jwise@unicast.com said:
> Yes, RQ_FILE is necessary for more than I thought it was

sock_host() complained about this - yes. But it appears to me that
the file descriptor isn't really needed anymore at this point -
it is used by sock_host() to get the peer address, but this was
already delivered by the RPC code, so there shouldn't be much
need to grovel in the RPC internals for the fd.

There is a strange assymmetry (or a bug?) within the
wrap library that sock_methods() is only applied within
sock_host() - this would be better done from request_init()
or request_set() imho to eliminate the need to call an
undocumented interface. (Otoh, sock_host() is not mentioned
in the manpace either...)

The current fix was reported to work by Andreas Wrede, so
I'd leave it as is as long no case shows up where it
fails.
I believe to remember that libwrap is going to be cleaned
up anyway - let's hope it gets a sane interface now...

best regards
Matthias