Re: local vs global object IDs

[Espen Skoglund]

[Ludovic Courtès]
> Hi,
> On Wed, Jun 04, 2003 at 12:38:45PM +0200, Espen Skoglund wrote:
>> If your model is that only a few tasks are persistent, then you
>> have a whole lot of other problems to deal with.  I.e., you
>> probably have many more open connections/dependencies between
>> transient and persistent tasks.  Making orthogonal persistence
>> completely transparent in such a system is difficult.

> I think it's probably hard, if not impossible, to have a persistent
> task not depend on a transient task.  Consider an X application: the
> application itself could be made persistent but it depends on the X
> server, so the X server would have to be persistent too.  The X
> server depends on the graphical hardware device drivers which
> *cannot* be made persistent.

> So you have to find another way to handle such cases.  For instance,
> the X server could be transient, and the system would log (in a
> "clever" way: it may not be necessary to keep everything in the log)
> communication between the persistent client and the X server in
> order to replay it upon restart.

My angle: keep the X-server persistent (including the framebuffer).
Some persistent tasks (e.g., the X-server) must be able to handle
communication with transient tasks (e.g., the graphics driver).  When
it detects that the driver changes it does some sort of
re-initialization (in our case, telling the graphics driver to reset
the framebuffer device to some specific state; resolution, depth,
etc.)

        eSk