Persistence & DTSTTCPW: ZODB clone?

goran.hultgren at bluefish.se goran.hultgren at bluefish.se
Fri Feb 1 12:56:35 UTC 2002


cg at home.cdegroot.com (Cees de Groot) wrote:
>  <goran.hultgren at bluefish.se> said:
[SNIP]
> Note that CATS wants you to subclass transactional objects (IIRC); that's

Are you sure? No, I think you needed to "enable" your classes using a
special compiler.
But I may be remembering Stephen's demo wrong.

Persistence should of course (most seem to agree on that nowadays) be
orthogonal to type/class.

> maybe something you don't want. However, I don't know another way of doing
> what CATS does with routing instance variable accesses to the transaction
> (except for some transparent proxy magic, which would work but maybe doing
> this through DNU is bound to be slow)
> 
> >I know he has been thinking about all the other parts too and a small
> >bird has whispered in my ear that there might be stuff coming...
> >
> Could you wisper a bit louder, bird? I'd hate to do double work ;-)

Stephen will have to whisper himself I think. ;-) Drop him an email.
 
> >As another sidenote, I have been playing with ImageSegments a bit and
> ><plug> if you sign up for an account at SqueakDot </plug> you can
> >actually see it in action right there - you can
> >checkpoint/revert/analyze your account using ImageSegments and you will
> >see some performance numbers right away.
> >
> The only thing I keep asking myself is whether ImageSegments are good for
> fine-grained storage (say a persistent objects and the reference graph to
> its non-persistent contained objects, as opposed to a big project). They seem
> to be a bit heavy-weighted, don't they?

Yep, they are a bit on the heavy side. But they may be useful
nevertheless! :-)
They are darn fast coming in, a bit slower going out of course given how
it's done,
you pay a price for the overhead of a markphase of course. But it's
neat.

Note: I use export segments at SqueakDot and that costs a bit more of
course. It's because
I want to be able to start fresh from an empty image.

> (OBTW: David Gorisek is contemplating an OmniBase port to Squeak; however,
> what's keeping him until now is that he cannot seem to do file locking from
> Squeak - anything coming there? Is there a lowest common denominator between
> the Squeak platforms for file locking?)

All the file stuff needs a big overhaul I guess. There are people
pondering that.

regards, Göran



More information about the Squeak-dev mailing list