gnuclient/gnuserv and local internet domain sockets

I'm trying to get multiple gnuserv processes running on the same machine at different ports so that I can selectively connect to a specific one. The high-level specific purpose is to enable me to connect to *the* XEmacs that is already running GNUS, or *the* XEmacs that already has my dissertation text loaded. I typically want to connect via a non-windowed attachment from home, ssh'd in to the machine on which the XEmacsen are running. My attempt to make this work has been to set GNU_PORT to a distinguished number before starting the XEmacs processes. Then I was hoping to be able to use, e.g., gnuclient -p 21033 and gnuclient -p 21035 to connect to two different XEmacsen with GNU_PORT set to 21033 and 21035 respectively. the connection *unless* I give the -h option to specify the hostname -- that then forces using internet sockets (and the GNU-SECURE-1 or MIT Magic Cookie authentication method). (The source code to gnuserv leads me to believe that the unix domain socket name cannot be altered.) Unfortunately, the -h option to gnuclient then requires that the -nw option (non-windowing) is *not* given. This is unfortunate, since it's far more convenient to not start a full new frame over a slow modem connection. To work around this difficulty, I've re-built my gnuclient.c to have line 464's conditional always fail to not report an error: if (suppress_windows_system && hostarg && 0 /*GJB:: added && 0 */) { fprintf (stderr, "%s: Remote editing is available only on X\n", progname); exit (1); } I don't doubt that there is some problem with -nw editing w/o X11, but here the fact that the hostarg is given really shouldn't imply that remote editing is requested-- perhaps the host IPs should be compared, instead. I.e., I'm using the -h argument strictly to permit the ability to use internet sockets instead of a unix domain socket. Another, more-direct, possibility to fix this difficulty is to simply permit naming the unix-domain socket to use. I'm not sure of the security implication (if any) of this suggestion, but it seems likely that mangling the UID + some user-specified number would be a nice way to provide this capability in a fairly clean way. I'd love to see a suitable change in the next versions of gnuclient/gnuserv to better support multiple servers. Thanks, Greg J. Badros gjb(a)cs.washington.edu Seattle, WA USA http://www.cs.washington.edu/homes/gjb