I am currently developing a multi-threaded C++ apllication on a HP workstation (HP-UX 10.20). The application contains a series of X windows (X11R5) that have been developed using UIM/X. The application receives input data from a variety of external sources. The input is received on a combination of CORBA interfaces, TCP/IP sockets and one serial port. I have set up threads to process the input data, but that data needs to be displayed in the applications windows. Since X11R5 is not thread-safe (and the possibility of being able to upgrade to X11R6 is very unlikely), I need to find a reliable, safe way of getting the application's input data from the various threads that process it into the main "default" thread that contains the X event main loop.
It has been suggested to me that I read the input data into a buffer and then use XSendEvent to generate a X event to pass a pointer to that buffer to a event handler routine for display processing. However, XSendEvent requires a display pointer and window ID, and some of my input will be requests to create a window that does not yet exist. Therefore, I may not always have a window ID to target the X event to.
Another suggestion has been to use UNIX pipes or TCP/IP sockets to pass the pointers to a callback routine to do the display processing. However, I am very unfamiliar with both pipes and sockets, and I am very concerned about error handling and error recovery (i.e. how do I receover if a pipe is broken?).
Therefore, what I am looking for is some information about the benefits and pitfalls of each of these approaches, as well as any suggestions and/or examples of other approaches that I may have not considered.
Another consideration you might look into is running the X application as a separate process and let the globals reside in shared memory (shmget, etc.) so you don't have to worry about thread safety on the X side.