fosdem.org

Hugo static site for fosdem.org. Schedule data is preprocessed with a Go program that reads the Pretalx JSON export and writes derived data files for Hugo to consume.

Prerequisites

All tools are provided by the Nix dev shell:

nix develop

This gives you go, hugo, pagefind, just, validate-ics and check-links.

Without Nix, install these manually: Go 1.25+, Hugo 0.155+, and Pagefind 1.4+.

Data Pipeline

The schedule data flows like this:

schedule.json  -->  go run main.go  -->  data/*.json  -->  Hugo templates
(Pretalx export)                         (derived data)

schedule.json is the raw Pretalx schedule export (~5 MB). It is gitignored. Download it from Pretalx and place it in the project root. Note: you must be logged in, otherwise you will get a 404.

main.go reads this file at startup and writes the following to data/:

File	Contents
schedule.json	Full parsed conference data
events.json	Map of event slug to event(s)
speakers.json	Map of speaker GUID to speaker + event list
tracks.json	Map of track name to events
tracklist.json	Tracks grouped by type (main/devroom/other)
keynotes.json	Keynote events (curated by slug)
roominfo.json	Per-room day/time ranges and track listings
roomtracks.json	Room x time grid for schedule overview

data/sponsors.json is maintained by hand.

All files under data/ are gitignored.

Build

# 1. Generate data files from schedule.json
go run main.go

# 2. Build the site
hugo build -D

# 3. Index for search
pagefind --site "public"

Dev Server

go run main.go
hugo server --baseURL=http://127.0.0.1/2026 -D --disableFastRender

Always test with a baseURL. The site will be deployed under a subpath, and all internal links use relURL (e.g. {{ url | relURL }}) to resolve paths correctly against it. Without a baseURL, broken links won't surface during development.

Deploy

Both targets are served from www-public0.fosdem.org and authenticate with Kerberos, so get a ticket first. It prompts for a password and lasts a few hours:

kinit {id}@FOSDEM.ORG
klist                      # confirm a krbtgt/FOSDEM.ORG ticket exists

rsync runs over ssh, which needs GSSAPI auth switched on. Either pass it per command or put it in ~/.ssh/config once:

Host *.fosdem.org
    GSSAPIAuthentication yes
    PreferredAuthentications gssapi-with-mic,publickey,password

Deploys run from the justfile. Both build first, then rsync:

just deploy-staging          # dry run
just deploy-staging go=1     # writes
just deploy-live go=1

The landing site sits under /2027/, mirroring production, so the relURL paths exercised on staging are the ones that ship. Staging builds with -D, production without, so drafts are reviewable on staging and never reach production.

Each target has its own output directory, dist/staging and dist/live, built with --cleanDestinationDir. Without it Hugo leaves stale output in place, leaking drafts and the wrong baseURL from the previous build.

--delete is scoped to the 2027/ subdirectory, so it cleans up files removed since the last deploy without touching anything else in the docroot. Never point it at the docroot root.

Production is /var/www/fosdem.org/public/2027/ as www-live. A dry run does not test write permission, so a clean -n pass is no guarantee the real sync will succeed.

PR Previews

Every pull request is published at https://staging.fosdem.org/pr-{number}/ and emptied when the PR closes. Locally:

just build-preview 123         # dist/pr-123, baseURL https://staging.fosdem.org/pr-123/
just deploy-preview 123        # dry run
just deploy-preview 123 go=1   # writes
just undeploy-preview 123 go=1 # removes the remote directory

Previews build the landing site only, with -D. pr must be a number.

Landing Site

hugo.landing.yaml is a separate site in this repo: the edition announcement, news, and the standing pages (/support/, /support/sponsors/, /support/donate/, /faq/, /contact/, /practical/conduct/). Own contentDir/layoutDir, so hugo.yaml, content/ and layouts/ are not involved and no schedule.json is needed.

hugo server --config hugo.landing.yaml --baseURL=http://127.0.0.1/2027 --disableFastRender
hugo build  --config hugo.landing.yaml -b https://fosdem.org/2027/

The feed is published at /rss.xml, not Hugo's default /index.xml. That is the URL subscribers use (fosdem.org/rss.xml redirects to the current edition). News items get real pages so <link>/<guid> resolve, and /news/ lists them, linked from the footer and the homepage.

