desktop/package.json has no "type": "module", and vite 8 warns about it

[dspury]

What

desktop/package.json declares no "type", so every .ts/.js file in the package is loaded as CommonJS by default. Two tools already complain, for the same reason.

Vite 8, on every dev-server start and build (surfaced by #137):

(!) Your Vite config uses features that are unsupported by `configLoader: 'native'`,
    which is planned to become the default in a future major version of Vite:
  - ESM syntax in a file loaded as CommonJS (vite.config.ts:1:1).
    Use a `.mjs` extension or set `"type": "module"` in the closest package.json
Set `VITE_CONFIG_NATIVE_IGNORE_WARNING=true` to suppress this warning.

oxlint, on every npm run lint:

(node:39406) [MODULE_TYPELESS_PACKAGE_JSON] Warning: Module type of
file:///.../desktop/tools/oxlint/anti-slop/index.ts is not specified and it
doesn't parse as CommonJS. Reparsing as ES module because module syntax was
detected. This incurs a performance overhead.
To eliminate this warning, add "type": "module" to .../desktop/package.json.

Nothing is broken today — both tools fall back correctly. The vite one is a deprecation notice with a stated end date: when configLoader: 'native' becomes the default in a future vite major, the fallback goes away.

Why it is not a one-line change

vite.config.ts uses __dirname, which does not exist in an ES module:

root: resolve(__dirname, 'renderer'),
build: { outDir: resolve(__dirname, 'dist/renderer'), … }

So adding "type": "module" requires replacing it with
dirname(fileURLToPath(import.meta.url)), and then checking every other file the package loads as CommonJS — .eslintrc.cjs is already explicitly .cjs and would be fine, but tools/oxlint/anti-slop/index.ts, scripts/stamp-release.js (which uses require), and anything else executed by node rather than bundled all need auditing.

Options

1. Add "type": "module" and fix the fallout. Correct long-term, and removes both warnings. Needs the __dirname replacement plus a sweep of node-executed files.
2. Rename vite.config.ts → vite.config.mts. Silences the vite warning specifically, with no package-wide change. Leaves the oxlint warning.
3. Set VITE_CONFIG_NATIVE_IGNORE_WARNING=true. Suppresses the message without addressing anything — only worth it as a deliberate, commented decision.

Option 1 is the real fix. Worth doing before the vite major that flips the default, rather than under time pressure when it lands.

Verification note

Whatever lands must be checked against the dev server, not just vite build — the ferry-dev-csp plugin in vite.config.ts rewrites the meta CSP so @vitejs/plugin-react can inject its refresh preamble, and if the config fails to load the dev window renders nothing at all.

[dspury]

The blast radius of adding "type": "module" just grew.

#144 migrates tsconfig.shared.json and tsconfig.electron.json from the deprecated moduleResolution: "Node" (node10) to module/moduleResolution: "node16" — required for TypeScript 6, and node10 is fully removed in 7.

Under node16, TypeScript decides CommonJS vs ESM emit from the nearest package.json "type" field. Today there is none, so .ts files compile to CommonJS and dist/electron/main.js still starts:

"use strict";
Object.defineProperty(exports, "__esModule", { value: true });

Adding "type": "module" would flip that to ESM. electron/main.ts is the app's main entry point, so getting this wrong is not a lint warning — it is the app failing to launch.

That does not make option 1 wrong; ESM main is supported by modern Electron. It does mean this issue is no longer "fix __dirname and sweep the node-executed files". Anything landing it now has to:

• confirm the compiled dist/electron/main.js and dist/electron/preload.js still load, by launching the app, not by typechecking;
• decide whether electron/*.ts should emit ESM at all, or whether those two configs should pin "module": "commonjs" explicitly so the package-level "type" cannot move them;
• re-check the packaged path too — asar + ESM has its own resolution rules, and Packaged app cannot start: sidecar.spec omits migration 003, so the frozen sidecar can't open a current database #139 is a fresh reminder that packaged behaviour differs from dev.

Option 2 from the original list (rename vite.config.ts → vite.config.mts) is now clearly the cheaper move if the goal is just silencing the vite warning: it touches one file and cannot affect the Electron emit.