xml-dev - Re: updategrams

[james anderson <james.anderson@mecomnet.de>]

I would have thought that one can do it by treating modifiables as
generational: one (transient) property is dynamic and changed upon
modification. The modifiables must be reified: each becomes an
identified object if it is not so already. The granularity is limited by
the amount of space which you are willing to devote.

We ended up using a generation id in addition to any object identity
which may have already been present and preventing mods if the
identified entity had changed. Web-DAV-like. Another option would be to
use the generation as part of the path, in which case an update failure
would be 'cause the generation was no longer "there" to modify.
It's sort of like multi-processed garbage collection with a fault on the
presence of a forwarding pointer.

? How about the other issue: do updategrams scale?
What do they propose for moby instances - those which go beyond single
relations, beyond tractable joins? I'm not worried about adding an entry
to an order invoice but rather something more like modifying a patient
medical history.

Roger Wolter wrote:
> 
> The way I understand optimistic concurrency control, I have to know what
> data values a record contained at the time my program read it so the
> concurrency control mechanism can compare these values to what's in the
> document to determine if it has been altered in the time since my program
> read the data.  Unless you are going to lock the document while you are
> modifying the XML (which could conceivably be a significant portion of your
> database) I don't see how you  have any choice in a loosely coupled
> environment.  Before/After images do both location and currency control.
> Yes, you can use XPath for location but I'm not sure how you would use it
> for concurrency control.
> 
> -----Original Message-----
> From: james anderson [mailto:james.anderson@mecomnet.de]
> ...
> 
> i read this in the short descriptions which i found online, but i'm
> sceptical about it. other tagging methods whould have been sufficient to
> achieve the same ends with pointers.
> has there been any experience with the matching costs against data
> objects more complex than a few items out of a single relational dataset?
> 
> i would wonder whether one has implemented xml pointers for this with
> yet another encoding...