News Posts

Posts are page bundles, created with the directory form. Passing index.md on the end silently falls back to Hugo's built-in archetype and you lose the slug:

nix develop -c hugo new content --config hugo.landing.yaml \
    --kind news news/2026-10-02-call-for-participation

The directory keeps the date so content-landing/news/ stays chronological, while the generated slug drops it from the URL (/2027/news/call-for-participation/). Posts start draft: true; preview with -D.

Images go beside index.md and are referenced by filename. A render hook resolves them as page resources and generates a srcset capped to the article width. A quoted title turns the image into a <figure> with that caption:

![Alt text](campus.png "Optional caption")

The feed and og:image use a featured image, defaulting to the first image in the directory and overridable with image: in front matter.

The draft 2026-08-31-example-post-with-image exercises every element a post can contain, callouts included. Preview it with -D rather than documenting it here.

/news/ is an index of title and date, newest first, grouped by year once posts span more than one. Full text lives on the post page.

Calendars

The homepage emits /2027/fosdem.ics from the DateStart/DateEnd params. Keep those in sync with DateFull.

A post ships its own calendar by listing dates and picking a calendar output:

outputs:
  - html
  - calendar-cfp
events:
  - date: 2026-10-04
    summary: "FOSDEM 2027: deadline for developer room proposals"

The filename comes from the output format, so another name means another format in hugo.landing.yaml (calendar-cfp writes fosdem-cfp.ics). No template needed: page.ics carries no format name, so it serves them all.

All-day VEVENTs, no timezone. Each carries the post link as URL and in DESCRIPTION, since Google Calendar drops URL on import.

UIDs derive from the post directory and the list position. Append to events:; reordering re-issues them and subscribers see the entries again as new.

Validation runs in CI after the landing build:

nix develop -c validate-ics public

Links

check-links fetches every external link in a built site. Internal links are covered by Hugo: an unresolved relref fails the build.

nix develop -c check-links dist/staging --skip payment-links.mollie.com

Non-zero exit on failure. Not in CI. Mollie payment links 404 on every path, valid ones included.

Metrics

The homepage emits /2027/metrics in Prometheus text exposition format, from the metrics output format and home.metrics:

# HELP website_last_build_time Starting time of the last website build
# TYPE website_last_build_time gauge
website_last_build_time 1792391980

now.Unix at build time, so a scrape that stops advancing means the build stopped. The scrape target carries the year prefix: https://fosdem.org/2027/metrics.

The extensionless filename comes from the text/prometheus media type: empty suffixes and delimiter.

Standalone Pages

Paths come from the file tree. Use url: only when the main site's path has no section here yet:

title: "Code of Conduct"
url: /practical/conduct/

Dateless, so the layout omits the dateline. Link them from _partials/footer.html, or _partials/header.html for the few that belong in the top nav.

Sponsors and FAQ entries are one file each, assembled into a single page with anchors and never rendered on their own:

content-landing/support/sponsors/<id>/index.md   tier: cornerstone|main|thanks
content-landing/support/sponsors/<id>/logo.svg   optional, page resource
content-landing/faq/<category>/_index.md         category title and weight
content-landing/faq/<category>/<question>.md     weight orders within the category

The directory name is the anchor. /support/ lists its own children.

Feed <guid>s are path-independent tag URIs derived from the directory name, so changing a slug is free. Renaming the directory re-issues the guid and subscribers see the post again as new.

The /rss.xml -> /<year>/rss.xml redirect is server-side, not in this repo. It must be repointed to /2027/ or subscribers keep getting the 2026 feed.

Project Structure

content/
  about.md, practical.md, ...   Static pages (Markdown)
  news/                         News articles
  schedule/
    event/_content.gotmpl       Generates a page per event from data
    speaker/_content.gotmpl     Generates a page per speaker from data
    track/_content.gotmpl       Generates a page per track from data
    room/_content.gotmpl        Generates a page per room from data

layouts/
  baseof.html                   Base template (nav, footer, dark mode)
  home.html                     Homepage
  event.html, speaker.html, ... Layout per page type
  partials/                     Shared template fragments

static/
  css/style.css                 All styles (CSS custom properties, dark mode)
  css/fonts/                    Bundled fonts (Signika, DejaVu Sans)
  js/leaflet.js                 Map library

Pages under content/schedule/ use Hugo content adapters (_content.gotmpl) to dynamically create pages from the data files, so no individual Markdown file is needed per event/speaker/track/room.