of always returning ISC_R_UNEXPECTED

                        by logging code. [RT #1293]

 814.   [bug]           Socket objects left over from accept() failures
                        were incorrectly destroyed, causing corruption
                        of socket manager data structures.

 813.   [bug]           File descriptors exceeding FD_SETSIZE were handled
                        badly. [RT #1192]

into hard errors - they should be soft errors

got wedged in a state where accept() always returns EBADF, the server would loop calling
internal_accept() repeatedly.  Implemented an alternative fix where a hard
error in accept() causes an event with a nonzero result code to be sent, and
the new socket object to be freed

a soft error as a hard error on send (that is, not retry it).  It also
modifies the code calling sendmsg() to retry on EINTR.

                        cause the server to retry the transmission
                        indefinitely. [RT #902]

a partially constructed event and flags.  The flags can include
ISC_SOCKFLAG_IMMEDIATE, which means that if the operation completes, the
event is filled in and not sent.

twice, and the socket object for the new socket was never destroyed,
causing the server to hang on exit (if multithreaded) or dump core on exit
(if singlethreaded).  Now the only difference between "hard" and "soft" errors
is that the latter are not logged.

accept() returns a valid peer address; instead just log an ugly error message
and continue if the peer address is left unchanged or insane by accept()
[RT #809]

the socket lock.  (Assuming the operation succeeds immediately, and doesn't
need to be queued.)

- Communications with the watcher thread tweaked to reduce the need for
locking.

assign the "pf" field in the new socket.

used the wrong errno value [RT #640]

I would have seen *before* I commited anything.  So, include isc/thread.h,
isc/mutex.h and isc/condition.h even if ISC_PLATFORM_USETHREADS is not defined.

(What caused me to bother with this at all was a problem that I resolved
a few days ago by fixing configure ... though now that I think about it,
that probably means there is some other latent problem with inconsistent
definitions that could maybe be handled better.  I'll look into it more
later, AFTER vacation.)

ISC_PLATFORM_USETHREADS is defined.

isc_msgcat_get() and speed things up.

least be duplicated consistently

                        should return EAFNOSUPPORT.  Work around this.

unexpected error [RT #215]