strafe-client/src/worker.rs

use std::thread;
use std::sync::{mpsc,Arc};
use parking_lot::Mutex;

//WorkerPool
struct Pool(u32);
enum PoolOrdering{
	Single,//single thread cannot get out of order
	Ordered(u32),//order matters and should be buffered/dropped according to ControlFlow
	Unordered(u32),//order does not matter
}
//WorkerInput
enum Input{
	//no input, workers have everything needed at creation
	None,
	//Immediate input to any available worker, dropped if they are overflowing (all workers are busy)
	Immediate,
	//Queued input is ordered, but serial jobs that mutate state (such as running physics) can only be done with a single worker
	Queued,//"Fifo"
	//Query a function to get next input when a thread becomes available
	//worker stops querying when Query function returns None and dies after all threads complete
	//lifetimes sound crazy on this one
	Query,
	//Queue of length one, the input is replaced if it is submitted twice before the current work finishes
	Mailbox,
}
//WorkerOutput
enum Output{
	None(Pool),
	Realtime(PoolOrdering),//outputs are dropped if they are out of order and order is demanded
	Buffered(PoolOrdering),//outputs are held back internally if they are out of order and order is demanded
}

//It would be possible to implement all variants
//with a query input function and callback output function but I'm not sure if that's worth it.
//Immediate = Condvar
//Queued = receiver.recv()
//a callback function would need to use an async runtime!

//realtime output is an arc mutex of the output value that is assigned every time a worker completes a job
//buffered output produces a receiver object that can be passed to the creation of another worker
//when ordering is requested, output is ordered by the order each thread is run
//which is the same as the order that the input data is processed except for Input::None which has no input data
//WorkerDescription
struct Description{
	input:Input,
	output:Output,
}

//The goal here is to have a worker thread that parks itself when it runs out of work.
//The worker thread publishes the result of its work back to the worker object for every item in the work queue.
//Previous values do not matter as soon as a new value is produced, which is why it's called "Realtime"
//The physics (target use case) knows when it has not changed the body, so not updating the value is also an option.

/*
QR = WorkerDescription{
	input:Queued,
	output:Realtime(Single),
}
*/
pub struct QRWorker<Task:Send,Value:Clone>{
	sender: mpsc::Sender<Task>,
	value:Arc<Mutex<Value>>,
}

impl<Task:Send+'static,Value:Clone+Send+'static> QRWorker<Task,Value>{
	pub fn new<F:FnMut(Task)->Value+Send+'static>(value:Value,mut f:F) -> Self {
		let (sender, receiver) = mpsc::channel::<Task>();
		let ret=Self {
			sender,
			value:Arc::new(Mutex::new(value)),
		};
		let value=ret.value.clone();
		thread::spawn(move || {
			loop {
				match receiver.recv() {
					Ok(task) => {
						let v=f(task);//make sure function is evaluated before lock is acquired
						*value.lock()=v;
					}
					Err(_) => {
						println!("Worker stopping.",);
						break;
					}
				}
			}
		});
		ret
	}

	pub fn send(&self,task:Task)->Result<(), mpsc::SendError<Task>>{
		self.sender.send(task)
	}

	pub fn grab_clone(&self)->Value{
		self.value.lock().clone()
	}
}

/*
QN = WorkerDescription{
	input:Queued,
	output:None(Single),
}
*/
//None Output Worker does all its work internally from the perspective of the work submitter
pub struct QNWorker<'a,Task:Send>{
	sender: mpsc::Sender<Task>,
	handle:thread::ScopedJoinHandle<'a,()>,
}

impl<'a,Task:Send+'a> QNWorker<'a,Task>{
	pub fn new<F:FnMut(Task)+Send+'a>(scope:&'a thread::Scope<'a,'_>,mut f:F)->QNWorker<'a,Task>{
		let (sender,receiver)=mpsc::channel::<Task>();
		let handle=scope.spawn(move ||{
			loop {
				match receiver.recv() {
					Ok(task)=>f(task),
					Err(_)=>{
						println!("Worker stopping.",);
						break;
					}
				}
			}
		});
		Self{
			sender,
			handle,
		}
	}
	pub fn send(&self,task:Task)->Result<(),mpsc::SendError<Task>>{
		self.sender.send(task)
	}
}

/*
IN = WorkerDescription{
	input:Immediate,
	output:None(Single),
}
*/
//Inputs are dropped if the worker is busy
pub struct INWorker<'a,Task:Send>{
	sender: mpsc::SyncSender<Task>,
	handle:thread::ScopedJoinHandle<'a,()>,
}

impl<'a,Task:Send+'a> INWorker<'a,Task>{
	pub fn new<F:FnMut(Task)+Send+'a>(scope:&'a thread::Scope<'a,'_>,mut f:F)->INWorker<'a,Task>{
		let (sender,receiver)=mpsc::sync_channel::<Task>(1);
		let handle=scope.spawn(move ||{
			loop {
				match receiver.recv() {
					Ok(task)=>f(task),
					Err(_)=>{
						println!("Worker stopping.",);
						break;
					}
				}
			}
		});
		Self{
			sender,
			handle,
		}
	}
	//blocking!
	pub fn blocking_send(&self,task:Task)->Result<(), mpsc::SendError<Task>>{
		self.sender.send(task)
	}
	pub fn send(&self,task:Task)->Result<(), mpsc::TrySendError<Task>>{
		self.sender.try_send(task)
	}
}

#[cfg(test)]
mod test{
	use super::{thread,QRWorker};
	use crate::physics;
	use strafesnet_common::{integer,instruction};
	#[test]//How to run this test with printing: cargo test --release -- --nocapture
	fn test_worker() {
		// Create the worker thread
		let test_body=physics::Body::new(integer::vec3::ONE,integer::vec3::ONE,integer::vec3::ONE,integer::Time::ZERO);
		let worker=QRWorker::new(physics::Body::ZERO,
			|_|physics::Body::new(integer::vec3::ONE,integer::vec3::ONE,integer::vec3::ONE,integer::Time::ZERO)
		);

		// Send tasks to the worker
		for _ in 0..5 {
			let task = instruction::TimedInstruction{
				time:integer::Time::ZERO,
				instruction:strafesnet_common::physics::Instruction::Idle,
			};
			worker.send(task).unwrap();
		}

		// Optional: Signal the worker to stop (in a real-world scenario)
		// sender.send("STOP".to_string()).unwrap();

		// Sleep to allow the worker thread to finish processing
		thread::sleep(std::time::Duration::from_millis(10));

		// Send a new task
		let task = instruction::TimedInstruction{
			time:integer::Time::ZERO,
			instruction:strafesnet_common::physics::Instruction::Idle,
		};
		worker.send(task).unwrap();

		//assert_eq!(test_body,worker.grab_clone());

		// wait long enough to see print from final task
		thread::sleep(std::time::Duration::from_millis(10));
	}
}