veloren/common/src/astar.rs
2024-02-05 19:51:50 +01:00

279 lines
9.5 KiB
Rust

use crate::path::Path;
use core::{
cmp::Ordering::{self, Equal},
fmt,
hash::{BuildHasher, Hash},
};
use hashbrown::HashMap;
use std::collections::BinaryHeap;
#[derive(Copy, Clone, Debug)]
pub struct PathEntry<S> {
// cost so far + heursitic
cost_estimate: f32,
node: S,
}
impl<S: Eq> PartialEq for PathEntry<S> {
#[allow(clippy::unconditional_recursion)] // false positive as we use .node
fn eq(&self, other: &PathEntry<S>) -> bool { self.node.eq(&other.node) }
}
impl<S: Eq> Eq for PathEntry<S> {}
impl<S: Eq> Ord for PathEntry<S> {
// This method implements reverse ordering, so that the lowest cost
// will be ordered first
fn cmp(&self, other: &PathEntry<S>) -> Ordering {
other
.cost_estimate
.partial_cmp(&self.cost_estimate)
.unwrap_or(Equal)
}
}
impl<S: Eq> PartialOrd for PathEntry<S> {
fn partial_cmp(&self, other: &PathEntry<S>) -> Option<Ordering> { Some(self.cmp(other)) }
// This is particularily hot in `BinaryHeap::pop`, so we provide this
// implementation.
//
// NOTE: This probably doesn't handle edge cases like `NaNs` in a consistent
// manner with `Ord`, but I don't think we need to care about that here(?)
//
// See note about reverse ordering above.
fn le(&self, other: &PathEntry<S>) -> bool { other.cost_estimate <= self.cost_estimate }
}
pub enum PathResult<T> {
/// No reachable nodes were satisfactory.
///
/// Contains path to node with the lowest heuristic value (out of the
/// explored nodes).
None(Path<T>),
/// Either max_iters or max_cost was reached.
///
/// Contains path to node with the lowest heuristic value (out of the
/// explored nodes).
Exhausted(Path<T>),
/// Path succefully found.
///
/// Second field is cost.
Path(Path<T>, f32),
Pending,
}
impl<T> PathResult<T> {
/// Returns `Some((path, cost))` if a path reaching the target was
/// successfully found.
pub fn into_path(self) -> Option<(Path<T>, f32)> {
match self {
PathResult::Path(path, cost) => Some((path, cost)),
_ => None,
}
}
pub fn map<U>(self, f: impl FnOnce(Path<T>) -> Path<U>) -> PathResult<U> {
match self {
PathResult::None(p) => PathResult::None(f(p)),
PathResult::Exhausted(p) => PathResult::Exhausted(f(p)),
PathResult::Path(p, cost) => PathResult::Path(f(p), cost),
PathResult::Pending => PathResult::Pending,
}
}
}
// If node entry exists, this was visited!
#[derive(Clone, Debug)]
struct NodeEntry<S> {
/// Previous node in the cheapest path (known so far) that goes from the
/// start to this node.
///
/// If `came_from == self` this is the start node! (to avoid inflating the
/// size with `Option`)
came_from: S,
/// Cost to reach this node from the start by following the cheapest path
/// known so far. This is the sum of the transition costs between all the
/// nodes on this path.
cost: f32,
}
#[derive(Clone)]
pub struct Astar<S, Hasher> {
iter: usize,
max_iters: usize,
max_cost: f32,
potential_nodes: BinaryHeap<PathEntry<S>>, // cost, node pairs
visited_nodes: HashMap<S, NodeEntry<S>, Hasher>,
/// Node with the lowest heuristic value so far.
