aboutsummaryrefslogtreecommitdiffstats
path: root/DOCS/performance_analysis.md
diff options
context:
space:
mode:
Diffstat (limited to 'DOCS/performance_analysis.md')
-rw-r--r--DOCS/performance_analysis.md99
1 files changed, 99 insertions, 0 deletions
diff --git a/DOCS/performance_analysis.md b/DOCS/performance_analysis.md
new file mode 100644
index 0000000..41f0bb9
--- /dev/null
+++ b/DOCS/performance_analysis.md
@@ -0,0 +1,99 @@
+# Performance Analysis & Recommendations
+
+## Executive Summary
+You are correct that the React implementation introduces inherent latency compared to the vanilla JS/Backbone version, particularly during initial load and interactions. While backend measurements show fast response times (~5ms) for the current dataset (where `full_content` is mostly empty), the React frontend's rendering strategy and bundle overhead are the primary contributors to the perceived slowness.
+
+## Key Findings
+
+### 1. React Rendering Overhead vs. Vanilla JS
+- **Initial Load**: The React application requires downloading a larger bundle (~241KB vs ~150KB for legacy) and then executing hydration before fetching data. This adds an initial delay (Time to Interactive) not present in the lightweight vanilla version.
+- **Excessive Re-rendering**: The current `FeedItems` component triggers a re-render of the **entire list** (and all children `FeedItem` components) whenever a single item is selected or modified. React's reconciliation process diffs every item, which is computationally expensive compared to Backbone's direct DOM manipulation of a single element.
+- **DOM Node Count**: Even with small content, rendering 15+ complex article cards with `dangerouslySetInnerHTML` involves significant DOM operations. React's synthetic event system and component overhead add micro-latency per item.
+
+### 2. Implementation Inefficiencies
+- **Observer Churn**: The `IntersectionObserver` in `FeedItems.tsx` disconnects and reconnects for *every item* whenever the list updates or state changes. This is an O(N) operation that degrades performance as the list grows.
+- **Lack of Memoization**: `FeedItem` components are not memoized (`React.memo`), causing them to re-render unnecessarily on parent state changes (like selecting an item with `j`/`k`).
+
+### 3. Backend Data Strategy (Minor Current Impact, Major Risk)
+- Although `full_content` is currently empty for most items, the backend still retrieves and serializes this field for every item in the list. This is "fast enough" now (5ms) but remains a structural inefficiency that will cause severe slowdowns if/when content is scraped.
+
+## Recommendations
+
+### 1. Optimize React Rendering (High Impact)
+The most effective way to restore "snappiness" is to reduce React's work:
+- **Memoize `FeedItem`**: Wrap the component in `React.memo` so it only re-renders when its specific props change. This prevents the entire list from flashing when you select one item.
+- **Virtualize Long Lists**: Implement `react-window` or `virtuoso` for both the sidebar (`FeedList`) and main view (`FeedItems`). This ensures only visible items are in the DOM, keeping the browser responsive regardless of list size.
+- **Stable References**: Use `useCallback` for event handlers passed to children to prevent breaking memoization.
+
+### 2. Fix Observer & Effect Logic (Medium Impact)
+- Refactor the `IntersectionObserver` in `FeedItems.tsx` to maintain a stable observer instance using `useRef` and only observe/unobserve specific elements as needed, rather than resetting the whole list.
+
+### 3. Backend Optimization (Defensive)
+- Even if not the current bottleneck, modifying `item.Filter` to exclude `full_content` on list views is a simple change that prevents future performance regressions.
+
+## Performance Optimizations Implemented (Current Session)
+
+Following the analysis above, the following optimizations have been applied to the React frontend:
+
+### 1. Component Memoization
+- **Change**: `FeedItem` is now wrapped in `React.memo`, and event handlers (`onToggleStar`, `onUpdate`) are memoized with `useCallback`.
+- **Impact**: Previously, clicking an item or pressing 'j'/'k' caused the **entire list** of items to re-render because the parent `FeedItems` state changed. Now, only the specific item being modified re-renders. This transforms interaction complexity from **O(N)** to **O(1)**, significantly improving apparent responsiveness.
+
+### 2. Stable IntersectionObserver
+- **Change**: The `IntersectionObserver` in `FeedItems` now uses a `useRef` to maintain a stable instance. It no longer disconnects and reconnects on every render or state change. The observer now references current state via refs (`itemsRef`, `hasMoreRef`) to avoid stale closures without triggering effect re-execution.
+- **Impact**: Removes the overhead of constantly destroying and recreating overlapping observers. Scrolling performance is smoother, and the "read" marking logic is more reliable and efficient.
+
+### 3. Event Handler Optimization
+- **Change**: Keyboard event handlers now use `refs` to access the latest state (`items`, `hasMore`, `loadingMore`) without needing to be re-attached on every render.
+- **Impact**: Reduces React's internal bookkeeping overhead and prevents event listener churn.
+
+---
+
+## Proposal: "Vanilla JS Optimized" Frontend (Modern Backdrop)
+
+Given your preference for the responsiveness of the legacy Backbone version, a modern Vanilla JS approach could offer the best of both worlds: the raw speed of direct DOM manipulation with the maintainability of modern ES6+ standards.
+
+### Core Philosophy
+**"No Framework, Just Platform."**
+Instead of React's Virtual DOM diffing (which adds overhead), we check state and update *only* the specific DOM nodes that change.
+
+### Architecture Proposal
+
+1. **State Management**:
+ - Use a simple **Store Pattern** using ES6 `Proxy` or a lightweight `Pub/Sub` module.
+ - State changes (e.g., `store.items[0].read = true`) automatically trigger specific DOM updates via subscribers, without re-evaluating a component tree.
+
+2. **Rendering**:
+ - **Initial Render**: Use **Tagged Template Literals** (like `lit-html` but simpler) to generate HTML strings efficiently.
+ ```javascript
+ const itemTemplate = (item) => `
+ <li class="feed-item ${item.read ? 'read' : ''}" data-id="${item.id}">
+ ...
+ </li>`;
+ ul.innerHTML = items.map(itemTemplate).join('');
+ ```
+ - **Updates**: Direct DOM manipulation.
+ ```javascript
+ // When item 101 becomes read:
+ document.querySelector(`li[data-id="101"]`).classList.add('read');
+ ```
+ - **Benefits**: Zero diffing cost. Instant updates.
+
+3. **Component Structure**:
+ - Use **Web Components (Custom Elements)** for encapsulated, reusable UI elements (e.g., `<feed-item>`).
+ - Native browser support means no framework overhead.
+
+4. **Routing**:
+ - Lightweight wrapper around **History API** to handle URL changes and view swapping without full reloads.
+
+5. **Build Tooling**:
+ - **Vite** (for dev server/HMR) but configured to bundle vanilla JS.
+ - **CSS**: Standard CSS variables for theming (already partially in place).
+
+### Why This Beats React for Neko
+- **Zero Hydration**: The browser parses HTML and executes minimal JS. TTI is effectively immediate.
+- **Memory Footprint**: No Virtual DOM copies of the tree.
+- **Predictable Performance**: You control exactly when and how the DOM updates, mirroring the Backbone experience but with cleaner, modern code.
+
+## Conclusion
+The "sluggishness" was primarily due to React's re-rendering of the entire list on interactions and the initial hydration cost. The **Memoization** and **IntersectionObserver** optimizations implemented have significantly reduced the re-rendering overhead, bringing responsiveness closer to the vanilla JS experience. However, a move to a modern Vanilla JS architecture (as proposed) would eliminate the inherent framework overhead entirely.