jebin2.github.io

The site — https://www.voidall.com. Static, no framework, no bundler, no runtime dependencies. Plain HTML, one stylesheet, and native ES modules.

Posts live in content/ in this same repo, so writing and publishing are one push. Every byte a reader receives is served by GitHub Pages; nothing is fetched from a third party at read time.

---

Publishing a post

Create a folder under content/ and a .md file inside it with the same name:

content/C/Bitfields/Bitfields.md   ✅ indexed as a post
content/C/Bitfields/header.png     ✅ image, referenced as ![](header.png)
content/C/Bitfields/draft.md       ❌ silently ignored — name doesn't match the folder

git add . && git commit -m "post: bitfields" && git push
git pull   # the build commits back to main; pull before your next push

Everything else is generated: the index, the date, the URL, the page, the description, the preview image, and the sitemap entry.

If a post never shows up, check the filename/folder rule first — a mismatch is skipped silently, by design, so a post can keep notes and images in its own directory.

Images and links inside a post

Use plain relative paths — ![](header.png). The build rewrites them to /content/… when it renders the page. Absolute https:// links are left alone.

---

The build

flowchart LR
    A[push content/] --> B[generate-manifest.mjs<br/>dates from git history]
    B --> C[prerender.mjs<br/>writing/slug/index.html]
    C --> D[generate-sitemap.mjs]
    D --> E[commit]
    E --> F[GitHub Pages]

Loading

.github/workflows/sitemap.yml (“Build site”) runs on a push touching content/, config.json, js/utils.js or scripts/** — or manually via Run workflow. It commits only when the output actually changes.

Script	Writes
generate-manifest.mjs	content/manifest.json	Indexes posts and SVGs. Dates come from each file's git history — a fast single pass over git log, falling back to --follow for files that have been renamed.
prerender.mjs	writing/<slug>/index.html	Renders each post's Markdown with marked, and writes real per-post <title>, description, og:image and canonical.
generate-sitemap.mjs	sitemap.xml	Static pages dated by their last commit, posts by last_modified_date.

Running it locally

npm install --no-save marked@9.1.6
node scripts/generate-manifest.mjs
node scripts/prerender.mjs
node scripts/generate-sitemap.mjs

The build reads everything off disk — no network — so it works offline and can't be broken by an upstream outage. Output is deterministic: the displayed date is formatted from the literal date in the manifest rather than the machine's timezone, so CI and a laptop produce identical files.

---

Layout

Path
content/	the posts, their images, data/*.json, svgs/, and the generated manifest.json
config.json	the one file that says where content lives — local paths under /content/
index.html	post listing (home)
writing/index.html	the writing listing
writing/<slug>/	generated — one prerendered page per post
projects.html, linksilike.html, stats.html	the other pages
404.html	not-found page, and the fallback renderer for posts not yet built
sitemap.xml	generated
.nojekyll	stops Pages filtering content/ through Jekyll
css/style.css	the whole stylesheet
js/*.js	one module per page, plus shared helpers
scripts/*.mjs	build scripts (Node, run in CI)

The JavaScript

Module
shared.js	loads config.json, injects header/footer, sets per-view metadata
cache.js	sessionStorage with a 5-minute TTL; falls back to a stale entry when a fetch fails
utils.js	escaping, date parsing, the post listing, and slugifyTitle()
blog.js	the writing listing, client-side post rendering, and the not-found view
post.js	runs on prerendered pages — chrome, analytics, highlighting, share UI
share.js	admin-only X share + article modal
analytics.js	fire-and-forget event inserts to Supabase
supabase.js	client config and the JSON fetch helper

slugifyTitle() in utils.js is imported by both the browser and the build, so listing links always resolve to the files the build wrote. Changing that function changes every post URL.

---

How a page renders

Prerendered post pages arrive with the article already in the HTML — a preview bot, a search crawler, or a reader with JS disabled all get the real post. JavaScript afterwards only adds the nav and footer, syntax highlighting, a read count, and (for admins) the share buttons.

The listing pages are still client-rendered: they fetch /content/manifest.json from this same origin plus two get_stats calls to Supabase — one for the all-time sort, one for the 24-hour counter beside each title.

Fallbacks. If a post is in the manifest but hasn't been prerendered yet, /writing/<slug>/ 404s, Pages serves 404.html, and blog.js resolves the slug against the manifest and renders it in the browser. Legacy /writing/?post=<path> URLs work the same way and point their canonical at the prerendered page. Any other unknown URL gets a real not-found.

Analytics

Writes go straight to a Supabase events table with a publishable key (insert-only RLS). Reads never touch the table — they go through the get_stats RPC, which aggregates server-side and takes an optional since_date. js/analytics-embed.js is a standalone snippet for tracking clicks from other projects.

Analytics failures are swallowed by design: counting reads must never block reading.

---

Worth knowing

Post dates come from git history. Never rewrite history under content/, and never copy posts in without it — a plain copy resets every created_date to the day of the copy. This is why the blog repo was merged in with its full history rather than copied.

Post URLs come from post titles. Pointer Arithmetic → /writing/pointer-arithmetic/. Renaming a post moves its URL and prunes the old directory, with no redirect. Rename freely before sharing a post; think twice afterwards. Two posts can't share a title — the build fails loudly rather than letting one page overwrite the other.

Bumping marked means regenerating its SRI hash. writing/index.html and 404.html load it from a CDN with integrity set; a stale hash blocks the script and the client-side fallback stops rendering. Keep the version pinned in the workflow matching the one in the HTML so build-time and browser rendering agree.

isAdmin() is a localStorage flag, trivially set by anyone. It only reveals share and export UI, nothing privileged. Don't let it grow into gating anything that matters.

---

History

Posts used to live in a separate repo, jebin2/blog, fetched at read time through the jsDelivr CDN and built via a cross-repo trigger. That repo is retired — its history now lives under content/ here. Pushing to it no longer affects this site.