I fixed the following bugs:
* enable-event & disable-event do not check if the event is already enabled/disabled. This can lead to enabling the idle callback more than once for a given window which can break the event loop in various ways. For instance, in current cl-opengl, rb-double enables idle once per mouseclick and this leads to multiple entires in *windows-with-idle-event* which in turn means that the idle method will get called multiple times breaking the logic.
* Currently the idle callback is disabled only in close, and only when *windows-with-idle-event* is nil. We need to do the same check in disable-event too. Currently this manifests in gears.lisp. If you minimize the window, (disable-event :idle) will get called but the callback will continue to run -> 100% cpu usage.
Now for darwin, i fixed all the issues i could find. It now works on sbcl/ccl/ccl64 with no issues. Instead of forcing the glut event loop on the initial thread, i used the trick from ccl opengl-ffi on sbcl too.
This means that, on sbcl and ccl/ccl64, a new thread is created (GLUT-thread) which will run the glut event loop. This also has the bonus of making cl-glut work without any changes under slime. The only issue when developing interactively is that one can not call glut functions from a thread other than GLUT-thread _while_ the glut event loop is running. I added warnings to indicate when the event loop runs/stops. If the glut event loop is not running (example: leave-main-loop is called), you can call glut functions from any thread and they will take effect when the glut event loop starts running again. This shouldn't affect normal development, i had to mention it because osx has the tendency to kill the process when one ignores this :-)
(All the changes apart from the bug fixes are only for darwin.)