scheduling:
- cursor mover created a new java.util.timer for every movement and never
cancelled it, leaking a live thread roughly every 10 seconds; timers are
replaced by a single shared scheduledexecutorservice with daemon threads,
shut down when the automation stops.
clicks:
- the double click sequence used thread.sleep() inside a timer task, blocking
the scheduler thread; it is now scheduled as one-shot actions on the shared
scheduler, with the same pacing as before.
- stopping the automation now also cancels in-flight movements and pending
clicks, instead of letting them run to completion.
movement math:
- when the destination shared one coordinate with the starting position, the
step on the other axis was computed as length/0 = infinity, and the stop
condition checked one axis with || instead of &&, so a movement could stop
(and click) before reaching its destination. steps are now guarded and the
movement stops only when both axes have converged.
ui and lifecycle:
- the app could not be closed: closing the window only hid it, and ESC
unregistered the hotkey without exiting. the window now exits cleanly on
close, stopping the automation and unregistering the native hook.
- toggleRunning() was called from the jnativehook thread and touched swing
components off the EDT; UI updates are now marshalled to the EDT.
- the hotkey thrashed: key repeats re-fired the toggle while ctrl+alt was
held, and any other key pressed with the combo held toggled it too. it now
fires exactly once, when the combo is completed.
away-detection (CursorMoveListener) stays intentionally unscheduled: it cannot
tell the user's movements apart from the app's own, and will be reworked in a
future iteration.
bea