retry_connect weirdness

[Helmut Eller <eller.helmut@gmail.com>]

I have some questions regarding retry_connect in Fmake_network_process.

1. if connect() is interrupted (EINTR) the socket is closed via
   emacs_close() and a new socket is created on the next iteration.
   Wouldn't it be better to retry with the same socket?  

   I have a server that accepts exactly one connection, i.e. the server
   closes the listening socket after accept(). I have the suspicion that
   retrying with a fresh socket leads to problems, unfortunately the
   scenario is hard to demonstrate due to timing and interrupts etc.

2. what's the matter with those turn_on_atimers/turn_on_atimers
   around connect()?  Why are timer interrupts allowed
   in other places but not during connect?  It's especially odd
   since connect() can take so long.

3. why is EISCONN ignored?

4. am I the only who's wondering how anything can work at all with this
   #ifdef/#ifndef jungle?

Helmut