Crafter.Network/implementations/Crafter.Network-ClientHTTP1.cpp

275 lines
11 KiB
C++
Raw Normal View History

feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
//SPDX-License-Identifier: LGPL-3.0-only
//SPDX-FileCopyrightText: Copyright (C) 2026 Catcrafts®
module;
#include <cerrno>
module Crafter.Network:ClientHTTP1_impl;
import :ClientHTTP1;
import :ClientTCP;
import :HTTP;
import :HTTP1;
feat(tls): add a libssl TLS transport and an https:// HTTP/1.1 stack HTTP/1.1 was plaintext-only, which left `https://` to either an HTTP/3 listener or a terminating proxy in front. Neither helps the callers this stack exists for — curl scripts, CI tooling, old proxies — so wrap the transport in libssl instead. Two new partitions: :Stream a ByteStream with per-call deadlines on both directions, plus the plaintext socket implementation. The HTTP/1.1 client and listener now hold a ByteStream& and never learn which transport they have, which is what lets one code path serve both schemes. :TLS TLSContext/TLSStream over OpenSSL 3, with credentials for both roles: chain and hostname verification on by default, private trust anchors, client certificates, mutual TLS, ALPN, and an in-process self-signed certificate for development. Both descriptors go non-blocking and every read and write is driven by poll() against a deadline. That is required for TLS — a blocking descriptor cannot express a handshake timeout — and it means a plaintext write can now time out too, instead of parking forever against a peer that stopped reading. ClientHTTP1 and ListenerHTTP1 gain credential-taking constructors; the existing ones still speak http://. The listener handshakes on the connection's own thread, so a peer that stalls mid-handshake costs one thread rather than the accept loop, and a failed handshake is counted rather than logged — on a public port it is ordinary traffic. MessageParser gains SetDefaultScheme so origin-form targets report the scheme the transport actually used; handlers shared with ListenerHTTP now see the same "https" they would over HTTP/3. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 20:13:30 +00:00
import :Stream;
import :TLS;
import :URL;
feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
import Crafter.Thread;
import std;
using namespace Crafter;
namespace {
// Host header value. The port is elided when it is the scheme default,
// which is what every other HTTP/1.1 client on the wire does and what
// virtual-host matching on the far side tends to expect.
feat(tls): add a libssl TLS transport and an https:// HTTP/1.1 stack HTTP/1.1 was plaintext-only, which left `https://` to either an HTTP/3 listener or a terminating proxy in front. Neither helps the callers this stack exists for — curl scripts, CI tooling, old proxies — so wrap the transport in libssl instead. Two new partitions: :Stream a ByteStream with per-call deadlines on both directions, plus the plaintext socket implementation. The HTTP/1.1 client and listener now hold a ByteStream& and never learn which transport they have, which is what lets one code path serve both schemes. :TLS TLSContext/TLSStream over OpenSSL 3, with credentials for both roles: chain and hostname verification on by default, private trust anchors, client certificates, mutual TLS, ALPN, and an in-process self-signed certificate for development. Both descriptors go non-blocking and every read and write is driven by poll() against a deadline. That is required for TLS — a blocking descriptor cannot express a handshake timeout — and it means a plaintext write can now time out too, instead of parking forever against a peer that stopped reading. ClientHTTP1 and ListenerHTTP1 gain credential-taking constructors; the existing ones still speak http://. The listener handshakes on the connection's own thread, so a peer that stalls mid-handshake costs one thread rather than the accept loop, and a failed handshake is counted rather than logged — on a public port it is ordinary traffic. MessageParser gains SetDefaultScheme so origin-form targets report the scheme the transport actually used; handlers shared with ListenerHTTP now see the same "https" they would over HTTP/3. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 20:13:30 +00:00
std::string DefaultAuthority(const std::string& host, std::uint16_t port, bool secure) {
// An IPv6 literal goes in brackets, or its colons read as the port's.
