Generate redirects and sitemap exclusions from front matter

[benbalter]

Why

redirect_from, redirect_to, and sitemap: false are in the content schema, but nothing read them. Redirects lived in a hand-maintained public/_redirects, and sitemap exclusions were a hardcoded list in astro.config.mjs. The two were kept in sync by hand. This PR makes front matter the source of truth.

Approach

• src/utils/front-matter-routes.ts (plus tests) reads src/content/{posts,pages} front matter with gray-matter, which is already a devDependency. It maps each entry to its URL: posts go through getPostUrl, and pages use permalink or /id/. From that it derives:
  • Sitemap exclusions: sitemap: false, plus redirect_to entries, since those URLs 301 off-site.
  • Redirect rules: redirect_from -> entry URL 301 and entry URL -> redirect_to 301. Each directory-style source is emitted in both /foo/ and /foo forms, matching the existing convention.
  • A throw if a redirect_from shadows a real page.
• astro.config.mjs
  • Loads the entries once at config load.
  • Feeds the exclusions into the @astrojs/sitemap filter. The non-content routes stay hardcoded: 404, _not-found, fine-print, resume/linkedin, and resume/print.
  • Adds a small front-matter-redirects integration that appends the generated rules to dist-astro/_redirects in astro:build:done. It fails the build if a source appears in both public/_redirects and front matter.
• public/_redirects keeps only the non-content rules: feed, sitemap, retired pages, and legacy slug fixes with no front matter.
• CI: astro-e2e.yml and build-and-deploy.yml now run script/validate-redirects dist-astro/_redirects after the build. The pre-build check on public/_redirects in ci.yml is unchanged.
• e2e: redirects.spec.ts and the redirect block in pages.spec.ts now read the built file, lazily.
• post-metadata.ts and the schema are untouched, so isExternal via redirect_to still works.

Before/after: effective redirect rules

These are the sorted non-comment rules in dist-astro/_redirects, from a build of main compared with this branch.

Identical: 63 rules both ways, empty diff.

Of those, 22 lines moved out of public/_redirects and into generated output. None were added and none were dropped. The moved lines are:

• Page renames (8 lines): /books, /books-for-geeks, /recommended-reading, and /cv, from other-recommended-reading.md and resume.md.
• Syndicated posts (6 lines): the three redirect_to posts.
• Slug corrections (8 lines): rules-of-communicating-at-github, types-of-pull-requests, eight-things-i-wish-i-knew-my-first-week, and moderating-open-source-conversations.

script/validate-redirects dist-astro/_redirects reports 63 redirects and 0 issues, with no duplicate sources and no loops.

Before/after: sitemap URLs

204 -> 201. The only removals are the three redirect_to posts:

• /2012/04/23/enterprise-open-source-usage-is-up-but-challenges-remain/
• /2015/04/27/eight-lessons-learned-hacking-on-github-pages-for-six-months/
• /2023/10/04/how-to-communicate-like-a-github-engineer/

This is the one intentional behavior change. Those URLs 301 off-site in production (curl -I on the 2012 URL returns a 301 to TechCrunch), so they shouldn't be in the sitemap. No content file currently sets sitemap: false, so there are no other changes.

Follow-up (not in this PR)

About 11 legacy slug-correction posts have redirects in public/_redirects but no redirect_from in front matter. Examples are what-is-a-hacker, open-collabortion, source-disclosed, jekyll-collections, and cathedral-bazaar-management. Moving them into front matter would finish the migration. I left them alone to avoid touching post files here.

Verification

• npm run check: 0 errors.
• Vitest: 15 new tests in front-matter-routes.test.ts, plus post-metadata.test.ts, all pass.
• SKIP_PDF=1 npm run build: succeeds and logs Appended 22 front matter redirects to _redirects.
• Redirect and sitemap diffs against a baseline build of main, as above.
• script/validate-redirects passes on both public/_redirects and dist-astro/_redirects.
• Playwright: redirects.spec.ts and the pages.spec.ts redirect tests, 22 pass.
• eslint on the new files and knip are clean.

🤖 Generated with Claude Code