#tree-traversal #graph-traversal #traversal #node-tree #stack #tree #cursor

no-std mutcursor

Safely stores mutable references to parent nodes, for backtracking during traversal of tree & graph structures

12 releases

0.2.2 Oct 8, 2024
0.2.1 Oct 7, 2024
0.1.9 Oct 6, 2024
0.1.8 Sep 27, 2024
0.1.1 Jun 28, 2024

#417 in Algorithms

MIT/Apache

39KB
697 lines

mutcursor

This crate provides types to safely store mutable references to parent nodes, for backtracking during traversal of tree & graph structures.

[MutCursor] is more efficient because it avoids dynamic allocation, while [MutCursorVec] provides for an arbitrarily deep stack.

[MutCursorRootedVec] supports mutable references to a separate root type and a different leaf type. In the future I may generalize this pattern to be more flexible.

Usage

use mutcursor::MutCursor;

let mut tree = TreeNode::new(5);
let mut node_stack = MutCursor::<TreeNode, 2>::new(&mut tree);

// Traverse to the last node
while node_stack.advance(|node| {
    node.traverse()
}) {}

assert_eq!(node_stack.top().val, 0);
assert_eq!(node_stack.depth(), 1);

node_stack.backtrack();
assert_eq!(node_stack.top().val, 1);
assert_eq!(node_stack.depth(), 0);

/// A simple stand-in for a recursive graph structure
struct TreeNode {
    val: usize,
    next: Option<Box<TreeNode>>
}
impl TreeNode {
    fn new(count: usize) -> Self {
        if count > 0 {
            Self {val: count, next: Some(Box::new(Self::new(count-1)))}
        } else {
            Self {val: 0, next: None}
        }
    }
    fn traverse(&mut self) -> Option<&mut Self> {
        self.next.as_mut().map(|boxed| &mut **boxed)
    }
}

Alternative(s)

This crate basically does the same thing as generic-cursors. However, there are several reasons to choose this crate:

  1. The fixed-size stack used by [MutCursor] has lower overhead than a Vec, and can be used in a no_std environment where dynamic allocation may be unavailable.

  2. The MutCursor::try_map_into_mut API enables some paterns that would be otherwise impossible.

Safety Thesis

Each &mut reference stored by a [MutCursor] mutably borrows the reference beneath it in the stack. The stack root takes a mutable (and therefore exclusive) borrow of the node itself. Therefore the stack's top is an exclusive borrow.

You can imagine unrolling tree traversal into something like the code below, but this isn't amenable to looping. In essence each level variable is preserved, but inaccessible because the level above is mutably borrowing it. The [MutCursor] object contains all the level variables but only provides access to the top

let level_1 = &mut root;
{
    let level_2 = level_1.traverse().unwrap();
    {
        match level_2.traverse() {
            Some(level_3) => {} // Do something with level_3
            None => {} // Fall back to work level_2 or level_1
        }
    }
}

Future Work

Macro to define cursor types

In the current design of [MutCursorRootedVec], there is a predefined pattern prescribing where RootT and NodeT types may exist on the stack. However, we may find it necessary in the future to support more than two types, in a more flexible pattern. It seems that a macro to define a bespoke cursor type is the best solutuion.

Internal enum for multiple-type support at runtime

For ultimate flexibility, we would want all the references to be stored by the stack as in an enum over the possible reference types. However, if ther user provided an enum as a type parameter to a cursor type, the result result would be double-indirection. Therefore the enum behavior would need to be internal to the MutCursor. Deriving a MutCursor from a user's enum type feels like a friendly way to define the types a cursor type is capable of storing.

No runtime deps

Features