const std::string name = host.contains(':') ? "[" + host + "]" : host;
if (port == (secure ? 443 : 80)) return name;
return name + ":" + std::to_string(port);
feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
}
}
struct ClientHTTP1::Impl {
std::string host;
std::uint16_t port;
feat(tls): add a libssl TLS transport and an https:// HTTP/1.1 stack HTTP/1.1 was plaintext-only, which left `https://` to either an HTTP/3 listener or a terminating proxy in front. Neither helps the callers this stack exists for — curl scripts, CI tooling, old proxies — so wrap the transport in libssl instead. Two new partitions: :Stream a ByteStream with per-call deadlines on both directions, plus the plaintext socket implementation. The HTTP/1.1 client and listener now hold a ByteStream& and never learn which transport they have, which is what lets one code path serve both schemes. :TLS TLSContext/TLSStream over OpenSSL 3, with credentials for both roles: chain and hostname verification on by default, private trust anchors, client certificates, mutual TLS, ALPN, and an in-process self-signed certificate for development. Both descriptors go non-blocking and every read and write is driven by poll() against a deadline. That is required for TLS — a blocking descriptor cannot express a handshake timeout — and it means a plaintext write can now time out too, instead of parking forever against a peer that stopped reading. ClientHTTP1 and ListenerHTTP1 gain credential-taking constructors; the existing ones still speak http://. The listener handshakes on the connection's own thread, so a peer that stalls mid-handshake costs one thread rather than the accept loop, and a failed handshake is counted rather than logged — on a public port it is ordinary traffic. MessageParser gains SetDefaultScheme so origin-form targets report the scheme the transport actually used; handlers shared with ListenerHTTP now see the same "https" they would over HTTP/3. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 20:13:30 +00:00
// Null for plaintext; shared by every connection this client dials, so the
// trust store is parsed once rather than per redial.
std::shared_ptr<TLSContext> tls;
feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
std::unique_ptr<ClientTCP> tcp;
feat(tls): add a libssl TLS transport and an https:// HTTP/1.1 stack HTTP/1.1 was plaintext-only, which left `https://` to either an HTTP/3 listener or a terminating proxy in front. Neither helps the callers this stack exists for — curl scripts, CI tooling, old proxies — so wrap the transport in libssl instead. Two new partitions: :Stream a ByteStream with per-call deadlines on both directions, plus the plaintext socket implementation. The HTTP/1.1 client and listener now hold a ByteStream& and never learn which transport they have, which is what lets one code path serve both schemes. :TLS TLSContext/TLSStream over OpenSSL 3, with credentials for both roles: chain and hostname verification on by default, private trust anchors, client certificates, mutual TLS, ALPN, and an in-process self-signed certificate for development. Both descriptors go non-blocking and every read and write is driven by poll() against a deadline. That is required for TLS — a blocking descriptor cannot express a handshake timeout — and it means a plaintext write can now time out too, instead of parking forever against a peer that stopped reading. ClientHTTP1 and ListenerHTTP1 gain credential-taking constructors; the existing ones still speak http://. The listener handshakes on the connection's own thread, so a peer that stalls mid-handshake costs one thread rather than the accept loop, and a failed handshake is counted rather than logged — on a public port it is ordinary traffic. MessageParser gains SetDefaultScheme so origin-form targets report the scheme the transport actually used; handlers shared with ListenerHTTP now see the same "https" they would over HTTP/3. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 20:13:30 +00:00
// Sits on top of `tcp` and must therefore be destroyed before it.
