If you enjoy using this crate, please star the github repo.
tree_iterators_rs is a library built to provide you with the utilities (iterators) to easily work with tree data structures in Rust. It provides a default implementation for the majority use cases. These include:
- TreeNode - This struct has a Vec-based list of children. documentation
- BinaryTreeNode - This struct contains an optional boxed reference to a left and right node. documentation
This crate is written such that you can build your own implementations as well using other collection types. The collection type simply need to implement the Iterator trait.
This crate only contains one feature flag:
- "serde" - this flag can be used to implement Serialize and Deserialize for TreeNode and BinaryTreeNode.
This crate uses no unsafe code!
The largest benefit of using this library is that the various tree iterators are interchangeable in the for and while loop syntax. This means you can write your code to use a breadth-first-search today and trivially swap it out for a depth-first pre or postorder search tomorrow without significantly changing the structure of your code. Syntactically, this library is also able to hide a lot of complexity and provide a friendly interface.
The other benefit of this library is its tight integration with the rest of Rust's iterators. Because each API returns an iterator, you can proceed to use the iterator APIs like filter, map, and reduce on your trees which can very powerfully simplify code.
The easiest way to get started is to simply add this crate as a dependency and
add a using statement to pull in its prelude
(use tree_iterators_rs::prelude::*;). You can then create your data structure
using the BinaryTreeNode or TreeNode structs provided and use their iterator
implementations. These structs provide default implementations of all
functionality available in this crate. The methods that come attached to these
structs include the following (all of which have open source code that can be
found in
this file):
- Owned Iterator APIs - these take ownership of the TreeNode similarly to an
into_iter() call.
- bfs()
- dfs_preorder()
- dfs_postorder()
- dfs_inorder() - this iterator is only available on BinaryTreeNode
- Mutable borrow APIs - these do not take ownership and behave similarly to an
iter_mut() call.
- bfs_iter_mut()
- dfs_preorder_iter_mut()
- dfs_postorder_iter_mut()
- dfs_inorder_iter_mut - this iterator is only available on BinaryTreeNode
- Borrow APIs - these do not take ownership and behave similarly to an iter()
call.
- bfs_iter()
- dfs_preorder_iter()
- dfs_postorder_iter()
- dfs_inorder_iter() - this iterator is only available on BinaryTreeNode
- In addition, all of these APIs have additional modifier methods which include .leaves() and attach_ancestors(). Each of these can be chained after the iterator call (more details in Examples).
2.0.0 - This version removes the Option wrapping of children iterators. This wrapping is redundant since iterators can represent empty collections in and of themselves. Removing it will improve performance going forward. All iterator methods (bfs(), dfs_preorder(), dfs_postorder(), etc.) can be used just the same as in previous versions.
For all of the following examples, we will use the tree structure as follows. It is complicated enough to convey the tree iterators' behavior thoroughly without being overwhelming. Code to build this tree using both the BinaryTreeNode and TreeNode structs is provided below.
flowchart TD
A[0] --> B{1}
A --> C{2}
B --> D{3}
B --> E{4}
D --> Z{None}
D --> Y{None}
E --> X{None}
E --> W{None}
C --> F{5}
F --> V{None}
F --> U{None}
C --> G{6}
G --> H{7}
G --> T{None}
H --> S{None}
H --> I{8}
I --> J{9}
I --> R{None}
J --> Q{None}
J --> K{10}
K --> P{None}
K --> O{None}
Using BinaryTreeNode
use tree_iterators_rs::prelude::*;
pub fn create_example_binary_tree() -> BinaryTreeNode<usize> {
BinaryTreeNode {
value: 0,
left: Some(Box::new(BinaryTreeNode {
value: 1,
left: Some(Box::new(BinaryTreeNode {
value: 3,
left: None,
right: None,
})),
right: Some(Box::new(BinaryTreeNode {
value: 4,
left: None,
right: None,
})),
})),
right: Some(Box::new(BinaryTreeNode {
value: 2,
left: Some(Box::new(BinaryTreeNode {
value: 5,
left: None,
right: None,
})),
right: Some(Box::new(BinaryTreeNode {
value: 6,
left: Some(Box::new(BinaryTreeNode {
value: 7,
left: None,
right: Some(Box::new(BinaryTreeNode {
value: 8,
left: Some(Box::new(BinaryTreeNode {
value: 9,
left: None,
right: Some(Box::new(BinaryTreeNode {
value: 10,
left: None,
right: None,
})),
})),
right: None,
})),
})),
right: None,
})),
})),
}
}Using TreeNode
use tree_iterators_rs::prelude::*;
pub fn create_example_tree() -> TreeNode<usize> {
use std::vec::Vec;
TreeNode {
value: 0,
children: vec![
TreeNode {
value: 1,
children: vec![
TreeNode {
value: 3,
