# ARCHITECTURE: Design & Implementation Details Understand why this library is designed the way it is, and how it works internally. **[Back to README](../README.md) • [See Performance](./PERFORMANCE.md) • [Real Examples](./GUIDES.md)** --- ## Table of Contents 1. [Design Philosophy](#design-philosophy) 2. [The Big Three Pain Points](#the-big-three-pain-points) 3. [Why Deque is 484x Faster](#why-deque-is-484x-faster) 4. [Iterator Protocol Design](#iterator-protocol-design) 5. [Self-Balancing Strategy](#self-balancing-strategy) 6. [V8 JIT Optimizations](#v8-jit-optimizations) --- ## Design Philosophy ### Core Principles 1. **Practicality**: Follows ES6 standards, ESNext patterns, simple method names 2. **Extensibility**: OOP inheritance for all data structures 3. **Modularization**: Independent NPM packages, no bloat 4. **Efficiency**: Time/space complexity comparable to native JS 5. **Maintainability**: Open-source standards, TDD, CI/CD 6. **Testability**: Automated unit + performance testing 7. **Reusability**: Fully decoupled, minimized side effects 8. **Security**: Read-write separation, safe member access ### API Design: Unified Interface Instead of learning different APIs for each library: ```javascript // ❌ Different libraries, different APIs const queue = new Queue(); queue.enqueue(1); // Library A queue.push(1); // Library B queue.offer(1); // Library C (Java style) // ✅ data-structure-typed: Consistent everywhere const queue = new Queue(); const deque = new Deque(); const stack = new Stack(); // All use THE SAME 4 methods queue.push(item); // Add deque.unshift(item); // Add to front stack.pop(); // Remove queue.shift(); // Remove from front ``` **Why?** Because developers already know `push`, `pop`, `shift`, `unshift` from Arrays. Zero new mental models needed. --- ## The Big Three Pain Points ### Pain Point 1: Performance Wall When operations become bottlenecks: ```javascript const games = [ { id: 'game_001', name: 'Tetris', score: 9850 }, { id: 'game_002', name: 'Minecraft', score: 9750 }, { id: 'game_003', name: 'Grand Theft Auto V', score: 9600 }, { id: 'game_004', name: 'Wii Sports', score: 9500 }, { id: 'game_005', name: 'PlayerUnknown\'s Battlegrounds', score: 9200 }, { id: 'game_006', name: 'Fortnite', score: 9100 }, { id: 'game_007', name: 'League of Legends', score: 8950 }, { id: 'game_008', name: 'The Legend of Zelda: Breath of the Wild', score: 8850 }, { id: 'game_009', name: 'Elden Ring', score: 8700 }, { id: 'game_010', name: 'Super Mario Bros', score: 8600 }, ]; // ❌ Repeated sorting slows everything const scores = []; for (const game of games) { scores.push(game.score); scores.sort((a, b) => b - a); // O(n log n) every time! } ``` ```javascript // ✅ Auto-maintained sorted order const rankings = new RedBlackTree(games, { toEntryFn: (game) => [game.id, game.score], comparator: (a, b) => b - a, }); // O(n log n) in total! // Iteration is always sorted! ``` **Real Impact:** - Competitive programming: TLE → AC (Time Exceeded to Accepted) - Real-time systems: P99 latency 500ms → 5ms - Message queues: 100 msg/sec → 10,000 msg/sec ### Pain Point 2: API Chaos Different libraries use completely different method names: | Operation | ArrayList | Queue | ArrayDeque | LinkedList | |--------------|-----------|--------------|---------------|---------------| | Add end | add() | offer() | push() | add() | | Remove end | remove() | - | pop() | removeLast() | | Remove start | remove(0) | poll() | removeFirst() | removeFirst() | | Add start | add(0, e) | offerFirst() | unshift() | addFirst() | **Result**: Developers must memorize N different APIs. **Solution**: Use 4 consistent methods everywhere: ```typescript // Every linear structure uses same API const deque = new Deque(); const queue = new Queue(); const list = new DoublyLinkedList(); // All support these 4 methods: structure.push(item); // Add to end structure.pop(); // Remove from end structure.shift(); // Remove from start structure.unshift(item); // Add to start ``` ### Pain Point 3: Conversion Hell Bouncing data between structures wastes cycles: ```typescript // ❌ Painful way: conversions everywhere const tree = new TreeLibrary(data); const filtered = tree.toArray() .filter(x => x > 5) // Convert to Array .map(x => x * 2) // Still Array .sort((a, b) => b - a); // Array sort is O(n log n) // Lost the tree's sorted benefits! ``` ```typescript // ✅ Clean way: operations work directly const tree = new RedBlackTree(); const result = tree .filter(v => (v ?? 