[Madraka]

RFC-0038 / ADR-0040, step 2 of 3. Refs #50.

Redirects on change

• SiteEngine.update → store/site-permalinks.ts. For each collection whose effective template changes, a RedirectPlan.move(oldPrefix, newPrefix, { match: "prefix", reason: "slug-change" }) is emitted into an OpBatch anchored on the site's new hash (OpBatch.anchorOn), in the same atomic batch as the site CAS (rule 4).
• Locales: redirects are written for the default locale and each other site locale. Entries exist in every locale; taxonomy only writes the default one (see open questions).
• Chains and loops: chains collapse (/posts → /blog → /journal sends /posts/x straight to /journal/x), and moving back to the default drops the self-redirect.
• Conflicts: a stale rev throws ConflictError before planning. A failed op_guards assertion counts as a missed CAS, so nothing is written.

Overlap checks, both ways (permalink-paths.ts)

• A changed template must not overlap: another collection's prefix (configured, or the default of a collection with entries), any taxonomy archive, a page at or under it, media, /search, or a locale prefix. The result is path_conflict / reserved_path, guarded on the CAS (noArchiveOverlapping, noPageWithin).
• Reverse direction: pages and taxonomy archives can no longer be created or moved inside a configured prefix (checks plus site-row SQL guards). A default /<collection> prefix keeps today's rule, where a page wins.

Search

planSearch rewrites every entry row when a published snapshot's templates differ from the indexed one.

Tests (one package at a time)

• core-ts 651/651: the new testing-permalinks.ts suite runs on sqlite, Miniflare D1 and Durable Object SQLite. It covers the moves, chains, back to default, overlaps, the reverse checks, stale and racing writers, and search hrefs.
• astro-renderer 256/256: a new routing test serves /blog/hello-world and 301s /posts/hello-world?ref=x → /blog/hello-world?ref=x, on live and snapshot sources.
• api 371/371.
• taxonomy.ts shrank, so the size baseline is lowered.

Open questions (follow-ups, not blockers)

1. Entry writes still check archives against the default /<collection>, and creating an entry in a new collection whose default prefix overlaps another collection's configured prefix is not refused. In that case entryAt prefers the configured one.
2. Should updateTaxonomy write per-locale redirects too, so both behave the same?
3. A move into a sub-prefix of itself (/posts → /posts/archive) is refused by RedirectPlan's loop check as invalid_redirect, not path_conflict.

🤖 Generated with Claude Code