Windows NT Porting

• Subject:
  This documentation deals with the porting of the VRMidtown client and viewer from SGI to Windows NT, done during Fall Quarter 1998.

• Known Bugs:
  • Problem when invalid server is entered
    The client/viewer will apparently keep trying to connect to a server that is not responding, which begins to eat up CPU time on the NT machine.

  • Incorrect exiting
    The client/viewer will exit immediately, without sending a disconnect message to the server. The abandoned client will hang around until a certain number of packets have been ignored by the (nonexistant) viewer before killing itself.

  • Lack of decent commenting
    A good amount of the commenting currently in the NT source is left over from the SGI code, much of it out of place in the files. What Windows-specific commentary that is in there is rather cryptic. This should be changed.

• Incompatibility Issues Encountered:
  • Threads
    Microsoft has no notion of what IRIX pthreads are. Much the same way, Microsoft threads, though well-done, have no similar implementation in IRIX. The changes I made here were done in two sections: the mutex.h and mutex.c needed to be re-written to take advantage of MS's built-in handles for synchronization, and the spot in network.c where a new thread was created.

  • Networking
    Not terribly different, the few changes made here were all in network.c and network.h -- fctrl does not exist in Windows, and ioctlsocket had to be used to set the socket as non-blocking.

  • Calling procedure
    This surprised me. Microsoft doesn't use the standard convention of argc and argv for windowed programs, instead any arguments are passed in in a single long string. Rather than worrying about parsing said string, I made use of the built-in file functionality inside the WinMain function (MS's version of main()) to make everything that had been command- line on the SGI into interactive on the NT machines. This was done as a wrapper around main(), so as not to vary too much from the SGI code.

• Future Plans/Suggestions:
  • More abstraction
    Make the main (changing) part of the code more-or-less platform independent, and have the platform specific stuff be called in similar functions.

  • Combination of source
    Sometime early next quarter, I would like to have the NT source and the SGI source together in a single source tree, the version being determined at compile-time by a #define, or a -D in the Makefile. I believe this is possible at this point, with some effort.

  • Easy installation method
    When this becomes the client/viewer of choice for 1502 students, it needs to be easily portable. I have begun researching InstallShield, and the possibility of using an evaluation version to create a simple, point-type-and-click install method.

  • Better logon screen
    Rather than re-typing in the same server time and again, a logon screen that would save the server and port last connected to, and put that up as a default would ease use.

  • Saved user preferences
    Doable through the registry, however, I have researched this very little.

  • Added functionality in 'meta' files
    With the advent of user interaction as the client/viewer opens, rather than reading completely from files, the possibility of creating what is now 'meta' files from scratch on the fly comes into play.

<< Back to Course View: Fall 1998