Per-port mTLS on the same SNI hostname — supported, or does it require distinct hostnames? (v3.7.4, Kubernetes CRD)

I'm trying to serve one hostname (app.example.com) with different client-cert requirements per port: cert-free on :443 (websecure) for the browser/login routes, and RequireAndVerifyClientCert on a second entrypoint :8443 for a couple of mTLS-only paths. Everything is via the Kubernetes CRD provider (IngressRoute + TLSOption), Traefik v3.7.4.

The config resolves correctly — the API (/api/http/routers) shows each router bound to the option I expect (default on websecure, pilot-mtls on the 8443 entrypoint) — but at runtime Traefik logs Found different TLS options for routers on the same host, so using the default TLS options instead and serves no client-cert request on either port. A verbose TLS trace confirms :8443 never sends a CertificateRequest. I tried it both ways: per-router tls.options, and binding the option at the entrypoint level (--entrypoints.pilot.http.tls.options=...). Same result each time. It only starts requesting the client cert when I give the mTLS routes a distinct hostname that isn't also served on :443.

So my question: is different clientAuth for the same SNI hostname across two entrypoints actually supported, or is the effective TLS option computed per-hostname globally (across entrypoints), making a distinct hostname the only option? If the former, what am I missing — and does a catch-all PathPrefix(/) router on the :443 side (like #10271) interfere with option resolution for the same host on the other entrypoint?

I’m seeing the same limitation when the same SNI hostname is used with different TLS options across entrypoints. The fact that a distinct hostname makes the client-certificate request work points to hostname-level option resolution rather than the Kubernetes CRD binding.