0) > 5) // Direct on tree .map((v, k) => [k, (v ?? 0) * 2]) // Still on tree .reduce((sum, v) => sum + (v ?? 0), 0); // Still on tree // Zero conversions, tree structure maintained! ``` --- ## Why Deque is 484x Faster ### The Problem: Array.shift() ```javascript // What happens when you shift from an array: const arr = [1, 2, 3, 4, 5]; arr.shift(); // Remove 1 // JavaScript engine must: // 1. Create new memory for index 0 // 2. Copy element 2 → index 0 // 3. Copy element 3 → index 1 // 4. Copy element 4 → index 2 // 5. Copy element 5 → index 3 // 6. Resize array // For 100,000 elements: O(n) = 100,000 operations each time! ``` Real benchmark: ```javascript const queue = []; for (let i = 0; i < 100000; i++) queue.push(i); for (let i = 0; i < 100000; i++) queue.shift(); // Time: 2829ms for 100,000 shifts ❌ ``` ### The Solution: Deque's Chunked Rings ```javascript // Deque implementation strategy: // Instead of physical array, use chunked buckets // bucket[0] bucket[1] bucket[2] // [1,2,3,4,5] [6,7,8,9,10] [11,12,13] // ^ // head pointer // When you shift(): // Move pointer forward in bucket // No copying until bucket is empty! // Only then delete bucket (batch operation) ``` Benchmark result: ```javascript const deque = new Deque(); for (let i = 0; i < 100000; i++) deque.push(i); for (let i = 0; i < 100000; i++) deque.shift(); // Time: 5.83ms for 100,000 shifts ✅ // 2829ms / 5.83ms = 485x faster! ``` ### Why This Works **Batching Operations**: Instead of reindexing on every shift, Deque batches operations into chunks. Only when an entire chunk is consumed does it get cleaned up. **Memory Locality**: Chunked structure is better for CPU cache than scattered reindexing operations. **Pointer Movement**: Shifting is just moving a pointer forward in memory, which is a CPU register operation (nanoseconds). --- ## Iterator Protocol Design ### The Hidden Superpower Every data structure implements JavaScript's iterator protocol: ```javascript // Why this matters: const tree = new RedBlackTree([5, 2, 8]); // All of these work automatically: [...tree] // Spread operator for (const x of tree) { } // for...of loop const [a, b, c] = tree // Destructuring new Set(tree) // Set constructor Array.from(tree) // Array.from // No special methods needed! // Just implement Symbol.iterator ``` ### Implementation Strategy ```typescript class CustomStructure { private items: T[] = []; // One method makes everything work * [Symbol.iterator]() { for (const item of this.items) { yield item; } } push(...item: T[]) { this.items.concat(item); } } // Now the structure works everywhere: const struct = new CustomStructure(); struct.push(1, 2, 3); // All of these work automatically: [...struct]; // [1, 2, 3] for (const x of struct) { } // Loops 1, 2, 3 ``` ### Why Iterator Protocol? 1. **Consistency**: Uses same interface as Arrays and Maps 2. **Zero Learning Curve**: Developers already know `for...of` 3. **Interoperability**: Works with spread, destructuring, Set constructor 4. **Future-Proof**: JavaScript standard, not library-specific --- ## Self-Balancing Strategy ### The Problem: Unbalanced Trees ```javascript // Create an unbalanced tree: const bst = new BST(); [1, 2, 3, 4, 5].forEach(x => bst.add(x)); // 1 // \ // 2 // \ // 3 // \ // 4 // \ // 5 // This becomes a linked list! O(n) instead of O(log n) ``` ### Red-Black Tree Solution **Rules:** 1. Every node is either Red or Black 2. Root is always Black 3. Red nodes have Black children (no consecutive Reds) 4. All paths to null have same number of Black nodes **Result**: Tree height limited to ~2 * log(n), guaranteeing O(log n) ```javascript const rbTree = new RedBlackTree([1, 2, 3, 4, 5]); // 3(B) Balanced! // / \ // 2(B) 4(R) // / \ // 1(R) 5(B) // Even with sequential inserts: O(log