r3dr_v2: HEAD /r3dr/v2/r/{slug} returns 405, not 302

[aaylward]

Problem

i.iili.uk Caddy forwards GET and HEAD /r/{slug} to r3dr_v2 (rewrite → /r3dr/v2/r/{slug}). The model only registers @http(method: "GET", …). smithy-cpp's router does exact method-bucket dispatch with no HEAD→GET fallback, so:

curl -I https://i.iili.uk/r/<slug>
# → 405 Method Not Allowed, Allow: GET
# no Location

Unfurlers and probes that lead with HEAD never see the redirect. Surfaced on #1432.

Expected

HEAD should get the same status and Location as GET (body empty), matching common redirect/unfurl behavior.

Options

1. Register HEAD on the Redirect operation (or router HEAD→GET for redirect routes).
2. Caddy-side map HEAD→GET before proxy (workaround; prefer fixing the service).

A Send("HEAD", …) wire/contract test should pin the 302 + Location once fixed.

Related

• iili: Caddy answers short links at i.iili.uk/r/{slug} #1432 (i.iili.uk Caddy site already forwards HEAD)
• smithy-cpp router at pin 46bcd4d: routes_.find(request.method)

[aaylward]

Fixed in #1435. Note the route renamed in the same PR: the path is /iili/v1/r/{slug} now, and the service is iili, so the repro above reads curl -I https://i.iili.uk/r/AQA (the public host is unchanged).

WithHeadAsGet in domains/iili/apis/iili/smithy_handler.cc re-dispatches HEAD as the GET it mirrors and clears the body — content on a HEAD response desynchronizes a keep-alive connection. main.cc and both test harnesses share the one composition, so the wire test exercises what production runs. That is option 1's parenthetical ("router HEAD→GET for redirect routes"), not the Caddy workaround.

Pinned by HeadGetsTheSameRedirectAsGetWithNoBody and HeadOnAnUnknownSlugIsTheSame404StatusAsGet in wire_test.cc. Mutation-checked: removing the shim from the harness fails the first with Which is: 405, the exact defect described here.

Two things this does not cover, both tracked elsewhere:

• Not yet verified in production — merged, but the host deploy hasn't run, so the curl -I above still returns 405 until it does. Confirming the 302 is on the deploy checklist in iili: rename the shortener from r3dr, and answer HEAD (#1359, #1433) #1435.
• HEAD is still not in the model. Routing was never the obstacle; framing was. ToWireResponse frames without seeing the request method, so a modeled HEAD operation would put the generated {} body on the wire. Filed upstream as HEAD responses carry a body: the transport frames without regard to request method smithy-cpp#192, with iili: model HEAD and delete WithHeadAsGet once smithy-cpp frames HEAD responses #1437 to model it here and delete the shim once the pin picks that up.

---

Generated by Claude Code