Mouse leave event... again (Windows specific)

Hi list,

(Longish post...)

A welcome addition to the output-pane input-model mechanism, in my opinion, 
would be the ability to capture the 'mouse leave' event (there has been some 
previous discussion of this). Up until recently I have been implementing this 
via a timer, which is scheduled when the mouse enters the pane, checking 
whether the mouse has left and unscheduling itself when it has.

This scheme has not proved altogether satisfactory, however, so I decided to 
see if I couldn't somehow intercept the Windows WM_MOUSELEAVE message, which is 
(as the name suggests) specifically designed for this purpose. This requires 
accessing the output-pane's underlying window procedure, which can be done via 
a well-established technique known as 'subclassing'. This is not subclassing in 
the familiar OO sense, but a means by which a new, user-defined window 
procedure can be substituted for the old one. All you need is the pane's 
handle, and a few Windows constants and procedures, such as SetWindowLongPtr. 
The address of the old procedure is retained so that it can be called to handle 
any messages not intercepted by the new one (or the same ones, if required). So 
I define a new window procedure as a lisp foreign-callable, and use it to 
intercept WM_MOUSELEAVE. It's also necessary to call TrackMouseEvent for this 
to work, which I do in my motion callback (once, on mouse entry).

This scheme *seems* to work OK. At least, I haven't experienced any noticeable 
issues with it in the few weeks I have been using it. Subclassing can also be 
used to implement 'owner drawn' controls, and in fact I have experimented 
lightly with this, creating (Windows look&feel purists stop reading now!) a 
menubar with a gradient background and a box highlight instead of the standard 
filled rectangle. I don't use any internal CAPI functions for this, it's all 
Windows FLI stuff (the handle of the 'top level' menu in an interface can be 
retrieved via GetMenu). I don't have any plans to introduce this into my 
application, but I wonder how far one could take it? Could this be dangerous in 
any way, maybe conflicting with low-level CAPI stuff? As I say, subclassing in 
this sense is a well-established technique. It's discussed at length on the 
MSDN page on window procedures - in fact it forms the bulk of that article:

http://msdn.microsoft.com/en-us/library/ms633569(v=VS.85).aspx

I have code I would be more than willing to share, if anybody is interested.

Best wishes,
Chris