Re: [PATCH 2/5] gdi32: Implement EMFDRV_AlphaBlend().

[Zhiyi Zhang]

On 4/15/21 3:31 PM, Huw Davies wrote:
> On Wed, Apr 14, 2021 at 03:27:57PM +0800, Zhiyi Zhang wrote:
> > Fix a bug that Tally produces a blank print preview when images have to be scaled.
> > Signed-off-by: Zhiyi Zhang
> >
> > dlls/gdi32/enhmfdrv/bitblt.c         | 85 ++++++++++++++++++++++++++++
> >  dlls/gdi32/enhmfdrv/enhmetafiledrv.h |  2 +
> >  dlls/gdi32/enhmfdrv/init.c           |  2 +-
> >  dlls/gdi32/tests/metafile.c          |  4 +-
> >  4 files changed, 89 insertions(+), 4 deletions(-)
> >
> > diff --git a/dlls/gdi32/enhmfdrv/bitblt.c b/dlls/gdi32/enhmfdrv/bitblt.c
> > index 1fc5c0b1040..f0be2b52fc0 100644
> > --- a/dlls/gdi32/enhmfdrv/bitblt.c
> > +++ b/dlls/gdi32/enhmfdrv/bitblt.c
> > @@ -27,6 +27,91 @@
> >  #include "enhmetafiledrv.h"
> >  #include "wine/debug.h"
> >
> > +BOOL CDECL EMFDRV_AlphaBlend( PHYSDEV dev_dst, struct bitblt_coords *dst, PHYSDEV dev_src,
> > -
> >                          struct bitblt_coords *src, BLENDFUNCTION func )
> >
> > +{
> > - UINT bits_size, bmi_size, emr_size, size, bpp;
> > - BITMAPINFOHEADER *bmih;
> > - EMRALPHABLEND *emr;
> > - HBITMAP hbitmap;
> > - BITMAP bitmap;
> > - BOOL ret;
> > -
> > - /* can't use a metafile DC as source */
> > - if (dev_src->funcs == dev_dst->funcs)
> > -
> >    return FALSE;
> > -
> > - hbitmap = GetCurrentObject(dev_src->hdc, OBJ_BITMAP);
> > - if (GetObjectW(hbitmap, sizeof(BITMAP), &bitmap) != sizeof(BITMAP))
> > -
> >    return FALSE;
>
> This would likely be cleaner using the BlendImage entry point
> (likewise I suspect StretchBlt would benefit from using PutImage).
>
> Huw.

Hi Huw,

I am afraid that it's not cleaner in practice. There are a few issues that make it less intuitive if I use the BlendImage entry point.
1. EMR_ALPHABLEND needs source DC to get world transform and background color, so we need to add a src_dev parameter to all BlendImage function calls.
2. EMR_ALPHABLEND records always use BI_BITFIELDS compression when the source image is 16 bit or 32 bit. Something like the following in gdi32/dib.c#line 1367 in GetDIBits()
needs to be added to the EMR_BlendImage entry point. So we can't copy BITMAPINFOHEADER directly using memcpy().
> /* if the src and dst are the same depth, copy the colour info across */
>     if (dst_info->bmiHeader.biBitCount == src_info->bmiHeader.biBitCount && coloruse == DIB_RGB_COLORS )
>     {
>         switch (src_info->bmiHeader.biBitCount)
>         {
>         case 16:
>             if (src_info->bmiHeader.biCompression == BI_RGB)
>             {
>                 src_info->bmiHeader.biCompression = BI_BITFIELDS;
>                 memcpy( src_info->bmiColors, bit_fields_555, sizeof(bit_fields_555) );
>             }
>             break;
>         case 32:
>             if (src_info->bmiHeader.biCompression == BI_RGB)
>             {
>                 src_info->bmiHeader.biCompression = BI_BITFIELDS;
>                 memcpy( src_info->bmiColors, bit_fields_888, sizeof(bit_fields_888) );
>             }
>             break;
>         }
>         copy_color_info( dst_info, src_info, coloruse );
>     }

Also need to reset biClrUsed to 0 as gdi32/dib.c#line 1462 in GetDIBits() to pass all the tests.
> if (info->bmiHeader.biSize != sizeof(BITMAPCOREHEADER))
>     {
>         info->bmiHeader.biClrUsed = 0;
>         info->bmiHeader.biSizeImage = get_dib_image_size( info );
>     }

So it turns out messier than using EMFDRV_AlphaBlend() if we need to add these special handling for bitmaps in different format.
I think it's better to stick with EMFDRV_AlphaBlend().

Thanks,
Zhiyi