darling-libkqueue/TODO

 * There are very likely to be lock-related bugs stemming from when the
   kevent_copyin() and kevent_copyout() functions stopped holding the
   kqueue lock. It would be nice to have a lock testing mechanism to 
   check that the correct locks are held.. something like:

      #define MTX_LOCKED 1
      #define MTX_UNLOCKED 0
      #if NDEBUG
      #define LOCK_MONITOR(x)  int x
      #define lock_assert(x,y) assert((x) == (y))
      #define lock_acquire(x,y) do { pthread_mutex_lock((x)); y = MTX_LOCKED } while (0)
      #define lock_release(x,y) do { pthread_mutex_unlock((x)); y = MTX_UNLOCKED } while (0)
      #else
      #define LOCK_MONITOR(x)  /**/
      #define lock_assert(x,y) do {} while (0)
      #define lock_acquire(x,y) do { pthread_mutex_lock((x)); } while (0)
      #define lock_release(x,y) do { pthread_mutex_unlock((x)); } while (0)
      #endif

      struct foo {
         LOCK_MONITOR(foo_lockstat);
      }
      ...

      lock_assert(&kn->kn_lockmon, MTX_LOCKED);
      lock_assert(&kn->kn_lockmon, MTX_UNLOCKED);

 * Add a counter that increments on each each kevent() call. When printing
   debug information within kevent(), display the value of the counter.
   This will be helpful when debugging a multithreaded program that may have
   multiple kevent() calls executing in parallel.

 * Add a dbg_printf() statement within kevent_wait() to report the value
   of the timeout.

 * Refactor kevent_copyin() to remove the goto statements.

 * Fix the crasher w/ corruption in test/vnode.c

 * Add the kevent64() syscall as implemented on MacOS X. This guarantees
   that 64-bit values can be used in the 'udata' field on 32-bit platforms.

 * Check other filters for the EV_DISPATCH bug that was fixed in r252.