Just so you know this didn't fall off my radar. I'm taking the time to solve a wrinkle in the win32 tests within the deferred connect failure on win32. The sockaddr still points at the original 0.0.0.0 generic address, meaning that apr_socket_connect fails. Investigating, and also verifing the need to add either Mladen's pre-unset the error code patch around WaitForMultipleObjects > 64 objects, and the need for any additional traps for either of the WaitFor when the holder of the event exits (the primary cases of this were patched quite a while ago.)
