Skip to content

search.md

Latest commit

 

History

History
77 lines (50 loc) · 3.82 KB

search.md

File metadata and controls

77 lines (50 loc) · 3.82 KB
Raw

Site search

Full-text search across the current edition plus every archived edition, served entirely client-side from a static Pagefind index. No server, no third-party analytics.

What gets indexed

make build invokes Pagefind after Lektor finishes. Pagefind walks site/ and indexes every page that:

  • Has a top-level <html> element (the lone exception is the Google verification file, which has no <html> tag — Pagefind warns and skips it).
  • Does not carry <meta name="robots" content="noindex"> (so redirect pages are skipped automatically).

Inside each indexable page, only content within an element marked data-pagefind-body is searched. The current markers:

Template Body element Notes
talk.html <article class="talk-article" data-pagefind-body> Title, abstract, full description, speakers, full transcript
speaker.html <article class="speaker-article" data-pagefind-body> Name, bio, company, talks list

Other pages (blog posts, sponsor pages, FAQs, static pages) are indexed via Pagefind's default <main> fallback.

Filter facets are declared on the same articles using hidden <span data-pagefind-filter="key:value" hidden></span> elements (one element per filter — HTML parsers collapse duplicate attributes on the same tag, so they have to be separate elements).

Talks declare: year, kind=Talk, format (talk / keynote), track, recording (yes / no), transcript (yes / no), python_skill, domain_expertise.

Speakers declare: year, kind=Speaker.

/search/ page

Single page at /search/ hosts the Pagefind UI bundle, scoped to the conference and pre-styled to match the site palette.

Features:

  • URL ?q= prefill so external links can deep-link a query (and so the SearchAction JSON-LD on the homepage works for Google's in-SERP search box).
  • / keyboard shortcut to focus the input from any page.
  • Filter sidebar shows the eight facets above with auto-generated value lists. A small post-render JS pass swaps underscored filter names (python_skill) for spaces (Python skill).
  • Search-action JSON-LD on the page declares the entry point at /search/?q={search_term_string} per Google's spec.

Local development

The Lektor dev server does not serve the Pagefind index — it doesn't know about the pagefind/ directory and prunes it on the next reload. For search-testing locally, use the static profile:

make build
python3 -m http.server 8088 --directory site

Or, with the bundled launch profile:

.claude/launch.json → "static" profile → port 8088

Pagefind itself is a make build step:

make pagefind        # site/ → site/pagefind/   (~12 MB, ~10 KB shards loaded on demand)

Updating the indexer

pagefind (Python wrapper) + pagefind_bin (bundled native binary) are pinned as dev dependencies in pyproject.toml. Bump together:

uv add --dev "pagefind>=1.5.2"
uv add --dev "pagefind_bin>=1.5.2"

The CSS / UI bundle ships with the same package, so a major version bump may require revisiting the override CSS in templates/search.html (--pagefind-ui-* variables).

Common follow-ups

Symptom Fix
New filter facet doesn't show up Add a <span data-pagefind-filter="key:value" hidden></span> inside the data-pagefind-body article in the relevant template, rebuild
Talk indexed but transcript missing from search Verify the transcript is rendered in the same data-pagefind-body article (it is, in talk.html) and that the talk's transcript_status != 'none'
Pagefind warns "page has no <html> element" Acceptable for verification files (Google site-verification HTML) — the page is intentionally bare and not search-relevant
Search returns no results in make run Expected; the Lektor dev server doesn't expose pagefind/. Use the static profile above