blog.hangerthem.com/lib/rehype-unwrap-images.tsx
HangerThem 30ef0d611e Bug fixes, SEO, performance and accessibility pass (#1)
## Summary

A cleanup pass across the blog: route and feed bugs fixed, SEO basics added, smaller and lazier images, accessibility fixes, and a real README. No content or design changes beyond small fixes.

## Bug fixes

- **Tag/category pages**: static params now use slugs, so pre-rendered pages match the links (before, `/tags/standard-model` was only rendered on demand and the pre-built pages said "Tag not found"). Unknown slugs now return a real 404, and there's a new not-found page.
- **Recently updated**: sorts by last modification *before* taking the top 5, so older posts that were updated recently now show up.
- **Atom feed**: `rel="self"` now points to `/feed.xml` (was `/feed`); `<updated>` uses the `updated` frontmatter field.
- **No-JS flash**: the `no-js` → `js` class swap runs in an inline `<head>` script before first paint, not after hydration.
- **Markdown links**: only internal paths go through `next/link`; `#anchors`, `mailto:` and links without an `href` no longer break.
- **Invalid HTML**: images in a paragraph together with text (e.g. "**Screenshots:**" followed by images on the Projects page) produced `<p><figure>` and a hydration error. Those paragraphs are now split.
- **Dates**: frontmatter dates are normalized to ISO, so `<time datetime>` values are valid. Formatting is pinned to Europe/Prague so a UTC build server can't shift a date by a day.

## SEO

- `sitemap.xml` and `robots.txt`
- Canonical URLs, Twitter cards and a feed link on every page, built by a shared `pageMetadata()` helper
- Generated Open Graph images: a site card, and for each post its cover image or a branded title card (pre-rendered at build time)
- `BlogPosting` JSON-LD on post pages

## Performance

- Only the post cover and the latest-post image are preloaded (was: every image)
- In-post image quality 100 → 90, plus responsive `sizes`
- Screenshots downscaled to 1800px wide; `linkboard-old.png` converted to WebP (3.1 MB → 60 KB)
- `CodeWrapper` is now a server component; only the copy button ships to the client
- Mermaid CDN script pinned to `11.17.2` for reproducible builds

## Accessibility

- Accessible names for icon-only links (share, social, mobile nav) and `aria-current` on active nav items
- Copy button: uses React state instead of mutating `innerHTML`, handles clipboard errors, announces the result to screen readers, and is no longer inside an `aria-hidden` header
- Code-block line numbers hidden from screen readers
- Image captions: the Markdown `"title"` becomes the caption; captions that only repeat the alt text are hidden from screen readers

## Content & maintenance

- **Frontmatter validation**: a missing or malformed field fails the build and names the file, instead of rendering a broken post
- **`draft: true` support**: drafts only render in development
- About/Projects/Setup pages share one `MarkdownPage` component; tag/category lists share `PostList`
- Markdown rendering is cached, so metadata and page share one render
- New README covering features, env vars, frontmatter schema, Markdown extensions and deployment
- Removed unused ESLint setup and the unrelated `hast` package; `oxlint`, `oxfmt` and `puppeteer` moved to devDependencies (locked versions unchanged)
- Typo fix in the Tessera cover alt text; removed unused `math`/`mermaid` flags from two posts

## Deployment notes

- **`SITE_URL` is now required**: the build fails with a clear error if it's missing.
- **Puppeteer is a devDependency now**: Nixpacks installs dev dependencies for the build by default, but check that Mermaid diagrams render on the first deploy (the Standard Model post has one).

## Testing

- `tsc --noEmit` and `bun run lint` pass
- `bun run build` passes; all 81 pages are generated statically
- Smoke-tested `next start`: posts, tags, categories, feed, sitemap and OG images return 200; unknown tags and routes return 404
- No block elements nested inside `<p>` in any built page

Reviewed-on: #1
2026-10-19 12:41:40 +00:00

61 lines
2 KiB
TypeScript

import type { Root, Element, ElementContent } from 'hast'
import type { Plugin } from 'unified'
import { SKIP, visit } from 'unist-util-visit'
function isBlank(node: ElementContent) {
return (
(node.type === 'text' && node.value.trim() === '') ||
(node.type === 'element' && node.tagName === 'br')
)
}
function isImage(node: ElementContent): node is Element {
return node.type === 'element' && node.tagName === 'img'
}
/**
* A rehype plugin that lifts <img> elements out of <p> tags, since images render as <figure>,
* which can't be nested inside a paragraph.
*
* A paragraph containing only an image is replaced by the image. A paragraph mixing text and
* images (e.g. "**Screenshots:**" followed by images on the next lines) is split: each image
* becomes a standalone node and the surrounding inline content stays in its own <p>.
*/
export const rehypeUnwrapImages: Plugin<[], Root> = () => (tree: Root) => {
visit(tree, 'element', (node: Element, index, parent) => {
if (node.tagName !== 'p' || !parent || typeof index !== 'number') return
if (!node.children.some(isImage)) return
const replacement: ElementContent[] = []
let run: ElementContent[] = []
const flushRun = () => {
// Drop whitespace and line breaks left over around the lifted images
if (run.some((child) => !isBlank(child))) {
replacement.push({ ...node, children: trimBlankEdges(run) })
}
run = []
}
for (const child of node.children) {
if (isImage(child)) {
flushRun()
replacement.push(child)
} else {
run.push(child)
}
}
flushRun()
parent.children.splice(index, 1, ...replacement)
return [SKIP, index + replacement.length]
})
}
function trimBlankEdges(nodes: ElementContent[]): ElementContent[] {
let start = 0
let end = nodes.length
while (start < end && isBlank(nodes[start])) start++
while (end > start && isBlank(nodes[end - 1])) end--
return nodes.slice(start, end)
}