sudipto@magnus.acs.ohio-state.edu (Sudipto Mukherjee) (06/01/91)
This has to do with HPUX 7.03 versus HPUX 7.05. We had a package called MOLDFLOW running on 7.03. We updated to 7.05 and moldflow would not run and kept giving a XOpenDisplay error. We changed the DISPLAY variable to unix:0:0 from Balki:0:0 where Balki is the name of our system. The package runs fine now. Now the question. Why do I have to set DISPLAY in this version of HPUX when I did no such thing to get it running on 7.03 ?? Right now, we set DISPLAY before we run the package. Is there any forseeable problem in setting this in our login scripts ?? I ought to mention that Moldflow runs on top of X. I would appreciate if anybody has clues to this, specially about why this change after the update. Thanks Sudipto sudipto+@osu.edu
ianhogg@cs.umn.edu (Ian J. Hogg) (06/01/91)
In article <1991May31.170820.14565@magnus.acs.ohio-state.edu>, sudipto@magnus.acs.ohio-state.edu (Sudipto Mukherjee) writes: > This has to do with HPUX 7.03 versus HPUX 7.05. We had a package > called MOLDFLOW running on 7.03. We updated to 7.05 and moldflow > would not run and kept giving a XOpenDisplay error. We changed the > DISPLAY variable to unix:0:0 from Balki:0:0 where Balki is the > name of our system. The package runs fine now. > > Now the question. Why do I have to set DISPLAY in this version of > HPUX when I did no such thing to get it running on 7.03 ?? > unix:0.0 used to be acceptable for a specification of the local display. The official value for a local connection is ":0.0". You don't mention it but I assume that have a different version of MOLDFLOW. If you are running the same binary on both versions I don't know why the "problem" came up. The only place that the DISPLAY variable is used is in Xlib. > Right now, we set DISPLAY before we run the package. Is there any > forseeable problem in setting this in our login scripts ?? > Possible performance degradation. When you create a connection to hostname:0.0, an AF_INET socket is created which causes all communication to go through the networking software (and possibly the hardware). When a connection to :0.0 is made an AF_UNIX socket is created which is just an named pipe. Also, I believe the :0.0 tells Xlib that it can communicate with the server via the most efficient method available. A vendor could support transport mechanism based on shared memory if they wanted. I think there is an extension that allows XImage data to be passed via shared memory. It is possible that Xlib is smart enough to detect if hostname is local and then create an AF_UNIX socket. I doubt it though. > I ought to mention that Moldflow runs on top of X. > > I would appreciate if anybody has clues to this, specially about > why this change after the update. > > Thanks > Sudipto > sudipto+@osu.edu
bcripe@hpcvlx.cv.hp.com (Brian E. Cripe) (06/04/91)
> This has to do with HPUX 7.03 versus HPUX 7.05. We had a package > called MOLDFLOW running on 7.03. We updated to 7.05 and moldflow > would not run and kept giving a XOpenDisplay error. We changed the > DISPLAY variable to unix:0:0 from Balki:0:0 where Balki is the > name of our system. The package runs fine now. Here is a guess. Some versions of Xlib will look at a display name such as "Balki:0.0" and use Unix-domain sockets if it specifies the local host -- other versions of Xlib will always use Internet sockets with this type of name. So ... If MOLDFLOW is built on the latter type of Xlib (always use Internet sockets if a hostname is specified), then when you changed DISPLAY from Balki:0.0 to unix:0.0 you switched from Internet sockets to Unix-domain sockets. Perhaps your update to 7.05 broke Internet sockets somehow? Brian Cripe