+1  A: 

Why don't you use LWA_ALPHA and build the shadow into the image?

Edit in reponse to your comment:

A) Doesn't stop you using an alpha channeled PNG for a shadow only. Blt the 2 images together and use as one single image.
B) Its not hard to generate a drop shadow. In the image you posted its black with 3 different alpha values.
C) But it doesn't work does it? ie Time to get creative.
D) Then don't try and get windows to do something it won't do for you.
E) Is entirely irrelevant. Layered windows handle that for you.

I strongly recommend you learn more about layered windows because they CAN help you to your goal.

Edit2: You have the bitmap. Its fairly easy to scan over the image and find which bits will be colour keyed (by identifying the black yourself) and hen modify that to have an alpha of 0 where everything else will have an alpha of 255 (Not: You may have to convert the image to a 32-bit image from a lower colour format, you can do this by creating a new DC and copying the image). This will give you the same effect with LWA_ALPHA as with LWA_COLORKEY. From there its fairly easy to identify the pixel at the edge, where the color changes to (R = 0, G = 0, B = 0, A = 0). You then change that first pixel to have a n alpha of 192, the one blow it to 128 and the one below to 64. You now have an alpha'd gradation below the image that will look like the shadow. You can adjust the alpha to get the effect you want.

Goz
a) i don't control the image; it comes from the current enabled visual theme b) i don't have an algorithm to generate a drop-shadow handy c) Windows already provides a mechanism to add a drop-shadow d) i'm trying to make it a visual clone of what Windows does. e) Sampling the pixels behind my window, in order to perform the alpha-blend, is not only potentially bad (because pixels can change), it is wrong - with the advent of desktop composition, and will cause things to run slow as the compositor generates me a picture of the desktop.
Ian Boyd
Why do you need to duplicate Windows' functionality instead of using what's already there?
Koro
Because it can't do what you want.
Goz
What image shall i use for the alpha-channeled PNG? i don't know the shape, or size, of the window after its transparency has been applied. And, yes (c) it doesn't work. Which is why i thought i'd ask a community of developers who might see what i've missed.
Ian Boyd
+1  A: 

CS_DROPSHADOW only works with standard rectangular windows. WS_EX_LAYERED windows don't support everything. They are more of a low-level, self-service method to draw exactly what you want.

To get a drop shadow, you'll have to generate the drop-shadow yourself from the alpha channel in the image.

Adrian McCarthy
Any idea how Microsoft is managing to draw a drop-shadow outside a window?
Ian Boyd
+1, and Accepted answer. Turns out Microsoft is creating the properly sized drop shadow not with a layered window, but by using a region. A call to `GetWindowRgn` on the tooltips hWnd returns the rounded rectangle region. So CS_DropShadow doesn't support layered windows.
Ian Boyd
+1  A: 

Transparent layered windows is documented as the preferred technique over regions.

However, CS_DROPSHADOW does pay attention to regions. If you crop or otherwise shape your window using a region, the drop-shadow will follow the new outline.

Fortunately, you can use regions with layered windows, and by combining the two get the effect you're looking for.

BTW: tooltips_class32 does use CS_DROPSHADOW - you won't see it in the window styles because it's a class style, not a window style.

Shog9
i *can* use regions with layered window? Since the two both cause transparency, but (as you say) only regions affect the window's true shape - is there any reason to use a layered window? (And if so, what? And if not, why does the tooltip use it?)
Ian Boyd
i will poke around with the window handle of a tooltip window, seeing if i can call the equivalent of `GetWindowRgn` and see if it has one. That should answer that part, although not why it's layered in the first place.
Ian Boyd
They can both be used to give a window a visible "shape", but they do very, very different things: the region effectively clips the window, excluding some areas completely while leaving what remains to be drawn normally; layered windows are still rectangular, but are drawn into an off-screen bitmap and composited (there's a hack in place to let you continue using the normal WM_PAINT drawing method, but the output is redirected). In addition to simple color-key transparency, layered windows allow for translucency and even per-pixel alpha values (if you handle drawing directly).
Shog9
@Ian: As for tool windows, they're probably layered for performance. You noted the lack of `CS_SAVEBITS` - this style is unnecessary for layered windows, since Windows has to draw and preserve the windows underneath them in order to composite the layered window on top (this can actually hurt performance if a very *large* layered window is used, although that may no longer be as much of an issue if the DWM is compositing the entire desktop anyway).
Shog9