std::unique_ptr<ByteStream> stream;
std::string protocol;
feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
void Connect(std::chrono::milliseconds connectTimeout, std::chrono::milliseconds handshakeTimeout) {
tcp = std::make_unique<ClientTCP>(host, port, connectTimeout);
feat(tls): add a libssl TLS transport and an https:// HTTP/1.1 stack HTTP/1.1 was plaintext-only, which left `https://` to either an HTTP/3 listener or a terminating proxy in front. Neither helps the callers this stack exists for — curl scripts, CI tooling, old proxies — so wrap the transport in libssl instead. Two new partitions: :Stream a ByteStream with per-call deadlines on both directions, plus the plaintext socket implementation. The HTTP/1.1 client and listener now hold a ByteStream& and never learn which transport they have, which is what lets one code path serve both schemes. :TLS TLSContext/TLSStream over OpenSSL 3, with credentials for both roles: chain and hostname verification on by default, private trust anchors, client certificates, mutual TLS, ALPN, and an in-process self-signed certificate for development. Both descriptors go non-blocking and every read and write is driven by poll() against a deadline. That is required for TLS — a blocking descriptor cannot express a handshake timeout — and it means a plaintext write can now time out too, instead of parking forever against a peer that stopped reading. ClientHTTP1 and ListenerHTTP1 gain credential-taking constructors; the existing ones still speak http://. The listener handshakes on the connection's own thread, so a peer that stalls mid-handshake costs one thread rather than the accept loop, and a failed handshake is counted rather than logged — on a public port it is ordinary traffic. MessageParser gains SetDefaultScheme so origin-form targets report the scheme the transport actually used; handlers shared with ListenerHTTP now see the same "https" they would over HTTP/3. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 20:13:30 +00:00
try {
if (tls) {
stream = TLSStream::Connect(tcp->socketid, tls, host, handshakeTimeout);
} else {
stream = std::make_unique<PlainStream>(tcp->socketid);
}
} catch (...) {
// A half-built connection must not be left pooled: the next Send()
// would treat it as reusable and read from a socket with no
// session on it.
tcp.reset();
throw;
}
protocol = std::string(stream->Protocol());
feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
}
void Close() {
feat(tls): add a libssl TLS transport and an https:// HTTP/1.1 stack HTTP/1.1 was plaintext-only, which left `https://` to either an HTTP/3 listener or a terminating proxy in front. Neither helps the callers this stack exists for — curl scripts, CI tooling, old proxies — so wrap the transport in libssl instead. Two new partitions: :Stream a ByteStream with per-call deadlines on both directions, plus the plaintext socket implementation. The HTTP/1.1 client and listener now hold a ByteStream& and never learn which transport they have, which is what lets one code path serve both schemes. :TLS TLSContext/TLSStream over OpenSSL 3, with credentials for both roles: chain and hostname verification on by default, private trust anchors, client certificates, mutual TLS, ALPN, and an in-process self-signed certificate for development. Both descriptors go non-blocking and every read and write is driven by poll() against a deadline. That is required for TLS — a blocking descriptor cannot express a handshake timeout — and it means a plaintext write can now time out too, instead of parking forever against a peer that stopped reading. ClientHTTP1 and ListenerHTTP1 gain credential-taking constructors; the existing ones still speak http://. The listener handshakes on the connection's own thread, so a peer that stalls mid-handshake costs one thread rather than the accept loop, and a failed handshake is counted rather than logged — on a public port it is ordinary traffic. MessageParser gains SetDefaultScheme so origin-form targets report the scheme the transport actually used; handlers shared with ListenerHTTP now see the same "https" they would over HTTP/3. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 20:13:30 +00:00
stream.reset();
feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
tcp.reset();
}
// One request/response exchange on the current connection. `received`
// is set as soon as any response byte arrives, which is what decides
// whether a failure is safe to replay on a new connection.
