Enclosed is a patch that addresses the following:

--Various minor cleanups.
Removed dead/obsolete code, made some style cleanups

--Mouse Events
Now keep track of what button(s) were pressed so we know when to send
the mouse up event. This fixes the case where the mouse is dragged
outside of the game window and released (in which case we want to send
the mouse up event even though the mouse is outside the game window).

--Input Grabbing
Here is my take on the grabbing situation, which is the basis for the
new implementation.

There are 3 grab states, ungrabbed (UG), visible (VG), and invisible
(IG). Both VG and IG keep the mouse constrained to the window and
produce relative motion events. In VG the cursor is visible (duh), in
IG it is not. In VG, absolute motion events also work.

There are 6 actions that can affect grabbing:

1. Set Fullscreen/Window (F/W). In fullscreen, a visible grab should do
nothing. However, a fullscreen visible grab can be treated just like a
windowed visible grab, which is what I have done to help simplify
things.

2. Cursor hide/show (H/S). If the cursor is hidden when grabbing, the
grab is an invisible grab. If the cursor is visible, the grab should
just constrain the mouse to the window.

3. Input grab/ungrab(G/U). If grabbed, the cursor should be confined to
the window as should the keyboard input. On Mac OS X, the keyboard
input is implicitly grabbed by confining the cursor, except for
command-tab which can switch away from the application. Should the
window come to the foreground if the application is deactivated and
grab input is called? This isn't necessary in this implementation
because the grab state will be asserted upon activation.

Using my notation, these are all the cases that need to be handled
(state + action = new state).

UG+U = UG
UG+G = VG or IG, if cursor is visible or not
UG+H = UG
UG+S = UG

VG+U = UG
VG+G = VG
VG+H = IG
VG+S = VG

IG+U = UG
IG+G = IG
IG+H = IG
IG+S = VG

The cases that result in the same state can be ignored in the code,
which cuts it down to just 5 cases.

Another issue is what happens when the app loses/gains input focus from
deactivate/activate or iconify/deiconify. I think that if input focus
is ever lost (outside of SDL's control), the grab state should be
suspended and the cursor should become visible and active again. When
regained, the cursor should reappear in its original location and/or
grab state. This way, when reactivating the cursor is still in the same
position as before so apps shouldn't get confused when the next motion
event comes in. This is what I've done in this patch.