experiment: Tauri v2 migration prototype

[seonglae]

Summary

• Prototype migration from electron-nuxt to Tauri v2
• Rust backend with system WebView
• 10-15x smaller bundle (~10-15MB vs ~150MB)

Why Tauri?

Aspect	Electron (current)	Tauri v2
Bundle size	~150MB	~10-15MB
Memory usage	High	Low
Runtime	Chromium	System WebView
Backend	Node.js	Rust
Auto-updater	Yes	Yes (built-in)

Trade-offs

Pros:

• Dramatically smaller bundle (10-15x reduction)
• Lower memory footprint
• Faster startup
• Better security (Rust, restrictive IPC)
• Production-ready, active development

Cons:

• Requires learning basic Rust for backend
• WebView2 dependency on Windows (usually pre-installed)
• Longer first build (Rust compilation)
• Different mental model than Electron

AHK Integration

Full process control via Rust's Command API:

use std::process::Command;

#[tauri::command]
async fn spawn_ahk(app_handle: tauri::AppHandle) -> Result<String, String> {
    let ahk_path = get_ahk_path(&app_handle);
    match Command::new(&ahk_path).spawn() {
        Ok(child) => Ok(format!("PID: {}", child.id())),
        Err(e) => Err(e.to_string()),
    }
}

Verdict: FULLY COMPATIBLE - Can spawn, track PID, kill process cleanly.

Files Changed

• src-tauri/Cargo.toml - Rust dependencies
• src-tauri/tauri.conf.json - Tauri configuration
• src-tauri/src/main.rs - Rust backend with AHK commands
• src-vue/ - Vue 3 frontend
• vite.config.ts - Vite configuration

Migration Path

1. Set up Tauri project structure
2. Learn basic Rust for backend commands (~50 lines)
3. Port Vue components to pure Vue 3 (from Nuxt)
4. Implement IPC for AHK control
5. Test on Windows thoroughly

Recommendation

PRIMARY RECOMMENDATION - Best overall choice for:

• Users who care about download size
• Background apps that need low memory
• Long-term maintainability

---

🤖 Generated with Claude Code

[seonglae]

CI npm ci was failing with ERESOLVE because the legacy @nuxtjs/composition-api devDependency requires Nuxt 2 while the project is on Nuxt 3. Added .npmrc with legacy-peer-deps=true to make npm ci resolve the existing lockfile. Verified locally: npm ci, npm run build, and cargo check all pass.

[seonglae]

Automated CI fix applied. All 5 checks are now green (test + 4-platform build matrix).

Root causes

• packages/mac-input declared os: ["darwin"], which made npm ci fail with EBADPLATFORM on Linux runners.
• package-lock.json was generated on darwin and was missing platform-specific optional dependencies (e.g. @rollup/rollup-linux-x64-gnu), so vite build crashed on Linux (npm bug #4828).
• The test job ran cargo check against the Tauri Rust crates on ubuntu-latest but never installed libwebkit2gtk-4.1-dev / libappindicator3-dev / librsvg2-dev / patchelf, so glib-sys and gobject-sys couldn't find their system libraries via pkg-config.
• The build matrix called cargo tauri build, but only @tauri-apps/cli is installed (via npm). That provides a tauri binary, not a cargo tauri subcommand.

Fixes

• Dropped the darwin-only os field from packages/mac-input/package.json and the matching entry in package-lock.json (mac-input has no postinstall and its scripts are manually invoked, so it installs harmlessly on every OS).
• Regenerated package-lock.json so all platform-specific rollup binaries are tracked.
• Added the Tauri Linux system-deps install step to the test job (mirroring the build job).
• Changed cargo tauri build to tauri build in the build matrix.