HTTPResponse Exchange(const std::string& wire, std::string_view method,
const HTTP1::MessageLimits& limits,
std::chrono::milliseconds timeout, bool& received) {
feat(tls): add a libssl TLS transport and an https:// HTTP/1.1 stack HTTP/1.1 was plaintext-only, which left `https://` to either an HTTP/3 listener or a terminating proxy in front. Neither helps the callers this stack exists for — curl scripts, CI tooling, old proxies — so wrap the transport in libssl instead. Two new partitions: :Stream a ByteStream with per-call deadlines on both directions, plus the plaintext socket implementation. The HTTP/1.1 client and listener now hold a ByteStream& and never learn which transport they have, which is what lets one code path serve both schemes. :TLS TLSContext/TLSStream over OpenSSL 3, with credentials for both roles: chain and hostname verification on by default, private trust anchors, client certificates, mutual TLS, ALPN, and an in-process self-signed certificate for development. Both descriptors go non-blocking and every read and write is driven by poll() against a deadline. That is required for TLS — a blocking descriptor cannot express a handshake timeout — and it means a plaintext write can now time out too, instead of parking forever against a peer that stopped reading. ClientHTTP1 and ListenerHTTP1 gain credential-taking constructors; the existing ones still speak http://. The listener handshakes on the connection's own thread, so a peer that stalls mid-handshake costs one thread rather than the accept loop, and a failed handshake is counted rather than logged — on a public port it is ordinary traffic. MessageParser gains SetDefaultScheme so origin-form targets report the scheme the transport actually used; handlers shared with ListenerHTTP now see the same "https" they would over HTTP/3. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 20:13:30 +00:00
stream->Write(wire.data(), wire.size(), timeout);
feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
HTTP1::MessageParser parser(HTTP1::MessageKind::Response, limits);
parser.SetRequestMethod(method);
std::vector<char> chunk(16 * 1024);
while (!parser.Complete()) {
feat(tls): add a libssl TLS transport and an https:// HTTP/1.1 stack HTTP/1.1 was plaintext-only, which left `https://` to either an HTTP/3 listener or a terminating proxy in front. Neither helps the callers this stack exists for — curl scripts, CI tooling, old proxies — so wrap the transport in libssl instead. Two new partitions: :Stream a ByteStream with per-call deadlines on both directions, plus the plaintext socket implementation. The HTTP/1.1 client and listener now hold a ByteStream& and never learn which transport they have, which is what lets one code path serve both schemes. :TLS TLSContext/TLSStream over OpenSSL 3, with credentials for both roles: chain and hostname verification on by default, private trust anchors, client certificates, mutual TLS, ALPN, and an in-process self-signed certificate for development. Both descriptors go non-blocking and every read and write is driven by poll() against a deadline. That is required for TLS — a blocking descriptor cannot express a handshake timeout — and it means a plaintext write can now time out too, instead of parking forever against a peer that stopped reading. ClientHTTP1 and ListenerHTTP1 gain credential-taking constructors; the existing ones still speak http://. The listener handshakes on the connection's own thread, so a peer that stalls mid-handshake costs one thread rather than the accept loop, and a failed handshake is counted rather than logged — on a public port it is ordinary traffic. MessageParser gains SetDefaultScheme so origin-form targets report the scheme the transport actually used; handlers shared with ListenerHTTP now see the same "https" they would over HTTP/3. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 20:13:30 +00:00
std::size_t read = 0;
const StreamStatus status = stream->ReadSome(chunk.data(), chunk.size(),
timeout, read);
if (status == StreamStatus::TimedOut) {
throw std::runtime_error("timed out waiting for the response");
}
if (status == StreamStatus::Closed) {
feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
// Peer closed. Completes a response framed by close;
// anything else throws out of Finish().
