@itslil/marked

This is not the official marked package. It is the parse path of marked@18.0.10 — parse, parseInline, setOptions / options, getDefaults, defaults, and marked() — rewritten in LilScript.

It matches marked@18.0.10 HTML on the GFM + CommonMark corpus (660 / 660). It does not have use(), Hooks, walkTokens, Renderer/Tokenizer subclassing, or the Marked class.

Site: yeargun.github.io/markedlil

npm install @itslil/marked

import { marked } from "@itslil/marked"

document.body.innerHTML = marked.parse("# hello")

Official marked does not sanitize HTML. Neither does this port. Run a sanitizer on the output if the markdown is untrusted.

Why the extension system is gone

LilScript is not JavaScript and is not trying to become JavaScript. It compiles to JS. Classes do not override methods. There is no any. A plugin ABI that installs user objects onto Lexer/Parser/Renderer cannot be expressed without lying about those rules.

The extension system was not stripped to win a size fight. It is absent because the language cannot host it. There is no plan to grow LilScript until every JavaScript pattern ports.

What is compared

Every size and speed number on this page is the same surface: official marked@18.0.10 Lexer, Parser, Tokenizer, Renderer, and helpers — the parse-only sources in official/marked-18.0.10/ — bundled, then run through Oxc and Terser with mangling on and off.

@itslil/marked is the LilScript compiler's own ESM: not bundled, not post-minified, license banner included.

The published npm marked.esm.js still contains use(), Hooks, walkTokens, and the Marked class. It is not a lane. Comparing this port to that file would be a different product against a subset.

LilScript scores a different artifact for each javascript.cost_model. Official Oxc/Terser rows are one file measured three ways. The LilScript library numbers below take raw from the raw compile, gzip from the gzip compile, and Brotli from the Brotli compile. The npm file is the Brotli compile.

• JS library ([mangle] extern_fields = true). extern class MarkedOptions / MarkedApi pin gfm, breaks, pedantic, silent, async, parse, parseInline, setOptions, options, getDefaults, and defaults.
• Closed LilScript (lilscript.closed.toml, extern_fields = false, Brotli compile). Not published. Host members such as string.length stay exact.

Measured with lilscript-codec gzip-9 / Brotli-11. LilScript lanes include the same license banner.

Lane	Raw	gzip-9	Brotli-11	vs Oxc on that codec
Official parse path	67,247	14,064	12,684	—
Official parse path · Oxc mangle on	37,022	10,930	10,092	baseline
Official parse path · Oxc mangle off	47,547	12,153	11,279	—
Official parse path · Terser mangle on	37,725	11,045	10,138	—
Official parse path · Terser mangle off	48,444	12,302	11,339	—
@itslil/marked · matched compiles	33,632	10,727	9,589	0.91× / 0.98× / 0.95×
@itslil/marked · cost_model brotli (npm)	35,985	10,766	9,589	0.95× Brotli
@itslil/marked · cost_model gzip	36,304	10,727	9,603	0.98× gzip
@itslil/marked · cost_model raw	33,632	10,605	9,504	0.91× raw
@itslil/marked · closed LilScript	35,705	10,730	9,533	0.94× Brotli

Against official parse path · Oxc mangle on, the matched library compiles are 5.0% smaller on Brotli-11, 1.9% smaller on gzip-9, and 9.2% smaller raw. Same 660-case HTML.

The closed Brotli compile is 56 Brotli bytes smaller than the npm file: the tax of keeping a JavaScript options object. A program written entirely in LilScript would not pay it.

Performance

Playwright Chromium. HTML checksummed against the official parse path and against published marked@18.0.10 on the spec corpus. Quiet median of 10 samples after discarding the first 3. Ratio is lane / official parse path (lower is faster).

Lane	32× document	660-case ×40	vs official parse path
Official parse path	52.5 ms	67.4 ms	1.00×
Official parse path · Oxc mangle on	52.7 ms	67.9 ms	1.00×
Official parse path · Oxc mangle off	52.5 ms	67.1 ms	1.00×
Official parse path · Terser mangle on	52.4 ms	67.6 ms	1.00×
Official parse path · Terser mangle off	52.3 ms	68.0 ms	1.00×
@itslil/marked	45.7 ms	62.1 ms	0.87×

The document suite is the joined GFM+CommonMark markdown repeated 32 times (~537 kB); the loop suite parses all 660 spec cases forty times. @itslil/marked is 13% faster than the official parse path on documents and 8% faster on the spec loop.

npm test
npm run measure
npx playwright install chromium
npm run bench

Oxc minify uses Vite 8 and needs Node ^20.19 || >=22.12.

Compatibility

The JS library (the npm file) is the only artifact that must keep these names readable after mangling. test/api.test.mjs locks the spellings in the compiler output and calls every entry on ESM, CJS, and UMD.

Name	What it is
parse(src, opt?)	block markdown → HTML
parseInline(src, opt?)	inline markdown → HTML
marked(src, opt?)	same as parse; throws if src is not a string
marked.parse	the marked function
marked.parseInline	same as parseInline
setOptions(opt) / options(opt)	mutate live defaults; return marked
marked.setOptions / marked.options	the same pair
getDefaults()	a fresh factory object (gfm: true, others false)
defaults / marked.defaults	the live options object

Option keys stay exact: gfm, breaks, pedantic, silent, async. async is present and always false — this port does not return a Promise. silent is accepted; on a thrown parse it matches official's error HTML.

The closed LilScript lane turns [mangle] extern_fields off. Those keys mangle. It is not the npm file.

• ESM, CJS, and UMD artifacts
• GFM on by default (breaks: false, pedantic: false)
• Nested links match published marked@18.0.10, not git master
• Spec fixtures live in test/specs/
• Official parse-path sources used for the comparison live in official/marked-18.0.10/

License

MIT. See LICENSE and NOTICE.md. marked is copyright Christopher Jeffrey / MarkedJS.