children: Vec::with_capacity(0),
},
TreeNode {
value: 4,
children: Vec::with_capacity(0),
},
],
},
TreeNode {
value: 2,
children: vec![
TreeNode {
value: 5,
children: Vec::with_capacity(0),
},
TreeNode {
value: 6,
children: vec![TreeNode {
value: 7,
children: vec![TreeNode {
value: 8,
children: vec![TreeNode {
value: 9,
children: vec![TreeNode {
value: 10,
children: Vec::with_capacity(0),
}],
}],
}],
}],
},
],
},
],
}
}All 3 APIs' usages are identical other than their borrowing model, so only one example will be given. Say we want to join all of the values in our test tree into a string. We can do this as follows:
use tree_iterators_rs::{
examples::create_example_tree,
prelude::*
};
// Tree creation (see above documentation)
let root = create_example_tree();
let mut result = String::new();
for value in root.bfs() {
result.push_str(&value.to_string());
result.push_str(", ");
}
// result: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10,
println!("{}", result);This code could also be written using Rust's iterator APIs:
use tree_iterators_rs::{
examples::create_example_tree,
prelude::*
};
// Tree creation (see above documentation)
let root = create_example_tree();
let result =
root.bfs()
.map(|val| val.to_string())
.collect::<Vec<String>>()
.join(", ");
// result: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10
println!("{}", result);An equivalent set of code not using this crate would look like this:
use tree_iterators_rs::examples::create_example_tree;
use std::collections::VecDeque;
// Tree creation (see above documentation)
let root = create_example_tree();
let mut result = String::new();
let mut queue = VecDeque::new();
queue.push_back(root);
while queue.len() > 0 {
if let Some(front) = queue.pop_front() {
for child in front.children {
queue.push_back(child);
}
result.push_str(&front.value.to_string());
result.push_str(", ");
}
}
// result: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10,
println!("{}", result);Similarly to the BFS example, all 3 APIs' usages are identical other than their borrowing model, so only one example will be given. Say we want to join all of the values in our test tree into a string. We can do this as follows:
use tree_iterators_rs::{
examples::create_example_tree,
prelude::*
};
// Tree creation (see above documentation)
let root = create_example_tree();
let mut result = String::new();
for value in root.dfs_preorder() {
result.push_str(&value.to_string());
result.push_str(", ");
}
// result: 0, 1, 3, 4, 2, 5, 6, 7, 8, 9, 10,
println!("{}", result);This code could also be written using Rust's iterator APIs:
use tree_iterators_rs::{
examples::create_example_tree,
prelude::*
};
// Tree creation (see above documentation)
let root = create_example_tree();
let result =
root.dfs_preorder()
.map(|val| val.to_string())
.collect::<Vec<String>>()
.join(", ");
// result: 0, 1, 3, 4, 2, 5, 6, 7, 8, 9, 10
println!("{}", result);An equivalent set of code not using this crate would look like this:
use tree_iterators_rs::examples::create_example_tree;
// Tree creation (see above documentation)
let root = create_example_tree();
let mut result = String::new();
let mut stack = vec![root];
while stack.len() > 0 {
if let Some(mut top) = stack.pop() {
top.children.reverse();
for child in top.children {
stack.push(child);
}
result.push_str(&top.value.to_string());
result.push_str(", ");
}
}
// result: 0, 1, 3, 4, 2, 5, 6, 7, 8, 9, 10,
println!("{}", result);Similarly to the other examples, all 3 APIs' usages are identical other than their borrowing model, so only one example will be given. Say we want to join all of the values in our test tree into a string. We can do this as follows:
use tree_iterators_rs::{
examples::create_example_binary_tree,
prelude::*
};
// Tree creation (see above documentation)
let root = create_example_binary_tree();
let mut result = String::new();
for value in root.dfs_inorder() {
result.push_str(&value.to_string());
result.push_str(", ");
}
// result: 3, 1, 4, 0, 5, 2, 7, 9, 10, 8, 6,
println!("{}", result);This code could also be written using Rust's iterator APIs:
use tree_iterators_rs::{
examples::create_example_binary_tree,
prelude::*
};
// Tree creation (see above documentation)
let root = create_example_binary_tree();
let result = root.dfs_preorder()
.map(|val| val.to_string())
.collect::<Vec<String>>()
.join(", ");
// result: 3, 1, 4, 0, 5, 2, 7, 9, 10, 8, 6
println!("{}", result);An equivalent set of code not using this crate would look like the following. It is important to note that the dfs_inorder API does not use recursion and thus does not incur the overhead of stack frames that this example does. For purposes of simplicity in the example, recursion was chosen. Rust's ownership model is difficult to work around with this traversal type and would make the example more complex than necessary.
