barely-a-dev-server

barely-a-dev-server

A thin, opinionated wrapper for esbuild as a .ts web server. Given an entryRoot folder, it:

• finds all .ts files under entryRoot and uses them as entry files to run esbuild in watch mode, and
• serves the built .js files together with a fallback to entryRoot for static files.
  • Paths ending in / are mapped to index.html in the corresponding folder.

When run with "dev": false, it writes these files to an output dir (dist/ + the entry root by default), ready to serve using your favorite static file server.

Install with:

npm install -D barely-a-dev-server

Example

// script/build.js
import { barelyServe } from "barely-a-dev-server";

barelyServe({
  entryRoot: "src", // the only required arg
  dev: true,
  port: 3333,
  esbuildOptions: {
    target: "esnext",
  },
});

---

<!-- src/index.html -->
<script src="./index.js" href="./index.ts" type="module" defer></script>

// src/index.ts
const a: number = 4;
console.log(a);

(Note that src must reference the generated .js file, not .ts. The example shows an ergonomic hack: you can use href to store a reference to the .ts source, so that you can e.g. "Follow link" in VSCode.)

Why barely-a-dev-server?

• A thin wrapper around esbuild, which is very fast and robust.
  • Automatically outputs source maps!
• Works just as well as fancy bundlers, if all your code is TypeScript.
• No dependencies other than esbuild.
• Less than 200 lines of source code (unminified).

Why not barely-a-dev-server?

• Hardcoded to assume that you are only using TypeScript for your source and ESM for your output.
• No CLI.
  • If you don't have a build script, you can do this: node -e 'import("barely-a-dev-server").then(s => s.barelyServe({entryRoot: "src"}))'
• No transformations (therefore no optimization) for non-script files.
• No automatic URL opening, no live refresh.
• Uses every .ts file under the entryRoot as an entry point. esbuild handles this very well, but this may result in significantly more output files than expected/needed.
  • A simple workaround is to put as many "library" files as possible outside the entry root, leaving mostly entry files themselves under the entry root.

These are mostly because it would make the codebase significantly larger to support them properly.