mdr at scn.org (Mike Rutenberg) wrote:
> This is part of a larger issue which is to be clear that the definitive
> state for Celeste is stored on the disk file, and what is in the image
> is a (very helpful) performance enhancing hint.  The current version of
> Celeste does little to detect inconsistancies between the two, or handle
> them gracefully.  Usually it works fine the way it is, but it can cause
> problems.
> 

Great idea, though let's weaken it to say categorization of messages
doesn't have to be current on disk.  Otherwise, we'd have to sync to
disk after every move or delete operation.

Overall, it will probably take at least a file locking primitive to get
really good guarantees, and I wonder if file locking is opening a can of
worms....


-Lex