use tree_iterators_rs::{
examples::create_example_binary_tree,
prelude::*
};
fn dfs_inorder(node: Option<Box<BinaryTreeNode<usize>>>, result: &mut String) {
match node {
None => {}
Some(node) => {
dfs_inorder(node.left, result);
result.push_str(&node.value.to_string());
result.push_str(", ");
dfs_inorder(node.right, result)
}
}
}
// Tree creation (see above documentation)
let root = create_example_binary_tree();
let mut result = String::new();
dfs_inorder(Some(Box::new(root)), &mut result);
// result: 3, 1, 4, 0, 5, 2, 7, 9, 10, 8, 6,
println!("{}", result);Similarly to the BFS and other DFS examples, all 3 APIs' usages are identical other than their borrowing model, so only one example will be given. Say we want to join all of the values in our test tree into a string. We can do this as follows:
use tree_iterators_rs::{
examples::create_example_tree,
prelude::*
};
// Tree creation (see above documentation)
let root = create_example_tree();
let mut result = String::new();
for value in root.dfs_postorder() {
result.push_str(&value.to_string());
result.push_str(", ");
}
// result: 3, 4, 1, 5, 10, 9, 8, 7, 6, 2, 0,
println!("{}", result);This code could also be written using Rust's iterator APIs:
use tree_iterators_rs::{
examples::create_example_tree,
prelude::*
};
// Tree creation (see above documentation)
let root = create_example_tree();
let result =
root.dfs_postorder()
.map(|val| val.to_string())
.collect::<Vec<String>>()
.join(", ");
// result: 3, 4, 1, 5, 10, 9, 8, 7, 6, 2, 0
println!("{}", result);An equivalent set of code not using this crate would look like the following. It is important to note that the dfs_postorder API does not use recursion and thus does not incur the overhead of stack frames that this example does. For purposes of simplicity in the example, recursion was chosen. Rust's ownership model is difficult to work around with this traversal type and would make the example more complex than necessary.
use tree_iterators_rs::{
examples::create_example_tree,
prelude::*
};
fn dfs_postorder(node: TreeNode<usize>, result: &mut String) {
for child in node.children {
dfs_postorder(child, result);
}
result.push_str(", ");
result.push_str(&node.value.to_string());
}
// Tree creation (see above documentation)
let root = create_example_tree();
let mut result = String::new();
dfs_postorder(root, &mut result);
// result: 3, 4, 1, 5, 10, 9, 8, 7, 6, 2, 0,
println!("{}", result);attach_ancestors() is a method that can be called after any of the above APIs to change the iterator structure into one that returns a slice of all ancestors and the current value in the tree. If one of these is called, the (now streaming) iterator will yield a slice where the item at index 0 is the root value, the item at index len() - 1 is the current value, and everything in between is the other ancestors. As an example, when we are at the value of 10 in our traversal (see above documentation), the slice will look like this: [0, 2, 6, 7, 8, 9, 10].
For example, we can use this API to filter down to only the values where all of the ancestors and the current node are even numbers in the example tree.
