Skip to content

PERFORMANCE_AUDIT.md

Latest commit

 

History

History
76 lines (56 loc) · 4.62 KB

PERFORMANCE_AUDIT.md

File metadata and controls

76 lines (56 loc) · 4.62 KB
Raw

Performance Audit: PaperCache V0.5.8

Date: June 29, 2026 Auditor: VariableThe
App Version: 0.5.8 (Tauri, custom arithmetic evaluator)

1. Executive Summary

PaperCache is built on Tauri v2, Rust, and React. After migrating from Electron (v0.5.0-beta), we've continued shrinking the footprint through dependency removal, custom evaluators, and lazy-loading.

Metric V0.5.0-beta V0.5.8 Delta
App Installer Size (DMG) ~7.3 MB ~6.8 MB -7%
Idle RAM Usage ~40 MB ~38 MB -5%
Idle CPU Usage 0.0% 0.0% Maintained
Bundle Dependencies ~600 packages ~590 packages Smaller

2. Testing Methodology & Environment

  • Hardware: MacBook Air M4, 16GB RAM
  • OS: macOS 15.7.5
  • Dataset: Workspace containing 500 markdown notes, averaging 2KB each.
  • Tooling: Activity Monitor, ls -lh, du -sh

3. Recent Performance Improvements (V0.5.3 — V0.5.8)

Custom Arithmetic Evaluator (V0.5.8)

Replaced expr-eval (~15KB, high-severity prototype pollution vulnerability, no fix available) with a custom recursive-descent parser at src/lib/evaluator.ts (~2KB).

  • Zero external dependencies — no eval, no Function constructors.
  • ~150 lines of TypeScript with 26 unit tests.
  • Not susceptible to prototype pollution.
  • Impact: Eliminated a security vulnerability with a smaller, faster evaluator.

Unused Dependency Removal (V0.5.6 — V0.5.8)

Removed 4 unused npm packages: @tauri-apps/plugin-fs, @tauri-apps/plugin-shell, @emnapi/core, @emnapi/runtime.

  • Impact: Reduced package.json weight and CI install times.

Lazy-Loaded WebGL Graph View (V0.5.3)

Replaced D3.js Canvas 2D graph with react-force-graph-3d (Three.js/WebGL), dynamically imported via React.lazy() + Suspense.

  • The Three.js bundle (~1.3 MB) loads only when the graph is first opened.
  • Impact: Graph rendering is offloaded to the GPU. No startup penalty. No UI freeze on large graphs.

DSL Regex Engine — Visible-Range Scanning (V0.5.3)

Created dslPlugin.ts — createRegexPlugin() scans only view.visibleRanges per update tick.

  • O(visible lines) complexity instead of O(document length).
  • Impact: Lag-free typing at any document size, even with many active regex rules. Eliminated the previous main-thread bottleneck for custom DSL parsing.

Drift-Corrected Timer Countdowns (V0.5.3)

Replaced setInterval with chained setTimeout loops in the timer panel.

  • Impact: No timer drift accumulation. Accurate countdowns regardless of browser event loop pressure.

Debounced Saves (V0.4.0, maintained)

500ms debounce on saveNote IPC calls.

  • Disk I/O reduced from ~3 writes/sec to ~1 write/sec during continuous typing.

TypeScript Strict Mode (V0.5.8)

Enabled strict: true in tsconfig, catching null/type issues at compile time rather than runtime.

  • Impact: Zero new type errors at enablement — the codebase was already compatible.

Coverage Thresholds (V0.5.8)

Added minimum coverage guardrails to vitest config (statements 65%, branches 50%, functions 55%, lines 65%).

  • Impact: Prevents silent coverage regression in CI.

4. Desktop Integration Overhead

  • Native API Key Storage: Uses the OS keychain (macOS Keychain, Linux Secret Service, Windows Credential Manager) via the Rust keyring crate. No user-visible memory or latency cost.
  • Global Shortcuts: Registered via tauri-plugin-global-shortcut — zero ongoing CPU overhead, fires events only on keypress.
  • Auto-Updates: Uses tauri-plugin-updater — checks version metadata on startup (~1 HTTP HEAD request). No persistent overhead.
  • Window State Persistence: Reads/writes a small JSON file (~200 bytes) on exit/startup. No runtime cost.

5. Known Limitations & Future Bottlenecks

  1. Full-Text Search: Currently searches are done client-side via JavaScript string matching on loaded notes. For workspaces exceeding 5,000+ notes, this could introduce noticeable search latency. A future optimization could implement a persistent search index in Rust (e.g., via tantivy or skim).

  2. Graph Layout on Very Large Datasets: The WebGL graph renders smoothly for typical workspaces (up to ~500 nodes). With 2,000+ nodes, force simulation convergence time and interaction framerate may degrade. Future optimization: implement LOD (level-of-detail) rendering or cluster-collapse for large folders.

  3. Startup Time (Cold): Initial load requires parsing note directory via Rust's walkdir (~5ms for 500 notes) plus React hydration. The JS bundle is ~200KB gzipped. Cold launch is typically sub-second but varies by filesystem speed and note count.