parser.Finish();
break;
}
received = true;
parser.Feed(chunk.data(), read);
}
if (!parser.Complete()) {
throw std::runtime_error("connection closed before a complete response arrived");
}
HTTPResponse response = parser.TakeResponse();
if (!parser.KeepAlive()) Close();
return response;
}
};
ClientHTTP1::ClientHTTP1(const char* host, std::uint16_t port)
: host(host), port(port), impl(std::make_unique<Impl>(std::string(host), port)) {}
ClientHTTP1::ClientHTTP1(std::string host, std::uint16_t port)
: ClientHTTP1(host.c_str(), port) {}
feat(tls): add a libssl TLS transport and an https:// HTTP/1.1 stack HTTP/1.1 was plaintext-only, which left `https://` to either an HTTP/3 listener or a terminating proxy in front. Neither helps the callers this stack exists for — curl scripts, CI tooling, old proxies — so wrap the transport in libssl instead. Two new partitions: :Stream a ByteStream with per-call deadlines on both directions, plus the plaintext socket implementation. The HTTP/1.1 client and listener now hold a ByteStream& and never learn which transport they have, which is what lets one code path serve both schemes. :TLS TLSContext/TLSStream over OpenSSL 3, with credentials for both roles: chain and hostname verification on by default, private trust anchors, client certificates, mutual TLS, ALPN, and an in-process self-signed certificate for development. Both descriptors go non-blocking and every read and write is driven by poll() against a deadline. That is required for TLS — a blocking descriptor cannot express a handshake timeout — and it means a plaintext write can now time out too, instead of parking forever against a peer that stopped reading. ClientHTTP1 and ListenerHTTP1 gain credential-taking constructors; the existing ones still speak http://. The listener handshakes on the connection's own thread, so a peer that stalls mid-handshake costs one thread rather than the accept loop, and a failed handshake is counted rather than logged — on a public port it is ordinary traffic. MessageParser gains SetDefaultScheme so origin-form targets report the scheme the transport actually used; handlers shared with ListenerHTTP now see the same "https" they would over HTTP/3. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 20:13:30 +00:00
ClientHTTP1::ClientHTTP1(const char* host, std::uint16_t port, TLSClientCredentials credentials)
: host(host), port(port), impl(std::make_unique<Impl>(std::string(host), port)) {
// Built here rather than on first Send() so bad credentials — an
// unreadable certificate, a trust anchor that is not a certificate —
// surface at construction, where the caller is still looking.
impl->tls = TLSContext::Client(credentials);
}
ClientHTTP1::ClientHTTP1(std::string host, std::uint16_t port, TLSClientCredentials credentials)
: ClientHTTP1(host.c_str(), port, std::move(credentials)) {}
feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
ClientHTTP1::ClientHTTP1(ClientHTTP1&&) noexcept = default;
ClientHTTP1::~ClientHTTP1() = default;
bool ClientHTTP1::Connected() const noexcept {
return impl && impl->tcp != nullptr;
}
feat(tls): add a libssl TLS transport and an https:// HTTP/1.1 stack HTTP/1.1 was plaintext-only, which left `https://` to either an HTTP/3 listener or a terminating proxy in front. Neither helps the callers this stack exists for — curl scripts, CI tooling, old proxies — so wrap the transport in libssl instead. Two new partitions: :Stream a ByteStream with per-call deadlines on both directions, plus the plaintext socket implementation. The HTTP/1.1 client and listener now hold a ByteStream& and never learn which transport they have, which is what lets one code path serve both schemes. :TLS TLSContext/TLSStream over OpenSSL 3, with credentials for both roles: chain and hostname verification on by default, private trust anchors, client certificates, mutual TLS, ALPN, and an in-process self-signed certificate for development. Both descriptors go non-blocking and every read and write is driven by poll() against a deadline. That is required for TLS — a blocking descriptor cannot express a handshake timeout — and it means a plaintext write can now time out too, instead of parking forever against a peer that stopped reading. ClientHTTP1 and ListenerHTTP1 gain credential-taking constructors; the existing ones still speak http://. The listener handshakes on the connection's own thread, so a peer that stalls mid-handshake costs one thread rather than the accept loop, and a failed handshake is counted rather than logged — on a public port it is ordinary traffic. MessageParser gains SetDefaultScheme so origin-form targets report the scheme the transport actually used; handlers shared with ListenerHTTP now see the same "https" they would over HTTP/3. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 20:13:30 +00:00
