[PATCH 9/9] d2d1: Partially implement ID2D1Bitmap1

[Nikolay Sivov]

On 11/23/2017 02:08 AM, Lucian Poston wrote:

> On Tue, Nov 21, 2017 at 5:52 AM, Nikolay Sivov <bunglehead at gmail.com> wrote:
>> On 21.11.2017 11:41, Lucian Poston wrote:
>>> @@ -69,6 +93,10 @@ static ULONG STDMETHODCALLTYPE d2d_bitmap_Release(ID2D1Bitmap1 *iface)
>> ...
>>> +            ID2D1ColorContext_Release(bitmap->color_context);
>>> +        if (bitmap->surface)
>>> +            IDXGISurface_Release(bitmap->surface);
>> ...
>>> +
>>> +    if (This->surface)
>>> +    {
>>> +        IDXGISurface_AddRef(This->surface);
>>> +    }
>>> +
>> Is it really possible that bitmap has no surface?
> Yes. Even prior to this patch, it could be initialized with a d3d
> device and no surface. This patch series adds a code path to create a
> bitmap with a surface and no device, namely
> CreateBitmapFromDxgiSurface.
I still don't see this. We have d2d_bitmap_create() that always creates 
a texture, and you can query a surface from it.
And we have d2d_bitmap_create_shared(), that is meant to share surface 
passed to it, or a surface used to initialize existing d2d bitmap.
>
>>> +static D2D1_BITMAP_PROPERTIES1 bitmap_properties_to_properties1(
>>> +        const D2D1_BITMAP_PROPERTIES *desc)
>>> +{
>>> +    D2D1_BITMAP_PROPERTIES1 d;
>>> +    d.bitmapOptions = D2D1_BITMAP_OPTIONS_NONE;
>>> +    d.colorContext = NULL;
>>> +    if (desc == NULL)
>>> +    {
>>> +        d.pixelFormat.format = DXGI_FORMAT_UNKNOWN;
>>> +        d.pixelFormat.alphaMode = D2D1_ALPHA_MODE_UNKNOWN;
>>> +        d.dpiX = 96.0f;
>>> +        d.dpiY = 96.0f;
>>> +    }
>>> +    else
>>> +    {
>>> +        d.pixelFormat.format = desc->pixelFormat.format;
>>> +        d.pixelFormat.alphaMode = desc->pixelFormat.alphaMode;
>>> +        d.dpiX = desc->dpiX;
>>> +        d.dpiY = desc->dpiY;
>>> +    }
>>> +
>>> +    return d;
>>> +}
>>> +
>> Maybe it's better to switch to PROPERTIES1 internally instead of
>> converting down like that.
> We have to convert from PROPERTIES to PROPERTIES1 at some point since
> some input parameters to the older public APIs are PROPERTIES. Are you
> saying the conversion should occur earlier, like immediately in the
> older public APIs?
>
>>> +    if (FAILED(hr = ID3D10Texture2D_QueryInterface(texture, &IID_IDXGISurface, (void **)&surface)))
>>> +    {
>>> +        surface = NULL;
>>> +        WARN("Texture2D had no underlying DXGISurface");
>>> +    }
>>> +
>> When does this fail? I mean it probably makes sense to have a warning,
>> but continue without a surface?
> The Texture2D implementation has code that is conditional on whether
> it has an internal surface object, including its QueryInterface
> implementation, so I wrote the bitmap code with that Texture2D
> assumption. If that's an invalid assumption, I can change this.
Texture2D supports it conditionally, yes. But textures are always 
created internally in d2d, in a way that surfaces are supported.
So if QI fails, it should be fatal failure, that's how existing code 
handles that already.