Re: prototype and consequent feedback on XDI primitives

[Giovanni Bartolomeo <giovanni.bartolomeo@uniroma2.it>]

Thanks Andy for your answer! I add some clarification below...

At 16.24 11/04/2007, andy.dale@ootao.com wrote:

> Hi Giovanni,
>
> Good to hear from you. 
>
> We have a newer version of the XDI server that we are
> planning on releasing as a 1.0 version in mid June.  This version is
> a C++ implementation. 
>
> I have cc'd Steve on this email so that if you want a
> version of the current deployment he can send it to you. It is available
> under free open source license. 

Yes please Steve, it would help me!


> I'm not sure what
> your NOTIFICATION primitive is for, could you explain the use case (maybe
> give me a sequence diagram?). 

For notification I refer to WSDL:notification, which is an asynchronous
message sent by the server to the client based on some subscription. A
silly example, your client might subscribed to =giovanni/+phone/work and
get a notification when this information changes; This is maybe of poor
use in traditional web-based application, but I've experimented that it
is highly required in highly dynamic environment (e.g. telecommunication
applications) and XDM for example does it transporting notification
messages using SIP. 
I think that once we'll have such a capability implemented and tested,
and if we find it useful, we could go back to the specification and add
some formal primitive to reflect this new feature, provided we've gained
enough experience to do that :-) 


> The new server has
> a robust plugin framework that lets you trigger custom code at any part
> of the get or set operation on any node in the graph. Ultimately we may
> convert the plugin API to Higgins IdAS but not in the short term. Plugins
> are defined and triggered much the same way as servlets are in tomcat;
> There is an xml server config file that has a plugins section, for each
> plugin you specify the location of the executable code, what point in the
> traversal do you want to trigger the plugin (before_get, after_set,
> before_set, etc...) and the xri pattern(s) that should trigger the
> plugin. For example: 
>
> $ALL/+email/home - would trigger the plugin for all home
> email address instance nodes 
>
> $ALL/+email/$children - would trigger the plugin for all
> email address instance nodes 
>
> $ALL/+email - would trigger on the email type node
>
> Steve can send you full documentation of the plugin
> triggers. It gets a little complicated when defining triggers that should
> (or should not) dereference links and refs.  We have found that this
> approach gives a lot of flexibility and power in setting up custom code
> to run whenever we need it to. 

Ok, I see here something which sounds familiar to me, e.g. concepts taken
from aspect oriented programming (after_, before_, around_, etc..). It
seems a really good approach with a great degree of freedom. However,
what we really need for the moment is a very simple implementation able
to execute basic actions (get, set, etc.) on legacy data repository, like
a relational database or xml files. Did you already implemented something
similar?
For what concerns the platform, in our demo we'll have
"laptop-servers (1.5GHz+, 1GB RAM) running some version of Linux
(probably Ubuntu 6.10) with MySQL (version 5.0.*) and Java (J2SE
5)". Is the c++ code tested on linux environment and do you think it
will be possible to write some JNI to offer a Java interface? 

Thanks to you and Steve for your great help!
Giovanni