bool ClientHTTP1::Secure() const noexcept {
return impl && impl->tls != nullptr;
}
std::string_view ClientHTTP1::Protocol() const noexcept {
return impl ? std::string_view(impl->protocol) : std::string_view();
}
feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
void ClientHTTP1::Disconnect() {
if (impl) impl->Close();
}
HTTPResponse ClientHTTP1::Send(const HTTPRequest& request) {
HTTPRequest prepared = request;
feat(tls): add a libssl TLS transport and an https:// HTTP/1.1 stack HTTP/1.1 was plaintext-only, which left `https://` to either an HTTP/3 listener or a terminating proxy in front. Neither helps the callers this stack exists for — curl scripts, CI tooling, old proxies — so wrap the transport in libssl instead. Two new partitions: :Stream a ByteStream with per-call deadlines on both directions, plus the plaintext socket implementation. The HTTP/1.1 client and listener now hold a ByteStream& and never learn which transport they have, which is what lets one code path serve both schemes. :TLS TLSContext/TLSStream over OpenSSL 3, with credentials for both roles: chain and hostname verification on by default, private trust anchors, client certificates, mutual TLS, ALPN, and an in-process self-signed certificate for development. Both descriptors go non-blocking and every read and write is driven by poll() against a deadline. That is required for TLS — a blocking descriptor cannot express a handshake timeout — and it means a plaintext write can now time out too, instead of parking forever against a peer that stopped reading. ClientHTTP1 and ListenerHTTP1 gain credential-taking constructors; the existing ones still speak http://. The listener handshakes on the connection's own thread, so a peer that stalls mid-handshake costs one thread rather than the accept loop, and a failed handshake is counted rather than logged — on a public port it is ordinary traffic. MessageParser gains SetDefaultScheme so origin-form targets report the scheme the transport actually used; handlers shared with ListenerHTTP now see the same "https" they would over HTTP/3. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-28 20:13:30 +00:00
if (prepared.authority.empty()) {
prepared.authority = DefaultAuthority(host, port, Secure());
}
feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
const std::string wire = HTTP1::SerializeRequest(prepared);
const std::string method = prepared.method.empty() ? std::string("GET") : prepared.method;
// Two attempts at most, and the second one only for the keep-alive
// race: the peer may have closed a pooled connection at the same moment
// we wrote to it, which is indistinguishable from success until the
// read fails. Nothing is replayed once a response byte has arrived.
for (int attempt = 0;; ++attempt) {
const bool reused = impl->tcp != nullptr;
if (!reused) impl->Connect(connectTimeout, handshakeTimeout);
feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
bool received = false;
try {
return impl->Exchange(wire, method, limits, timeout, received);
feat(http1): add an HTTP/1.1 client and listener HTTP/3-only is not a deployable position yet: plenty of clients, proxies and CI tooling still speak nothing but HTTP/1.1. This adds that path using the request/response types the HTTP/3 stack already uses, so a route handler or call site moves between the two protocols by changing the class name. - :HTTP1 — transport-free wire format. Serialisation with the framing headers owned by the serialiser, and an incremental parser that takes arbitrary socket chunks and yields one message at a time: keep-alive, pipelining, content-length and chunked bodies (with trailers), read-to-EOF responses, interim 1xx skipping, HEAD/204/304 framing and Expect: 100-continue. Ambiguous framing is rejected rather than guessed at (content-length with transfer-encoding, disagreeing content-lengths, whitespace before a colon), and CR/LF in a value we are asked to serialise is refused. - ClientHTTP1 — persistent connection, redialling once when a pooled connection turns out to have been closed by the peer, which is the race HTTP/1.1 keep-alive cannot avoid. Nothing is replayed after a response byte has arrived. - ListenerHTTP1 — one thread per connection (keep-alive connections are idle most of their life and would pin every ThreadPool thread), automatic Date, HEAD, 100-continue, handler-requested close, idle and request timeouts, and 400/404/500 responses. Routes fall back to the query-stripped path so `/thing?x=1` reaches the handler for `/thing`. No TLS: this is `http://` only. Encrypted traffic still goes over HTTP/3, or through a TLS-terminating proxy. Tests: codec unit tests including the malformed inputs above, a client/server round-trip, keep-alive and stale-connection recovery, a 10 MiB body both ways, and interop both directions against curl and python3's http.server (skipped when those are not installed). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-27 00:45:09 +00:00
