After some consideration and washing off the feeling that I am wasting
my time, I have reworked my previous patch to test at boot time whether
GDI+ is available (gdiplus_v3.diff). It implements a generic
"native-image-api" option which can be extended to OSX (I have a patch,
but I have no machines to test it so it is not attached).

It does not implement a new image type. Instead, it places the native
backends at the front and continues searching the image list when those
backends are not active. I find that treating the native backend as a
second class citizen, as ImageMagick, is not ok. But most important, I
find that the abstraction of backend = image-type is broken.

Regarding previous discussions, I believe the concerns about GDI+
becoming deprecated are irrelevant in comparison to the fact that Emacs
is using already a deprecated API, namely GDI, which is older and
equally prone to disappear in favor of Direct2d/WIC.

Regarding unicode, GDI+ requires WCHAR. Therefore the backend refuses to
initialize if w32-unicode-filenames is NIL, defaulting to other
backends. However, if the user changes this variable from T to NIL after
the backend has been initialized, the backend will remain active but
refuse to load images.

I have fixed the use of terminal hooks, due to insistance. However, that
meant one function declaration had to be added to w32term.h

I have not changed the use of SSDATA in the :data field of images. If
this is a problem, it is so in all of image.c.

I have touch my Emacs installation to avoid removing the double spaces
after a dot, but I am not 100% sure all is Kosher.

I go to bed. If you still find that this patch is not acceptable, I am
not going to work any more on it. I developed it to minimize the number
of dependencies of Emacs on the Windows platform, but that does not seem
to be a priority around here.

Best,

-- 
Juan José García Ripoll
http://juanjose.garciaripoll.com
http://quinfog.hbar.es