n) guaranteed! ``` ### AVL Tree Alternative **Stricter balance requirement**: Height difference ≤ 1 ```javascript // 3 // / \ // 2 4 // / \ // 1 5 // More strictly balanced = better search // Trade-off: Slower insertions/deletions (more rebalancing) ``` ### SkipList: Probabilistic Balancing Unlike Red-Black Tree or AVL Tree which rebalance deterministically, **SkipList** uses randomization: ``` Level 3: ──────────────────────────► 50 Level 2: ────────► 20 ──────────────► 50 Level 1: ──► 10 ──► 20 ──► 30 ──► 50 ──► 70 ``` Each node is promoted to a higher level with probability 0.5 (configurable). This gives O(log n) **average** performance — equivalent to Red-Black Tree in practice, but implemented with much simpler code. **When to use SkipList vs TreeMap:** - Both provide identical APIs (`NavigableMap` semantics) - SkipList: simpler codebase, constant factors slightly higher - TreeMap (Red-Black Tree): guaranteed O(log n) worst-case --- ## Range Query Structures ### SegmentTree Stores aggregated values in a complete binary tree backed by a flat array: ``` Array: [1, 3, 5, 7] Tree: 16 (root = sum all) / \ 4 12 (sum of halves) / \ / \ 1 3 5 7 (leaves = original values) Internal array: [_, 16, 4, 12, 1, 3, 5, 7] (1-indexed) ``` Query `[1,2]` (3+5=8): combines nodes covering exactly that range in O(log n). Supports any associative operation via `merger` function: sum, min, max, GCD, product, etc. ### BinaryIndexedTree (Fenwick Tree) A more compact structure for prefix sums only: ``` Array: [1, 3, 5, 7, 9] BIT tree (1-indexed): T[1] = 1 (covers [1,1]) T[2] = 1+3 = 4 (covers [1,2]) T[3] = 5 (covers [3,3]) T[4] = 1+3+5+7=16 (covers [1,4]) T[5] = 9 (covers [5,5]) ``` Prefix sum uses bit tricks (`i -= i & -i`) to traverse only O(log n) nodes. **SegmentTree vs BinaryIndexedTree:** | | SegmentTree | BinaryIndexedTree | |---|---|---| | Operations | sum/min/max/any | sum only | | Update | O(log n) | O(log n) | | Query | O(log n) | O(log n) | | Space | O(2n) | O(n) | | Complexity | Higher | Lower | --- ## V8 JIT Optimizations ### How V8 Makes This Fast ```javascript // V8 JIT optimizes data structures that: // 1. Have predictable shapes (hidden classes) // 2. Use consistent types // 3. Have stable method calls // ✅ Good for V8 JIT: class Node { constructor(value) { this.value = value; // Always same type this.left = null; // Always null or Node this.right = null; // Always null or Node } } ``` ```javascript // ❌ Bad for V8 JIT: class BadNode { constructor(value) { this.value = value; this.left = value; // Sometimes Node, sometimes number this.meta = null; // Added dynamically later } } // Our library uses strict typing for this reason! ``` ### Performance Benefit ```javascript // Predictable structure → V8 caches optimizations // First call: Interpreted // Subsequent calls: JIT compiled to native code // Result: typically faster after warm-up (benchmarks should include warm-up runs) const tree = new RedBlackTree(); for (let i = 0; i < 1000000; i++) { tree.set(i, Math.random()); } // Becomes near-native speed through JIT! ``` --- ## Method Chaining Architecture ### Design Pattern ```typescript class TreeStructure { // Methods return `this` for chaining filter(predicate: (v: V, k: K) => boolean): this { // ... filter logic ... return this; // Return structure, not Array! } map(mapper: (v: V, k: K) => V): this { // ... map logic ... return this; // Chain continues! } reduce(reducer: (acc: A, v: V, k: K) => A, init: A): A { // ... reduce logic ... return result; // Terminal operation } } // Result: Chainable on any structure tree .filter(x => x > 5) .map(x => x * 2) .reduce((a, v) => a + v, 0); ``` ### Why This Matters Traditional approach loses structure type: ```javascript // ❌ Loses type information const result = tree.toArray() // Now it's Array[] .filter(x => x > 5) // Still Array .map(x => x * 2); // Still Array // Lost O(log n) properties! ``` ```javascript // ✅ Preserves structure const result = tree .filter(x => x > 5) // Still RedBlackTree .map(x => x * 2) // Still RedBlackTree .reduce((a, v) => a + v); // Properties maintained! ``` --- ## Memory Efficiency ### Comparison | Structure | Overhead | Notes | |------------|----------|---------------------| | Array | Low | Fixed size | | LinkedList | High | Pointer per node | | BST | Medium | 2 pointers per node | | Deque | Very Low | Chunked, batched | | Heap | Very Low | Array-based | ### Deque Memory Strategy ```javascript // Instead of one big array: // [ ][ ][ ][ ][ ][ ][ ][ ][ ][ ] // Use chunks (buckets): // bucket[0]: [1][2][3][4][5] // bucket[1]: [6][7][8][9][10] // Benefits: // 1. Only allocate chunks needed // 2. Reuse empty buckets // 3. Better cache locality // 4. Less fragmentation ``` --- ## Type Safety Architecture ### Generic Type Parameters ```typescript // Custom object types interface User { id: number; name: string; age: number; } const userTree = new RedBlackTree(); userTree.set(1, { id: 1, name: 'Alice', age: 30 }); // Type inference works: const user = userTree.get(1); // Type: User | undefined user?.name; // Type-safe access ``` ### Comparator Custom Logic ```typescript type CustomObject = { name: string; priority: number; }; // Default comparator (ascending) const ascTree = new RedBlackTree(); // Reversed comparator (descending) const descTree = new RedBlackTree([], { comparator: (a, b) => b - a // Reverse comparison }); // Works with any type const objectTree = new RedBlackTree([], { comparator: (a, b) => a.priority - b.priority, }); ``` --- ## Order-Statistic Tree Architecture ### How `enableOrderStatistic` Works When enabled, every node maintains a `_count` field — the number of nodes in its subtree (including itself). ``` 8 (count=5) / \ 3 10 (count=1) (count=3) / \ 1 6 (count=1) (count=1) ``` ### Count Maintenance Counts are updated in every mutation path: - **Insert** (`set`/`add`): `_updateCountAlongPath` increments counts from new node to root - **Delete**: `_updateCountAlongPath` decrements counts from deleted position to root - **Rotations** (AVL: `_balanceLL/LR/RR/RL`, RBT: `_leftRotate/_rightRotate`): `_updateCount` recalculates after structural changes - **Balanced rebuild** (`setMany`, `perfectlyBalance`): counts rebuilt during tree construction ### Rank Operations All O(log n) by walking down the tree using counts: - **`getByRank(k)`**: Start at root. If left subtree count > k, go left. If equal, return current. Otherwise subtract and go right. - **`getRank(key)`**: Walk to key, accumulating left subtree counts. Returns count of elements preceding key in tree order. - **`rangeByRank(start, end)`**: Combine `getByRank` to find boundaries, then in-order collect. ### Opt-in Design Order-statistic is **opt-in** (`enableOrderStatistic: true`) because: - Extra count maintenance on every mutation adds O(log n) overhead - Most users don't need rank queries - `_snapshotOptions` preserves the flag through `clone()` and `map()` --- ## Error Handling: `raise()` ### Unified Error Strategy All error throwing goes through `raise(ErrorType, message)` in `src/common/error.ts`: ```typescript raise(TypeError, ERR.comparatorRequired('TreeMap')); raise(RangeError, ERR.indexOutOfBounds(index, length)); ``` ### Why `raise()` Instead of Direct `throw` 1. **Single chokepoint** — all errors flow through one function for consistent behavior 2. **Error message templates** — `ERR` object provides standardized, reusable messages 3. **Future extensibility** — can add logging, telemetry, or error transformation without touching call sites ### Error Categories | Error Type | When | Example | |------------|------|---------| | `TypeError` | Invalid input type | NaN key, missing comparator, non-function callback | | `RangeError` | Out of bounds | Index beyond array length, invalid rank | --- ## Summary: Design Checklist - ✅ Unified API across all structures - ✅ Iterator protocol implementation - ✅ Method chaining architecture - ✅ Self-balancing guarantees - ✅ V8 JIT-friendly code - ✅ Memory-efficient algorithms - ✅ Full TypeScript support - ✅ Order-statistic tree with opt-in `enableOrderStatistic` - ✅ Unified error handling via `raise()` + `ERR` templates --- **Next:** [Check PERFORMANCE.md](./PERFORMANCE.md) for benchmarks. **Or:** [See GUIDES.md](./GUIDES.md) for implementation examples.