En réponse à Isaac Yonemoto <ityonemo@xxxxxxxxxxxx>: > > > It is different having an option to do that and making > > > that automatic. What Helmar sugested was that all apps > > > the user had opened when the computer shutdown would be > > > reopened when it restarted. > > I'm sorry, we seemed to have deviated from what I was suggesting: > > I suggested that apps get "dehydrated" at the kernel-land level (and > not the userland level). This means that there is no message-passing, > or > app-server involved (in fact, app_server, itself, would be dehydrated > in > the process). This already exist, it's what laptops do when you stop them. But it needs ACPI support in kernel *and* drivers. > > As to malicious apps: > > 1. They would have to be running already. > 2. BeOS is excellent at killing unresponsive apps already. > 3. There's no "startup" message/app loading. It gets immediately put > back to where it was. As far as the app knows, it just suffered a > time > lapse. This might interfere with media apps that use strict time > codes, > so perhaps a "you've just entered a timewarp" message needs to be > passed > systemwide. > > As to the idea of rebootless kernel upgrades: > > The entire state of the kernel need not be saved, in fact only certain > critical information (you can't bootstrap from nothing, anyway, so > there > must at least be some rudimentary kernel functions active, like FS, > for > example). Whatever future versions of the kernel naturally need to be > able to understand whatever stored data there is, but there's nothing > stopping the data format from being augmented or changed a little. > As I said not even GNU Hurd does it, so I bet it's pretty hard (nothing is impossible, but...) to implement. François.