links.yml

name: Links

on:
  pull_request:
  push:
    branches: [main]
  workflow_dispatch:

permissions:
  contents: read
  issues: write

jobs:
  linkcheck:
    runs-on: ubuntu-latest
    if: github.repository == 'textrefs/textrefs.org'
    steps:
      - uses: actions/checkout@v6
        with:
          submodules: recursive

      # Extract external URLs directly from sources instead of crawling the
      # built site. The rendered site expands each citation system into one
      # resolver URL per reference (~200k URLs for ~10 works), but every
      # distinct target already appears in the YAML source. Internal links,
      # anchors, and i18n are validated at build time by
      # starlight-links-validator, so this workflow only checks externals.
      - name: Collect external URLs from sources
        run: |
          mkdir -p lychee
          grep -rhoE 'https?://[^"'\''<>) ]+' data/ src/content/docs/ \
            | sed -E 's/[.,;:]+$//' \
            | sort -u > lychee/urls.txt
          echo "Collected $(wc -l < lychee/urls.txt) unique URLs"

      - name: Check links with lychee
        id: lychee
        uses: lycheeverse/lychee-action@v2
        with:
          args: --config lychee.toml lychee/urls.txt
          output: ./lychee/out.md
          fail: false

      # Report only: never fail CI. On main (or manual dispatch) open/update a
      # tracking issue when broken links are found. PR runs only show the job summary.
      - name: Create or update link report issue
        if: >-
          steps.lychee.outputs.exit_code != 0 &&
          (github.ref == 'refs/heads/main' || github.event_name == 'workflow_dispatch')
        uses: peter-evans/create-issue-from-file@v6
        with:
          title: Link Checker Report
          content-filepath: ./lychee/out.md
          labels: report, automated