[news.eclipse.technology.cosmos] Re: parsing query results from multiple

[David Loewenstern <davidloe@xxxxxxxxxx>]

Thanks for the quick response, Dave.

I'm afraid to create defects or enhancements until I understand COSMOS a 
bit better.  I don't want to spam defects that have already been 
addressed or that really work as designed.  If we establish that I have 
in fact uncovered either a defect or a feature that needs to be 
addressed, I'll submit a bug report/feature request.

I thought of using a serializer aggregate but I couldn't figure out what 
it would key off of.  The problem is that the metadata you'd need to 
know which serializer to use (e.g., a schema name or an MDR identifier) 
isn't available at the point that the serializer is called.

Yes, a serializer registry could work.

The idea for the last sentence came from the javadoc for 
IRecord.setValueFromString:

	 * Set the value for this record from a string.
	 * This API is needed because the value coming in
	 * from XML will be raw text.  Implementers are
	 * responsible for converting this raw text to
	 * their format of choice.

So, if I created my own implementation of IRecord, I could have it parse 
the XML appropriately, using its own serializer instead of the one 
registered with CMDBfServicesUtil.  But how do I get 
QueryOutputStreamHandler to use my IRecord implementation?  Here is line 
108 of QueryOutputStreamHandler.java:

currentRecord = 
CommonArtifactFactory.getInstance().createRecord(currentResultEntity, null);

Why isn't it this?

currentRecord = getArtifactFactory().createRecord(currentResultEntity, 
null);

If it were, then I could create different artifact factories for 
different MDRs, although that might not be the easiest method.

Dave


David Whiteman wrote:
> Hi David,
>
> Is it possible to create a serializer aggregate that keys off something 
> in the query response, perhaps a namespace, to delgate to model-specific 
> serializers for the actual work?  That might not be an elegant solution 
> but might work for you in the near term.  We would also like to hear 
> your ideas on what sort of API you would like to see, maybe some sort of 
> serializer/datatype mapping registry?
>
> It sounds like the last sentence of your post is regarding a separate 
> problem.  When you say it ignores it for most object types, can you give 
> an example of one for which it does, or a place you think it is doing this?
>
> In general, I would encourage you to create defects and/or enhancements 
> in our Bugzilla system where appopriate, including specific details and 
> recreation scenarios, as this will make it more likely that we can get 
> the work scheduled, and offer more timely solutions.  Plus, being the 
> creator of the defect, you will get notified as the status changes.
>
> Thanks,
> David
>
> David Loewenstern wrote:
> > I apologize if this question has been answered before, but I have not 
> > seen it in the archives.
> >
> > Say I have an application that queries two _different_ MDRs with 
> > _different_ underlying object models.  The application can be 
> > configured so it knows which model goes with which MDR.   To make this 
> > concrete, say we have one MDR that handles personnel and a different 
> > MDR that handles IT assets, and we need to construct queries across 
> > both (e.g., "list all assets that are assigned to managers with pay 
> > grade above 10").
> >
> > I realize it isn't Cosmos's or CMDBf's job to tear apart the query and 
> > route the subqueries to the appropriate MDR, and I'm not worried about 
> > that.  The problem I have is specific, and has to do with handling the 
> > query responses.
> >
> > Both MDRs will send back query responses containing records.  For one, 
> > the records will be personnel, for the other, IT assets.  As far as I 
> > can tell, the CMDBf will handle both by way of the 
> > QueryOutputStreamHandler, ultimately calling 
> > CMDBfServicesUtil.getRecordSerializer().  There can be only one record 
> > serializer -- it's a class variable, so even if I have two separate 
> > instances of QueryOutputStreamHandler, they will both use the same 
> > record serializer.  But I need two serializers, one for the personnel 
> > data model, one for the IT data model.  How do I accomplish this?
> >
> > It seems that there is a problem in that QueryOutputStreamHandler 
> > accepts a user-defined ArtifactFactory but ignores it in favor of 
> > CommonArtifactFactory when creating most object types.
> >
> > David Loewenstern
> > davidloe@xxxxxxxxxx