>A current theory (that not all "facts" support) is that the application may be leaving something in memory after a normal shutdown--something that persists across RDC sessions and applies to all users. I didn't think that was possible: remote sessions are supposed to be isolated from each other. I would expect all handles and connections to be terminated and memory released as the remote session ends--even if it terminated abnormally--which it didn't (at least there were no errors upon log out.) The client is convinced that the problem started with a specific version --the one where the connection was changed from string to handle--but maybe that has nothing to do with it.

In Terminal Services (or Citrix) there's somewhere deep a setting on the server which, pardon the pun, serves to ignore short glitches in the connection. So if the user disconnects before closing all the apps, they will stay open for a while - the length of which is this setting. I think I saw once, while I was administering one Citrix machine, but the default was just fine, so I didn't experiment with it and don't know how long is it by default. My guess is 5 to 15 minutes, judging by how long an absence does it take to have to restart my TotalCommander after I logged out. But I haven't really measured.

It may depend on how your app exits when the system tells it to quit. It should be equivalent to the shutdown via the closebox, but maybe it's not completely the same. And if there's a chance that it requires some user interaction when shutting down - if disconnected, there's nobody to click OK, and it may appear to have hung, while it just hangs in there. To simulate this behavior, try to shut your app from task manager and see how it behaves then.

The app may become invisible - then they really can't see it but it may be visible to task manager and any other process, if not to the windows taskbar. So... if they experience this again, just tell them to invoke the task manager and see whether the app has become an invisible zombified process. Then they should be able to close it from there (as a process, if it won't as an app).

Take the previous paragraph with a grain of salt - that behavior is something I see on my machine and a few others, but may not be a rule across the board. Also, I'm actually using TaskKiller for most of the terminations, because it kills the processes which Task Manager won't (but which die just fine when shutting down the system - TaskManager is toothless sometimes), so the behavior when shutting down an app may not be the same as the one I see.