///
/// (node, heuristic value)
closest_node: Option<(S, f32)>,
}
/// NOTE: Must manually derive since Hasher doesn't implement it.
impl<S: Clone + Eq + Hash + fmt::Debug, H: BuildHasher> fmt::Debug for Astar<S, H> {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.debug_struct("Astar")
.field("iter", &self.iter)
.field("max_iters", &self.max_iters)
.field("potential_nodes", &self.potential_nodes)
.field("visited_nodes", &self.visited_nodes)
.field("closest_node", &self.closest_node)
.finish()
}
}
impl<S: Clone + Eq + Hash, H: BuildHasher + Clone> Astar<S, H> {
pub fn new(max_iters: usize, start: S, hasher: H) -> Self {
Self {
max_iters,
max_cost: f32::MAX,
iter: 0,
potential_nodes: core::iter::once(PathEntry {
cost_estimate: 0.0,
node: start.clone(),
})
.collect(),
visited_nodes: {
let mut s = HashMap::with_capacity_and_hasher(1, hasher);
s.extend(core::iter::once((start.clone(), NodeEntry {
came_from: start,
cost: 0.0,
})));
s
},
closest_node: None,
}
}
pub fn with_max_cost(mut self, max_cost: f32) -> Self {
self.max_cost = max_cost;
self
}
pub fn poll<I>(
&mut self,
iters: usize,
// Estimate how far we are from the target? but we are given two nodes...
// (current, previous)
mut heuristic: impl FnMut(&S, &S) -> f32,
// get neighboring nodes
mut neighbors: impl FnMut(&S) -> I,
// have we reached target?
mut satisfied: impl FnMut(&S) -> bool,
) -> PathResult<S>
where
I: Iterator<Item = (S, f32)>, // (node, transition cost)
{
let iter_limit = self.max_iters.min(self.iter + iters);
while self.iter < iter_limit {
if let Some(PathEntry {
node,
cost_estimate,
}) = self.potential_nodes.pop()
{
let (node_cost, came_from) = self
.visited_nodes
.get(&node)
.map(|n| (n.cost, n.came_from.clone()))
.expect("All nodes in the queue should be included in visisted_nodes");
if satisfied(&node) {
return PathResult::Path(self.reconstruct_path_to(node), node_cost);
// Note, we assume that cost_estimate isn't an overestimation
// (i.e. that `heuristic` doesn't overestimate).
} else if cost_estimate > self.max_cost {
return PathResult::Exhausted(
self.closest_node
.clone()
.map(|(lc, _)| self.reconstruct_path_to(lc))
.unwrap_or_default(),
);
} else {
for (neighbor, transition_cost) in neighbors(&node) {
if neighbor == came_from {
continue;
}
let neighbor_cost = self
.visited_nodes
.get(&neighbor)
.map_or(f32::MAX, |n| n.cost);
// compute cost to traverse to each neighbor
let cost = node_cost + transition_cost;
if cost < neighbor_cost {
let previously_visited = self
.visited_nodes
.insert(neighbor.clone(), NodeEntry {
came_from: node.clone(),
cost,
})
.is_some();
let h = heuristic(&neighbor, &node);
// note that `cost` field does not include the heuristic
// priority queue does include heuristic
let cost_estimate = cost + h;
if self
.closest_node
.as_ref()
.map(|&(_, ch)| h < ch)
.unwrap_or(true)
{
self.closest_node = Some((node.clone(), h));
};
// TODO: I think the if here should be removed
// if we hadn't already visited, add this to potential nodes, what about
// its neighbors, wouldn't they need to be revisted???
if !previously_visited {
self.potential_nodes.push(PathEntry {
cost_estimate,
node: neighbor,
});
}
}
}
}
} else {
return PathResult::None(
self.closest_node
.clone()
.map(|(lc, _)| self.reconstruct_path_to(lc))
.unwrap_or_default(),
);
}
self.iter += 1
}
if self.iter >= self.max_iters {
PathResult::Exhausted(
self.closest_node
.clone()
.map(|(lc, _)| self.reconstruct_path_to(lc))
.unwrap_or_default(),
)
} else {
PathResult::Pending
}
}
fn reconstruct_path_to(&mut self, end: S) -> Path<S> {
let mut path = vec![end.clone()];
let mut cnode = &end;
while let Some(node) = self
.visited_nodes
.get(cnode)
.map(|n| &n.came_from)
.filter(|n| *n != cnode)
{
path.push(node.clone());
cnode = node;
}
path.into_iter().rev().collect()
}
}