Update readme.md and remove some platforms from the GitHub workflow.

Author: user

.github/workflows/rust.yml

@@ -19,8 +19,8 @@ jobs:
       matrix:
         os: [
           ubuntu-latest, ubuntu-24.04, ubuntu-22.04, ubuntu-24.04-arm, ubuntu-22.04-arm,
-          macos-latest, macos-14, macos-15, macos-26, macos-13, macos-15-intel
-          windows-latest, windows-2025, windows-2022, windows-11-arm,
+          macos-latest, macos-14, macos-15, macos-26, macos-13
+          windows-2025, windows-2022, windows-11-arm,
         ]
     runs-on: ${{ matrix.os }}
 

readme.md

@@ -6,42 +6,39 @@ When implementing an AVL tree, there are some choices:
 * Storing the balance factor or height in a node.
 * Taking a recursive or iterative approach.
 
-This is an iterative implementation of an AVL tree that stores the height in a node and does not store the parent pointer in it. Key points:
+This is an iterative implementation of an AVL tree that stores the height in a node
+and does not store the parent pointer in it. Key points:
 * Written in pure Rust.
-* The key data type can be any type that has Ord.
+* The key type can be any type that has `Ord`.
 * Each node is identified by a key, and so there are no nodes with the same key.
 
 ## Description
 <picture>
   <source media="(prefers-color-scheme: dark)" srcset="img/AVLTreeDarkTheme.png">
   <source media="(prefers-color-scheme: light)" srcset="img/AVLTreeLightTheme.png">
-  <img src="img/AVLTreeDarkTheme.png" width="650">
+  <img src="img/AVLTreeDarkTheme.png" width="620">
 </picture>
 
 **A diagram of an AVL tree with a balance factor labeled below each node.**
 
-
 The AVL tree, named after its inventors, is a self-balancing binary search tree (BST)
 that has a worst-case time complexity of $O(log(n))$ on lookup, insertion, and deletion.
 
 An ordinary (unbalanced) BST has a worst-case time complexity of $O(n)$ for the basic operations.
-Without rebalancing, the tree height keeps increasing on insertion, and the performance degrades to that of a linked list, that is, linear time.
+Without rebalancing, the tree height keeps increasing on insertion,
+and the performance degrades to that of a linked list, that is, linear time.
 
 Therefore, rebalancing through tree rotations is necessary to keep the useful property of BSTs.
 
-Here is
-[an interactive tool to visualize how AVL trees work](https://www.cs.usfca.edu/~galles/visualization/AVLtree.html).
-
-## License
-All code, including the test module, is released under the MIT license.
-
-The diagram above is my own work and is released into the public domain (CC0).
+Here is an
+[interactive tool to visualize how AVL trees work](https://www.cs.usfca.edu/~galles/visualization/AVLtree.html).
 
 ## Definitions
 Below I provide the definitions used throughout this repository.
 
 ### Depth
-The depth of a node is the depth of its parent plus one. An empty node's depth is zero. A root node has no parent, thus its depth is $0 + 1 = 1$.[^1]
+The depth of a node is the depth of its parent plus one. An empty node's depth is zero.
+A root node has no parent, thus its depth is $0 + 1 = 1$.[^1]
 
 ### Height
 The height of a node $X$ is the maximum of the heights of its children plus one:
@@ -71,3 +68,21 @@ In the diagram above, the root node $R$ is left-heavy, since $BF(R) = Height(S)
 [^1]: This differs from the usual definition, where a root's depth is zero.
 [^2]: This differs from the usual definition, where a leaf's height is zero.
 [^3]: In the usual definition, it's three.
+
+## Implementation notes
+All primitive integer types have `Ord`, so they can be used as the key type. So does `String`, since it has `Ord`.
+
+IEEE 754 floating-point arithmetic has a special value called "Not a Number" (NaN),
+which is not equal to anything, not even to itself.
+Because NaN violates the [trichotomy](https://en.wikipedia.org/wiki/Law_of_trichotomy),
+floating-point types like `f32` cannot have `Ord`, so they cannot be used as the key type.
+
+A key-value pair is stored in the node itself.
+During a tree rotation, everything of the node is swapped with another (using `std::mem::swap`).
+Therefore, if the key or value were large, expensive memory copies might take place during insertion or deletion.
+
+## License
+All code, including the [test module](src/avl_test.rs), is released under the MIT license.
+
+The diagram above is my own work and is released into the public domain
+([CC0](https://creativecommons.org/publicdomain/zero/1.0/)).