_init_socks5_connection ignores timeout — SyncStream.read() puts socket into infinite blocking mode

[easy6699]

When using a SOCKS5 proxy with a sync httpx.Client, requests can hang indefinitely despite a timeout being configured. The hang occurs during the SOCKS5 handshake, before any HTTP traffic is sent.

Root cause

_init_socks5_connection in socks_proxy.py calls stream.read() three times — for the auth method response, the username/password response, and the connect response — without passing a timeout argument:

# socks_proxy.py, lines 62, 81, 97
incoming_bytes = stream.read(max_bytes=4096)

SyncStream.read() in sync.py accepts timeout: float | None = None and calls self._sock.settimeout(timeout) before every recv():

# sync.py, line 127
def read(self, max_bytes: int, timeout: float | None = None) -> bytes:
    exc_map: ExceptionMapping = {socket.timeout: ReadTimeout, OSError: ReadError}
    with map_exceptions(exc_map):
        self._sock.settimeout(timeout)  # timeout is None here
        return self._sock.recv(max_bytes)

Calling settimeout(None) puts the socket into infinite blocking mode, overriding:

• any timeout the socket was created with via socket.create_connection(..., timeout=...)
• any process-wide default set via socket.setdefaulttimeout()

As a result, if a SOCKS5 proxy accepts the TCP connection but then stalls — a common occurrence with Tor circuit establishment — recv() blocks forever and the entire process hangs. The timeout passed to httpx.Client(timeout=...) has no effect because it is never forwarded to these stream.read() calls.

The async backend does not have this bug — _init_socks5_connection in socks_proxy.py passes timeout through correctly.

To reproduce

import httpx

# Connect to a SOCKS5 proxy that accepts the TCP connection but stalls
# on the handshake response (simulate with: nc -l 9150)
proxy = httpx.Proxy("socks5://127.0.0.1:9150")
with httpx.Client(proxy=proxy, timeout=5.0) as client:
    client.get("http://example.com")  # hangs forever, timeout has no effect

Expected behavior

The timeout value passed to httpx.Client (and forwarded via request.extensions["timeout"]["connect"]) should apply to all I/O during the SOCKS5 handshake, including the stream.read() calls in _init_socks5_connection. A ConnectTimeout should be raised after the configured timeout elapses.

Actual behavior

recv() blocks indefinitely. No exception is raised. Ctrl+C shows the process stuck at:

File "httpcore/_sync/socks_proxy.py", line 97, in _init_socks5_connection
    incoming_bytes = stream.read(max_bytes=4096)
File "httpcore/_backends/sync.py", line 128, in read
    return self._sock.recv(max_bytes)
KeyboardInterrupt

Suggested fix

_init_socks5_connection receives the stream but not the timeout. The timeout is available in Socks5Connection.handle_request as timeouts.get("connect", None). It should be passed through to _init_socks5_connection and forwarded to each stream.read() call:

# socks_proxy.py
def _init_socks5_connection(
    stream: NetworkStream,
    *,
    host: bytes,
    port: int,
    auth: tuple[bytes, bytes] | None = None,
    timeout: float | None = None,   # <-- add this
) -> None:
    ...
    incoming_bytes = stream.read(max_bytes=4096, timeout=timeout)  # all 3 call sites

And in handle_request:

kwargs = {
    "stream": stream,
    "host": ...,
    "port": ...,
    "auth": self._proxy_auth,
    "timeout": timeout,   # <-- forward it
}

Note: the async counterpart (socks_proxy.py) already passes timeout correctly and does not need fixing.

My Test Environment

httpcore: 1.0.9
httpx: 0.28.1
Python: 3.14.4
OS: Debian-based Linux
Proxy type: SOCKS5