Socket: treat read()/recv()==0 as EOF before consulting errno

[lhoward]

A read()/recvfrom()/recvmsg() return of 0 is an orderly EOF, not an error, and does not set errno. The error-classification guard checked errno == EWOULDBLOCK before count == 0:

guard count > 0 else {
    if errno == EWOULDBLOCK { throw SocketError.blocked }
    else if errnoSignalsDisconnected() || count == 0 { throw SocketError.disconnected }
    ...
}

So an EOF read whose errno was left as EWOULDBLOCK by an earlier would-block read on the same thread is misclassified as .blocked instead of .disconnected. The reader then re-suspends waiting for more data instead of closing.

Because errno is per-thread, this only manifests under concurrency (a would-block read on one connection contaminating an EOF read on another). The consequences:

• The connection never closes — it lingers in CLOSE-WAIT (fd leak).
• epoll keeps re-reporting the readable EOF socket, so the pool spins re-arming (EPOLL_CTL_ADD/DEL) and re-reading it (read()==0), pegging CPU.

Reproducible with a handful of concurrent half-closed keep-alive connections (shutdown(SHUT_WR), held open): the server pins a core and accumulates CLOSE-WAIT sockets. With the fix, CPU stays at 0 and no sockets leak.

Fix: check count == 0 first; errno is only meaningful after a -1 return. Applied to read, recvFrom, and recvMsg.

[codecov]

Codecov Report

❌ Patch coverage is 86.66667% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 92.91%. Comparing base (0a6f327) to head (a196c2e).

Files with missing lines	Patch %	Lines
FlyingSocks/Sources/Socket.swift	86.66%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #224      +/-   ##
==========================================
- Coverage   92.92%   92.91%   -0.01%     
==========================================
  Files          71       71              
  Lines        3719     3728       +9     
==========================================
+ Hits         3456     3464       +8     
- Misses        263      264       +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

[lhoward]

FWIW this appears to be the final fix in my Claude-assisted to quest to eliminate FlyingFox-related CPU usage spikes and FD leaks on Linux.