Function send_mouse 

pub unsafe fn send_mouse(
    window: *mut c_void,
    kind: u64,
    x: f64,
    y: f64,
    clicks: i64,
)

Hand a synthetic mouse event straight to window, bypassing the application’s event queue.

post_mouse enqueues on NSApp and needs the run loop to dequeue it; sendEvent: is the method NSApp would itself call once it had, so the event arrives without waiting on the queue.

This is not a way around needing focus, and neither is anything else. With the process inactive and its window on screen but not key — measured, the window was isVisible: true and occlusion-visible on a real screen — a click was delivered by none of: this, NSApp sendEvent:, NSApp postEvent:, mouseDown: at the hit-tested view, or any of those with acceptsFirstMouse: installed on the hosting view’s class at run time. Nor by changing what the window is: an NSPanel, and a nonactivatingPanel — which becomes key without activating its application, and is the obvious thing to reach for here — both stayed non-key and delivered nothing, with and without a makeKeyWindow or an activate.

What decides it is CGSSessionScreenIsLocked. A locked login session has no active application and cannot be given one, so every route above is closed at the same place; unlock it and all of them work. That is the point of the lock rather than a gap, so code here should ask whether the session is locked and say so, not try to route around it.

Below AppKit is worse, not better: Quartz Event Services (CGEventPostToPid, CGEventPost) is gated on the Accessibility permission, and an unbundled binary reports AXIsProcessTrusted() == false, so both of those deliver nothing even when the application is frontmost. See examples/activation_probe.rs, which measures each route side by side.

Safety

window must be a live NSWindow.