Programming

You're not trying to measure how long a proces has been inactive, you're trying to measure how long a user has been inactive. The process could well still be occasionally busy with other things even when not processing user input.

Configure your window environments to logout after x minutes of inactivity, instead of trying to measure that from the outside in. They actually deal with user input and know when it's happening -- the WM will know that not just keystrokes but mouse movements mean they're not idle, for example.

This will mean that different login methods will need to be controlled in different ways if they can be controlled at all, but has the advantage that it might actually be possible.

Thank you. That makes sense now. So my flow will be something like
1) Have window sessions configured - With this idle sessions will be terminated anyway
2) Find other mechanisms to handle other applications - I still need to figure this out! It's very likely that I need to fiddle with the /proc system as I cannot modify the source code of the applications.

I repeat: /proc statistics can't tell you how long a user has been idle. You can't count on user inactivity being correlated with process inactivity, particularly for complicated GUI things with threading and event loops.

You should watch I/O, not processes. The GUI watches how long a session is idle by tracking keyboard and mouse input(or the lack of it!). The equivalent for a remote user would be watching when a terminal is idle.

This, again, is best done from inside, hence the existence of variables like TMOUT. Maybe you can find a way to snoop on the terminals, though I wouldn't want to imagine the security holes that could open.

Oh I didn't think of it that way before! Thanks for reminding Then I need to find out a way from the application side itself! Let's see how it goes.