Stable release with performance improvements. - Reduce number of dynamic memory allocations and improve cache locality. - Reduce stack consumption and increase buffer sizes.

Stable release with performance, portability, and correctness improvements. BSD support isn't tested, but should work. - Use lockless list for gc thread list. - Use accept4() on BSDs that support it. - Ensure that basic POSIX accept() path works. - Handle uncommon Linux-specific accept() errors. - Correct a fd leak on OOM error path. - Run GC more often to reduce peak memory consumption. - Keep a cache of unused thread structs to reduce malloc()/free() calls.

Stable release with some incremental improvements. - Written in plain C; no dependency on C++. - Binary size and memory footprint is smaller with same or better performance. - Proxy connections to localhost are disabled by default. - Logging can be enabled or disabled at runtime via command switches. - Logging is improved to record disconnections and total bytes sent/received.