Expand description
Enough HTTP/1.1 to be an honest gateway, and no more.
Same reasoning as everywhere else in this workspace: probe.rs writes a raw
GET because a health check is a status line; hearth-pull drives curl
because a registry download needs TLS and resume, which is a real library’s
job. This is the third case — a server on localhost speaking to clients that
already exist, where the whole surface is read a request, write a response,
and relay bytes. A framework would bring an async runtime and a dependency
tree to do what four hundred lines of std does, and the parsing is the only
part with a rule in it, so it lives here, pure and tested.
What this deliberately does NOT do: TLS (put it behind a proxy, and the default bind is loopback so there is nothing to expose), HTTP/2, or keep-alive pipelining. A model gateway handles a few requests per second that each take seconds; connection setup is not the cost.
Structs§
- Request
- A parsed request. Headers are lowercased on the way in, because HTTP header names are case-insensitive and half of the clients disagree about casing.
- Server
- The listening socket.
Functions§
- parse_
header - Parse one header line into a lowercased name and a trimmed value.
- parse_
request_ line - Parse a request line. Pure —
GET /v1/models HTTP/1.1. - proxy
- Forward a request to a model’s endpoint and relay the response back.
- read_
request - Read one request off a stream.
- render_
response - Serialize a response we generated ourselves.