Solid 2.0’s client runtime, compiled with LilScript and published as @itslil/solidjs. Source: github.com/yeargun/lil-solidjs.

Official js-framework-benchmark keyed table: 68.4% smaller Brotli (11,420 B → 3,609 B) and 0.92× CPU versus Solid 2.0. 18/18 paired browser demos also ship smaller after Brotli (76.3% smaller in total). Solid 2.0 on one side, LSX (LilScript JSX) on the other. Live apps, raw / gzip-9 / Brotli-11 sizes, and measured performance are on the solidlil demo lab.

The keyed-table row is the fair app-sized comparison: both implementations are the official jumbotron + 1,000-row table. Solid’s JSX is compiled by the Solid Vite plugin; solidlil’s LSX is compiled by LilScript to the same cloneNode template style, then terser. The 18 demos are the same programs compiled as closed-world apps; they look even smaller because LilScript can delete unused runtime, while Solid 2.0’s npm browser entry is a pre-bundled dist/solid.js + dist/web.js.

Official js-framework-benchmark CPU, same machine, Chrome 151, 15 blocks, CPU throttling on. Same jumbotron table: Solid’s JSX and solidlil’s LSX both compile to cloneNode templates. Geometric mean of the nine keyed totals is 0.921× Solid 2.0.

npm install @itslil/solidjs

import { createSignal, createMemo, createEffect, flush, createRoot } from "@itslil/solidjs"

createRoot(() => {
  const [count, setCount] = createSignal(0)
  const doubled = createMemo(() => count() * 2)

  createEffect(
    () => doubled(),
    (value) => console.log(value),
  )

  setCount(1)
  flush()
})

@itslil/solidjs targets Solid 2.0 (solid-js@2.0.0-rc.0, @solidjs/web@2.0.0-rc.0): queued writes until flush(), split createEffect(compute, apply), For / Repeat, Loading / Errored, draft-first stores, onSettled, isPending / latest. There is no batch, createResource, or startTransition.

LilScript apps are written in LSX (.lilx) — JSX for LilScript — and compile closed-world (import … from "solidlil"). JavaScript consumers use the tuple helpers on the default entry. A DOM app should import only from @itslil/solidjs/web so it shares one reactive graph; mixing @itslil/solidjs with @itslil/solidjs/web duplicates the runtime.

import { Signal, append, createIntSignal, render } from "solidlil";

func()->void dispose = render("#app", (Element root) => {
  Signal<int> count = createIntSignal(0);
  Element view = (
    <button type="button" onClick={() => { count.write(count.read() + 1); }}>
      {`Count ${count.read()}`}
    </button>
  );
  append(root, view);
});

import { render, bindText, keyedEach, createIntSignal } from "@itslil/solidjs/web"

This is not a different tree-shaking trick on Solid’s side. A normal Solid 2.0 web build is a single bundled file — js-framework-benchmark’s keyed entries are Vite library mode + terser, sideEffects: false, the same as a typical app. The size gap is still large because Solid 2.0 ships the browser runtime as already-concatenated files (solid-js/dist/solid.js and @solidjs/web/dist/web.js). Rollup can drop unused exports from those files, but it cannot recover the fine-grained module graph Solid 1.9 had. Official keyed Solid 1.9 in the same harness is 11,563 B raw; Solid 2.0 is 33,701 B raw; solidlil LSX is 10,020 B raw.

The 18 lab demos are matching closed-world application builds of the same UI — official Solid 2.0 (JSX or TypeScript + Vite + terser) versus LSX — reported as raw, gzip-9, and Brotli-11. They overstate the gap relative to a full table app because a Solid 2.0 counter still carries most of that pre-bundled client. Use the js-framework-benchmark keyed row as the app-level number.

CPU is not inferred from size. The lab’s Playwright medians (npm run bench:perf) are a small demo; the krausest harness (throttled Chrome, 15 blocks) is the number that should be compared to other frameworks.

The reusable package ESM is a different artifact from those app builds. npm run test:size prints package and named-import sizes.

Keep solidlil next to a LilScript checkout, or point at a release compiler:

npm ci
SOLIDLIL_LILSCRIPT_BIN=/path/to/lilscript npm run build
npm run build:apps
npm run bench:perf
npm run check

Set SOLIDLIL_BUILD_MODE=development for a faster local library build. Production is the default. SOLIDLIL_SKIP_BROWSER=1 skips Playwright if Chromium is not installed.

npm test           # Node reactivity + Vite consumer bundle
npm run check      # tests, TypeScript declarations, Pages artifact, npm tarball
npm run test:size  # package raw/gzip/brotli report
npm run build:site # GitHub Pages lab

The implementation is MIT licensed. See NOTICE.md for upstream attribution.