} catch (const std::exception&) {
impl->Close();
if (attempt == 0 && reused && !received) continue;
throw;
}
}
}
void ClientHTTP1::SendAsync(const HTTPRequest& request,
std::function<void(HTTPResponse)> onSuccess,
std::function<void(std::string)> onError) {
HTTPRequest copy = request;
ThreadPool::Enqueue([this, copy = std::move(copy),
onSuccess = std::move(onSuccess),
onError = std::move(onError)]() mutable {
try {
HTTPResponse response = this->Send(copy);
if (onSuccess) onSuccess(std::move(response));
} catch (const std::exception& e) {
if (onError) onError(e.what());
} catch (...) {
if (onError) onError("unknown error");
}
});
}
namespace {
std::int32_t StatusCode(const HTTPResponse& response) {
std::int32_t code = 0;
std::from_chars(response.status.data(), response.status.data() + response.status.size(), code);
return code;
}
bool IsRedirect(std::int32_t code) {
return code == 301 || code == 302 || code == 303 || code == 307 || code == 308;
}
}
FetchResult ClientHTTP1::Fetch(std::string_view url, const FetchOptions& options) {
URL current = URL::Parse(url);
if (current.scheme != "http" && current.scheme != "https") {
throw std::invalid_argument("ClientHTTP1::Fetch: not an http:// or https:// URL: " + std::string(url));
}
std::string method = options.method.empty() ? std::string("GET") : options.method;
std::unordered_map<std::string, std::string> headers = options.headers;
std::string body = options.body;
// Every URL the chain has been sent to: a redirect back to one of them
// would loop.
std::set<std::string> visited { current.ToString() };
FetchResult result;
while (true) {
// A client per hop: one ClientHTTP1 is one origin, and a redirect
// may leave it. Destroying the client closes the hop's connection
// once its response, body included, has been read.
ClientHTTP1 client = current.Secure() ? ClientHTTP1(current.host, current.PortOrDefault(), options.credentials)
: ClientHTTP1(current.host, current.PortOrDefault());
client.connectTimeout = options.connectTimeout;
client.handshakeTimeout = options.handshakeTimeout;
client.timeout = options.timeout;
client.limits = options.limits;
HTTPRequest request;
request.method = method;
request.scheme = current.scheme;
request.authority = current.Authority();
request.path = current.Target();
request.headers = headers;
request.body = body;
result.response = client.Send(request);
result.url = current;
const std::int32_t code = StatusCode(result.response);
if (!IsRedirect(code)) return result;
const auto location = result.response.headers.find("location");
if (location == result.response.headers.end()) {
result.stop = FetchStop::NoLocation;
return result;
}
URL next;
try {
next = current.Resolve(location->second);
} catch (const std::invalid_argument&) {
result.stop = FetchStop::NoLocation;
return result;
}
if (next.scheme != "http" && next.scheme != "https") {
result.stop = FetchStop::OtherScheme;
return result;
}
if (result.redirects >= options.maxRedirects) {
result.stop = FetchStop::Limit;
return result;
}
if (!visited.insert(next.ToString()).second) {
result.stop = FetchStop::Loop;
return result;
}
++result.redirects;
// RFC 9110 §15.4: a 303 asks for a GET of the Location; a 301 or 302
// keeps the method, but a POST becomes a GET, as the user agents the
// sections describe have always done; a 307 or 308 keeps the method
// and the body. A request that turned into a GET loses the body and
// the headers that described it.
const bool toGet = (code == 303 && method != "HEAD") || ((code == 301 || code == 302) && method == "POST");
if (toGet) {
method = "GET";
body.clear();
std::erase_if(headers, [](const auto& header) { return header.first.starts_with("content-"); });
}
// Credentials stay with the origin they were meant for.
if (!next.SameOrigin(current)) {
headers.erase("authorization");
headers.erase("proxy-authorization");
headers.erase("cookie");
}
current = std::move(next);
}
}