NOTE: Be sure to add a use statement for streaming_iterator::StreamingIterator to pull in the filter, map, reduce, for_each, etc. methods from the streaming_iterator crate.
use streaming_iterator::StreamingIterator;
use tree_iterators_rs::{
examples::create_example_tree,
prelude::*
};
let root = create_example_tree();
let mut result = String::new();
root.dfs_preorder_iter()
.attach_ancestors()
.filter(|slice|
slice.iter().all(|value| **value % 2 == 0)
)
.map(|slice| slice[slice.len() - 1])
.for_each(|value| {
result.push(' ');
result.push_str(&value.to_string())
});
// result: 0 2 6
println!("{}", result);We can do the same with the in order iterator if we use a binary tree:
use streaming_iterator::StreamingIterator;
use tree_iterators_rs::{
examples::create_example_binary_tree,
prelude::*
};
let root = create_example_binary_tree();
let mut result = String::new();
root.dfs_inorder_iter()
.attach_ancestors()
.filter(|slice|
slice.iter().all(|value| **value % 2 == 0)
)
.map(|slice| slice[slice.len() - 1])
.for_each(|value| {
result.push(' ');
result.push_str(&value.to_string())
});
// result: 0 2 6
println!("{}", result);We can do the same with the postorder iterator to get the result in reverse:
use streaming_iterator::StreamingIterator;
use tree_iterators_rs::{
examples::create_example_tree,
prelude::*
};
let root = create_example_tree();
let mut result = String::new();
root.dfs_postorder_iter()
.attach_ancestors()
.filter(|slice|
slice.iter().all(|value| **value % 2 == 0)
)
.map(|slice| slice[slice.len() - 1])
.for_each(|value| {
result.push(' ');
result.push_str(&value.to_string())
});
// result: 6 2 0
println!("{}", result);And we can do the same with a breadth-first search. This just so happens to yield the same result as the depth first preorder iterator:
use streaming_iterator::StreamingIterator;
use tree_iterators_rs::{
examples::create_example_tree,
prelude::*
};
let root = create_example_tree();
let mut result = String::new();
root.bfs_iter()
.attach_ancestors()
.filter(|slice|
slice.iter().all(|value| **value % 2 == 0)
)
.map(|slice| slice[slice.len() - 1])
.for_each(|value| {
result.push(' ');
result.push_str(&value.to_string())
});
// result: 0 2 6
println!("{}", result);leaves() is a method that can be called after any of the above APIs (including attach_ancestors() - see here) to change the iterator structure to one that only returns leaves of the tree. In the example tree (see above documentation), this will always result in the sequence 3, 4, 5, 10. Once this method is called, the iterator transforms to be either a breadth-first (if the iterator was previously breadth-first) or a depth first postorder search (if the iterator was previously one of pre-, in-, or post-order depth first searches).
I will be using the depth first preorder search in the examples, but this works with all of the traversal types. This method can be called immediately if you wish to only receive the leaves of the tree like so:
use tree_iterators_rs::{
examples::create_example_tree,
prelude::*
};
let root = create_example_tree();
let result = root.dfs_preorder()
.leaves()
.map(|val| val.to_string())
.collect::<Vec<String>>()
.join(", ");
// result: 3, 4, 5, 10
println!("{}", result);Alternatively, this method can be used to perform a normal traversal and then switch to a leaves-only traversal partway through the regular one. This can be done like so (again, all traversal types support this):
use tree_iterators_rs::{
examples::create_example_tree,
prelude::*
};
let root = create_example_tree();
let mut dfs_preorder = root.dfs_preorder();
let mut results = Vec::new();
// take the first 2 non-leaves before switching to a leaves-only iterator
results.push(dfs_preorder.next().unwrap().to_string());
results.push(dfs_preorder.next().unwrap().to_string());
// once leaves is called, iteration switches to a depth-first postorder search
for leaf in dfs_preorder.leaves() {
results.push(leaf.to_string());
}
let result = results.join(", ");
// result: 0, 1, 3, 4, 5, 10,
println!("{}", result);This crates' APIs are powered by 3 traits. The traits include:
- OwnedTreeNode/OwnedBinaryTreeNode - these traits powers the bfs(), dfs_preorder(), dfs_inorder() and dfs_postorder() APIs.
