iili: Caddy answers short links at i.iili.uk/r/{slug}

[aaylward]

Short links mint as i.iili.uk/r/{slug} now: Caddy on the consolidated box rewrites those onto r3dr_v2, and the Cloudflare Worker at iili.uk is SPA-only. Pulled the old r3dr.net / www.r3dr.net Caddy sites (Go service still in compose until chunk 3).

Deploy config tests pin the new host and the gone ones; iili_web unit tests cover the mint base. After deploy, curl a known slug on i.iili.uk and mint once from the SPA to confirm.

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	1d4-web	07088b8	Commit Preview URL Branch Preview URL	Aug 21 2026, 11:30 PM

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	iili	07088b8	Commit Preview URL Branch Preview URL	Aug 21 2026, 11:30 PM

[aaylward]

Single-pass code review at xhigh effort (the Agent tool was unavailable in this context, so this is one reviewer working all angles in sequence — not the usual multi-agent fan-out with an independent verify pass). 6 findings, ranked most-severe first in the inline comments.

Headline: the method GET HEAD matcher forwards HEAD to r3dr_v2, but the smithy-cpp router (verified at the pinned 46bcd4d) does exact method-bucket dispatch and the model registers only GET — so HEAD i.iili.uk/r/{slug} answers 405 Allow: GET, not the 302 the PR's comments, READMEs, and the deleted worker test all promise. Verified against the router source; a Send("HEAD", ...) wire_test would pin it.

Also verified clean: the bare respond 404 after handle is correct under Caddy's directive order; rewrite * /r3dr/v2{path} preserves the query string; trailing-slash tolerance is re-established by the router's NormalizedTarget (so that deleted worker test's invariant survives); the {path} braces don't break the Go tests' depth counting; compose wiring (service name, TRUSTED_PROXY_CIDRS, db init) checks out; localStorage recent links self-heal to the new base.

---

Generated by Claude Code

[aaylward]

Folded review follow-ups in 07088b8:

• add scala 3 build support #1 HEAD → 405: filed #1433; Caddy still forwards HEAD, comments point at the issue.
• improve scala support #2 old iili.uk/r/{slug}: skipped (none minted); noted in Worker + README.
• try bazel_deps #3–IntelliJ? 🍕 #6: substring host guard, r3dr_v2:8091 pin, compose orphan cleanup, docs corrected.

[aaylward]

Rechecked 07088b8 against the review — all six findings verified addressed. Checked out the head, re-read the diff, and ran the deploy config suite (go test ./deploy/consolidated → ok; gofmt clean).

• add scala 3 build support #1 HEAD → 405: deferred honestly. r3dr_v2: HEAD /r3dr/v2/r/{slug} returns 405, not 302 #1433 exists and states the mechanism correctly (exact method-bucket dispatch at smithy-cpp pin 46bcd4d, no HEAD fallback); the Caddyfile comment, test comment, and both READMEs now say 405-today instead of claiming HEAD works.
• improve scala support #2 mint window: closed as attested — "no iili.uk/r/{slug} links minted" is stated in the Worker header and README, which is the note the review asked for. (Author-knowledge claim; not independently verifiable from the repo.)
• try bazel_deps #3 substring guard: verified. strings.Contains over directiveLines, which strips full-line and inline comments before matching, so the guard can't false-positive on comments; no remaining line in the Caddyfile contains r3dr.net. The www/r3dr.net double-report on a hit is cosmetic.
• fix readme grammar #4 port pin: verified. Test requires reverse_proxy r3dr_v2:8091 verbatim and the i.iili.uk block has it (Caddyfile line 184); suite passes.
• srsly? 🤦 #5 compose orphans: verified removed (/var/www/r3dr mount, depends_on: r3dr), and the README inventory now lists the internal-only r3dr service. One crumb of the same class remains: deploy.sh:355 still scps r3dr_web assets to ~/r3dr-assets and deploy.sh:379-380 copies them into /var/www/r3dr, which nothing mounts or serves anymore. Harmless dead work each deploy; fine to leave for chunk 3 with the files themselves, just flagging so chunk 3's sweep includes it.
• IntelliJ? 🍕 #6 docs: verified. Both READMEs now say the r3dr.net frontage and its links died with this change and only the service body waits for chunk 3.

CI on 07088b8: everything completed is green (test-deploy, test-iili-web, format-check included); sanitizers and build-and-test still running at time of writing.

---

Generated by Claude Code