* Tuomo Valkonen <[EMAIL PROTECTED]> [040218 21:06]: > > My thoughts on session management: > > Cheap over-engineered plastic clone of "orthogonal persistence". > > The most common reason for me to have to restart all X apps on > my home computer is the console going into an unusable state. > The OS kernel should take care of taking snapshots of the system > state at intervals and return to a recent snapshot in case of > power outage, system crash etc. It should also be possible to > do this for just a group of processes, and to mark some important > snapshots stored permanently.
> Networked programs are a problem, yes, but otherwise applications > should not need to have any kind of session management support. > After all, what at least I want is to return to a snapshot of > the system state an instant before my system was willing or > unwillingly terminated. System snapshot is not very trivial, in the time between the snapshot was taken and the crash, you might have written some new files or modified old ones. Now the snapshot needs to reverse the filesystem transactions too. Maybe some of these transaction were e-mail downloads that were deleted from the server, now if the file transaction is reversed the e-mails are gone. The only thing that really knows what is the state of the program is the program itself, it has the real context and it should have the brains to know if something changed under its feet. I myself don't bother with sessions, they don't work for my needs. I'd be happy if I could turn off the computer and return back to the same state when I powered it up again, but the same state may be different if the files on the networked filesystem were modified in the meantime. Baruch
