WireGuard connection fails on Python 3.14: run_in_executor socket checks always return False in tcpcheck.py

[decorahensisire]

Summary

WireGuard connections fail on Python 3.14 with Error: Connection failed. Try connecting to a different server or check your network settings. The root cause is that tcpcheck.is_any_port_reachable() always returns False when called from within the CLI's asyncio event loop on Python 3.14, even though the target ports are genuinely reachable.

Environment

• OS: EndeavourOS (Arch Linux)
• python-proton-vpn-api-core: 5.2.5-1
• proton-vpn-cli: 1.0.1-1
• Python: 3.14.6
• NetworkManager: 1.58.0

Steps to Reproduce

1. Set protocol to WireGuard (~/.config/Proton/VPN/settings.json → "protocol": "wireguard")
2. Run protonvpn --verbose connect

Observed Output

CONN.CONNECT:START | Protocol: wireguard
TCP check: ip=<server_ip> ports=[443, 7770, 8443]
TCP check result: False
VPN server NOT reachable.
CONN:STATE_CHANGED | Error

This happens for every server selected. The same ports are confirmed reachable via direct socket tests (/dev/tcp, socket.connect_ex).

Root Cause

is_any_port_reachable() in tcpcheck.py wraps loop.run_in_executor() socket calls via asyncio.create_task + asyncio.as_completed. On Python 3.14, this combination silently returns False for all ports when called from within the CLI's existing event loop, even though the sockets would connect successfully if called directly.

Verified:

• is_port_reachable(ip, port) called synchronously → True
• asyncio.run(is_any_port_reachable(ip, ports)) in a fresh event loop → True
• Same call from within the CLI's event loop during connect → False

There is also a pre-existing logic bug in the original as_completed loop: it returns on the very first completed task regardless of its result, meaning a single False (port refused quickly) would short-circuit before other ports succeed.

Workaround

Setting server_reachable = True unconditionally in networkmanager.py allows WireGuard to connect successfully, confirming the tunnel itself is fully functional.

Suggested Fix

Replace the create_task + as_completed pattern with asyncio.gather:

results = await asyncio.gather(
    *[_is_port_reachable(port) for port in ports],
    return_exceptions=True
)
return any(r is True for r in results)

Note: this also fixes the logic bug. Further investigation into why run_in_executor socket results are wrong inside the CLI event loop on Python 3.14 may be warranted.