On Wed, 2003-07-23 at 19:19, Lex Spoon wrote:
> It would be perfectly recoverable, if you make swiki pages get into
> changesets and into the changes file.
> 
After some talk on #squeak about the direction I want to move in (full
capabilities for security, for starters), I'm inclined to put storage
for Gardner's Wiki pages back into the image, iff I can find some decent
way to keep a transaction log (snapshot the image daily, and cycle it
weekly, that'd probably let me sleep well enough ;-)). There are two
issues to solve: the best format to store message sends in a txn log
(maybe with some data like timestamping?) and the best place to actually
decide that a message send should land in the log. 

VW has immutability support that would come in handy here: in VW, I'd
probably attack this problem by flagging a 'persistent' object immutable
- if some message send changes the object's state, then mutate the
object by flagging it mutable, log the message send, and make the object
immutable again. In Squeak, I'll probably tackle it on the application
level for the time being.

> Incidentally, do you know what you are doing to get unopenable
> snapshots?  
I had one crashing image that was busy talking to non-existent TCP
sockets, IIRC. It might have something to do with Seaside, with all its
stack juggling, leaving stuff in an unexpected state (I might, for
example, have pressed 'Save' in the middle of a Seaside request - of
course, if this were the case, it becomes tricky to have an
auto-snapshotting server image...).

Anyway, if someone is interested in doing an autopsy, I've got an image
that starts up with a black screen and eats 100% CPU time -
http://tai42.xs4all.nl/~cg/hosed.tar.gz - with my VM (3.4-1 #1, it says;
Ian's produce).