- MutBorrowedTreeNode/MutBorrowedBinaryTreeNode - these traits powers the bfs_iter_mut(), dfs_preorder_iter_mut(), dfs_inorder_iter_mut and dfs_postorder() APIs.
- BorrowedTreeNode/BorrowedBinaryTreeNode - these traits powers the bfs_iter(), dfs_preorder_iter(), dfs_inorder_iter() and dfs_postorder_iter() APIs.
You may pick and choose which of the traits you implement to suit your needs. They are identical apart from some naming conventions and the ownership of the value returned from their required method implementation.
As an example, we can implement another TreeNode variant that uses a Linked List to hold its children. All code here is MIT licensed and you are free to directly copy or modify it as you see fit subject to the conditions of the license.
use tree_iterators_rs::prelude::*;
use std::collections::LinkedList;
struct LLTreeNode<T> {
value: T,
children: LinkedList<LLTreeNode<T>>
}This is a good start, but realistically we want it to implement all of the tree traversal APIs available to us.
We can start with the OwnedTreeNode implementation as follows. Since we chose to use a LinkedList as our children property, we have to wrap its iterator in a Some() Option variant. The hardest part of implementing this is figuring out the type of LinkedList's into_iter() method return type.
use tree_iterators_rs::prelude::*;
use std::collections::{
LinkedList,
linked_list::IntoIter
};
struct LLTreeNode<T> {
value: T,
children: LinkedList<LLTreeNode<T>>
}
impl<T> OwnedTreeNode for LLTreeNode<T> {
type OwnedValue = T;
type OwnedChildren = IntoIter<LLTreeNode<T>>;
fn get_value_and_children(self) -> (Self::OwnedValue, Self::OwnedChildren) {
(
self.value,
self.children.into_iter()
)
}
}Now that we have implemented OwnedTreeNode, our type has the bfs(), dfs_preorder(), and dfs_postorder() methods.
The mutable borrow implementation is very similar to the owned one. The only difference is that the 'Value' associated type changed to a mutable reference and we are calling iter_mut() instead of into_iter().
use tree_iterators_rs::prelude::*;
use std::collections::{
LinkedList,
linked_list::IterMut
};
struct LLTreeNode<T> {
value: T,
children: LinkedList<LLTreeNode<T>>
}
impl<'a, T> MutBorrowedTreeNode<'a> for LLTreeNode<T>
where Self: 'a {
type MutBorrowedValue = &'a mut T;
type MutBorrowedChildren = IterMut<'a, LLTreeNode<T>>;
fn get_value_and_children_iter_mut(&'a mut self) -> (Self::MutBorrowedValue, Self::MutBorrowedChildren) {
(
&mut self.value,
self.children.iter_mut()
)
}
}Now that we have implemented MutBorrowedTreeNode, our type has the bfs_iter_mut(), dfs_preorder_iter_mut(), and dfs_postorder_iter_mut() methods.
The borrow implementation is also very similar to the owned one. The only difference is that the 'Value' associated type changed to an immutable reference and we are calling iter() instead of into_iter().
use tree_iterators_rs::prelude::*;
use std::collections::{
LinkedList,
linked_list::Iter
};
struct LLTreeNode<T> {
value: T,
children: LinkedList<LLTreeNode<T>>
}
impl<'a, T> BorrowedTreeNode<'a> for LLTreeNode<T>
where Self: 'a {
type BorrowedValue = &'a T;
type BorrowedChildren = Iter<'a, LLTreeNode<T>>;
fn get_value_and_children_iter(&'a self) -> (Self::BorrowedValue, Self::BorrowedChildren) {
(
&self.value,
self.children.iter()
)
}
}Now that we have implemented BorrowedTreeNode, our type has the bfs_iter(), dfs_preorder_iter(), and dfs_postorder_iter() methods.
This library is purposely very loose in its trait declarations, so it opens the door to some pretty exotic data transformations and use cases. Each iterator is as lazy as possible and only calls the get_value_and_children method variant once per node in the tree. This means you are able to lazily build/generate the tree as traversal progresses if you choose to. This opens the door to using this library to do some crazy things. I would love to hear about any use cases like this that you come up with. Please leave a note/issue on the github repo!