Re: winex11: Fix alpha blending in X11DRV_UpdateLayeredWindow.

[Alexandre Julliard]

Sebastian Lackner <sebastian(a)fds-team.de> writes:
> @@ -2627,14 +2627,12 @@ BOOL CDECL X11DRV_UpdateLayeredWindow( HWND hwnd, const UPDATELAYEREDWINDOWINFO
>      {
>          IntersectRect( &rect, &rect, info->prcDirty );
>          memcpy( src_bits, dst_bits, bmi->bmiHeader.biSizeImage );
> -        PatBlt( hdc, rect.left, rect.top, rect.right - rect.left, rect.bottom - rect.top, BLACKNESS );
> -    }
> -    ret = GdiAlphaBlend( hdc, rect.left, rect.top, rect.right - rect.left, rect.bottom - rect.top,
> -                         info->hdcSrc,
> -                         rect.left + (info->pptSrc ? info->pptSrc->x : 0),
> -                         rect.top + (info->pptSrc ? info->pptSrc->y : 0),
> -                         rect.right - rect.left, rect.bottom - rect.top,
> -                         (info->dwFlags & ULW_ALPHA) ? *info->pblend : blend );
> +    }
> +    ret = BitBlt( hdc, rect.left, rect.top, rect.right - rect.left, rect.bottom - rect.top,
> +                  info->hdcSrc,
> +                  rect.left + (info->pptSrc ? info->pptSrc->x : 0),
> +                  rect.top + (info->pptSrc ? info->pptSrc->y : 0),
> +                  SRCCOPY );

BitBlt is not guaranteed to set a valid alpha channel.

-- 
Alexandre Julliard
julliard(a)winehq.org

winex11: Fix alpha blending in X11DRV_UpdateLayeredWindow.

[Sebastian Lackner]

On 07.02.2017 17:08, Alexandre Julliard wrote:
> Sebastian Lackner <sebastian(a)fds-team.de> writes:
> > @@ -2627,14 +2627,12 @@ BOOL CDECL X11DRV_UpdateLayeredWindow( HWND hwnd, const UPDATELAYEREDWINDOWINFO
> >      {
> >          IntersectRect( &rect, &rect, info->prcDirty );
> >          memcpy( src_bits, dst_bits, bmi->bmiHeader.biSizeImage );
> > -        PatBlt( hdc, rect.left, rect.top, rect.right - rect.left, rect.bottom - rect.top, BLACKNESS );
> > -    }
> > -    ret = GdiAlphaBlend( hdc, rect.left, rect.top, rect.right - rect.left, rect.bottom - rect.top,
> > -                         info->hdcSrc,
> > -                         rect.left + (info->pptSrc ? info->pptSrc->x : 0),
> > -                         rect.top + (info->pptSrc ? info->pptSrc->y : 0),
> > -                         rect.right - rect.left, rect.bottom - rect.top,
> > -                         (info->dwFlags & ULW_ALPHA) ? *info->pblend : blend );
> > +    }
> > +    ret = BitBlt( hdc, rect.left, rect.top, rect.right - rect.left, rect.bottom - rect.top,
> > +                  info->hdcSrc,
> > +                  rect.left + (info->pptSrc ? info->pptSrc->x : 0),
> > +                  rect.top + (info->pptSrc ? info->pptSrc->y : 0),
> > +                  SRCCOPY );
>
> BitBlt is not guaranteed to set a valid alpha channel.

Could you please clarify in which situation this code will not work as expected?
I have checked the relevant code, but so far I have not found any reason why it wouldn't work here.
Also, it works perfectly fine with the test app written to demonstrate this bug
(which also uses alpha channels). Am I missing something?

Regards,
Sebastian

winex11: Fix alpha blending in X11DRV_UpdateLayeredWindow.

[Alexandre Julliard]

Sebastian Lackner <sebastian(a)fds-team.de> writes:
> On 07.02.2017 17:08, Alexandre Julliard wrote:
> > Sebastian Lackner <sebastian(a)fds-team.de> writes:
> > > @@ -2627,14 +2627,12 @@ BOOL CDECL X11DRV_UpdateLayeredWindow( HWND hwnd, const UPDATELAYEREDWINDOWINFO
> > >      {
> > >          IntersectRect( &rect, &rect, info->prcDirty );
> > >          memcpy( src_bits, dst_bits, bmi->bmiHeader.biSizeImage );
> > > -        PatBlt( hdc, rect.left, rect.top, rect.right - rect.left, rect.bottom - rect.top, BLACKNESS );
> > > -    }
> > > -    ret = GdiAlphaBlend( hdc, rect.left, rect.top, rect.right - rect.left, rect.bottom - rect.top,
> > > -                         info->hdcSrc,
> > > -                         rect.left + (info->pptSrc ? info->pptSrc->x : 0),
> > > -                         rect.top + (info->pptSrc ? info->pptSrc->y : 0),
> > > -                         rect.right - rect.left, rect.bottom - rect.top,
> > > -                         (info->dwFlags & ULW_ALPHA) ? *info->pblend : blend );
> > > +    }
> > > +    ret = BitBlt( hdc, rect.left, rect.top, rect.right - rect.left, rect.bottom - rect.top,
> > > +                  info->hdcSrc,
> > > +                  rect.left + (info->pptSrc ? info->pptSrc->x : 0),
> > > +                  rect.top + (info->pptSrc ? info->pptSrc->y : 0),
> > > +                  SRCCOPY );
> >
> > BitBlt is not guaranteed to set a valid alpha channel.
>
> Could you please clarify in which situation this code will not work as expected?
> I have checked the relevant code, but so far I have not found any reason why it wouldn't work here.
> Also, it works perfectly fine with the test app written to demonstrate this bug
> (which also uses alpha channels). Am I missing something?

If the source doesn't have alpha, it's assumed to be all 0xff.  In that
case, GdiAlphaBlend will set the alpha correctly in the destination, but
BitBlt won't.

-- 
Alexandre Julliard
julliard(a)winehq.org