Close device socket in dispatcherClean to unblock EventReader thread

[alicespetma-stack]

Summary

When DispatcherWaitEventComplete times out (e.g. after the timeout fixes in PR #104), DispatcherClean is called to clean up the dispatcher state. However, the EventReader thread remains stuck in recv() because the TCP socket is never closed during cleanup. This leaks a thread and socket for each failed connection attempt, preventing clean retry.

Problem

The cleanup flow today:

1. DispatcherWaitEventComplete times out, returns error
2. Caller invokes DispatcherClean
3. dispatcherClean sets resetXLink = 1 and destroys semaphores
4. EventReader thread checks resetXLink at the top of its loop...
5. ...but never reaches the check because recv() is blocked forever

The EventReader's recv() call (in tcpipPlatformRead) has no SO_RCVTIMEO, so it blocks until data arrives or the connection breaks. If the remote device is unresponsive, neither happens.

Fix

Close the device socket at the start of dispatcherClean, before cleaning up events and semaphores. The shutdown(SHUT_RDWR) inside closeDeviceFd causes recv() to return an error immediately, allowing the EventReader thread to exit its loop.

The dispatcherDeviceFdDown flag prevents double-close when dispatcherReset (which also closes the FD) calls dispatcherClean. The flag check is inlined rather than calling dispatcherDeviceFdDown() to avoid deadlock on the non-recursive reset_mutex, which dispatcherReset already holds when it calls us.

Why not SO_RCVTIMEO?

SO_RCVTIMEO was considered but would affect all recv() calls on the socket, including healthy connections where large data transfers may legitimately take a long time. Closing the socket on cleanup is more surgical — it only interrupts recv() when we already know the connection is dead (timeout has occurred).

Files Modified

File	Change
src/shared/XLinkDispatcher.c	Close device socket early in dispatcherClean

Companion PR

This works together with PR #104 (configurable timeouts for XLinkConnect and XLinkOpenStream). PR #104 enables timeout-based error return from the main thread; this PR ensures the EventReader thread also cleans up.

Test Plan

• Build on Linux and Windows
• After XLinkConnect timeout, verify EventReader thread exits (no leaked threads)
• After timeout + cleanup, verify subsequent XLinkConnect retry succeeds on a healthy device
• Normal device operation unaffected (socket only closed during cleanup, not during active use)
• dispatcherReset path still works correctly (no double-close)