSBUFFERDESC *ds
>   TRACE("capf = 0x%08x, device->drvcaps.dwFlags = 0x%08x\n", capf, device->drvcaps.dwFlags);
>
> /* Allocate an empty buffer */
> -	dsb->buffer = HeapAlloc(GetProcessHeap(),0,sizeof(*(dsb->buffer)));
> +	dsb->buffer = HeapAlloc(GetProcessHeap(),0,sizeof(*(dsb->buffer)) + dsb->buflen);
>          if (!dsb->buffer) {
>                  IDirectSoundBuffer8_Release(&dsb->IDirectSoundBuffer8_iface);
>       return DSERR_OUTOFMEMORY;
>   }
>
> -	/* Allocate system memory for buffer */
> -	dsb->buffer->memory = HeapAlloc(GetProcessHeap(),0,dsb->buflen);
> -        if (!dsb->buffer->memory) {
> -		WARN("out of memory\n");
> -                IDirectSoundBuffer8_Release(&dsb->IDirectSoundBuffer8_iface);
> -		return DSERR_OUTOFMEMORY;
> -	}
> +	/* Assign subsequent memory for buffer
> +           to make GlobalHandle(dsb->buffer->memory) call fail */
> +	dsb->buffer->memory = (LPBYTE)dsb->buffer + sizeof(*(dsb->buffer));

I think we need to make sure memory is aligned -- e.g. round up the
size of dst->buffer to nearest alignment. Or, some other solution
might be more elegant; Zeb suggested VirtualAlloc might solve this?

Andrew