[lhoward]

HTTPServer leaves Nagle's algorithm on for the sockets it accepts. When a small write follows one the client hasn't yet acknowledged, the kernel holds it until the ACK arrives, and most clients delay that ACK by up to 40 ms. Typical cases are the next WebSocket message on a quiet connection, or a response written in more than one piece.

We noticed this on a WebSocket relaying a stream of small binary messages. About 2% of them reached the client 40 to 42 ms late, each one sent just after the client's delayed ACK. Most HTTP servers turn Nagle off on accepted connections for this reason; nginx's tcp_nodelay is on by default, for example.

This change:

• FlyingSocks: adds a tcpNoDelay socket option (IPPROTO_TCP, TCP_NODELAY), with the two constants added to each platform's Socket extension (Darwin, Glibc, Musl, Android, WinSock2).
• HTTPServer: sets the option on every connection it accepts on an IPv4 or IPv6 address.
  • A Unix domain socket has no Nagle delay, so it's left alone.
  • If setting the option fails, the server logs a warning and serves the connection anyway.
• HTTPConnection: socket changes from private to internal, so a test can read the option back. Nothing else uses that.

Tests:

• SocketTests.socket_Sets_And_Gets_TCPNoDelayOption round-trips the option on an IPv4 stream socket.
• HTTPServerTests.server_SetsTCPNoDelay_OnAcceptedConnections checks the option on the server's accepted connection. Without the server change it fails.
• On Linux (Swift 6.4), FlyingSocks passes all 155 tests. In FlyingFox, the only failures out of 314 tests are HTTPDecoderTests' standardizedPath and naughtyPath_IsParsed, which fail the same way on main.
• I haven't built it on Windows, Musl or Android. On Windows the constants follow the existing IPPROTO_IPV6.rawValue pattern.

If you'd rather make it a Configuration flag than unconditional, I'm happy to change it.

🤖 Generated with Claude Code

https://claude.ai/code/session_01CEgKfBwi5LpKqwfbctCEdH

[codecov]

Codecov Report

❌ Patch coverage is 92.30769% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 93.75%. Comparing base (e845dc0) to head (01d567c).

Files with missing lines	Patch %	Lines
FlyingFox/Sources/HTTPServer.swift	90.90%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #255      +/-   ##
==========================================
- Coverage   93.76%   93.75%   -0.01%     
==========================================
  Files          72       72              
  Lines        3847     3859      +12     
==========================================
+ Hits         3607     3618      +11     
- Misses        240      241       +1     

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

[swhitty]

TIL! thank you for this I will merge it and then just extract the switch into its on func on main 🙏