[UR-WG] Comments/Questsions on usage#draft3 schema

[Donal K. Fellows]

Matt Ford wrote:
> I have some comments/questions on your latest release of the proposed 
> 2.0 spec.
> 
> - It's much more legible now; your effort in this is much appreciated by 
> me at least! :-)

Heh! :-D

> - I like the much cleaner type derivations.
> 
> - With regard to the AtomicActivityUsageRecord type: 
>   * I don't like how we call it an AtomicActivity and then use "JobName" 
> as an element name.  The activity may have nothing to do with jobs; I 
> still feel as though we have a mix of the specific and generic in this type.

That bit was a cut-n-paste job done in a hurry. We can change it.

>   * In a similar line of thinking I wonder if elements such as "Status" 
> should not be global and instead be local to a derived type such as 
> "AtomicActivity" i.e., declared instead of referenced.  Not all Usage 
> records will necessarily have "Status" and it's meaning can in principle 
> vary wildly.

I think I could repeat myself here.

> - With regard to the CompoundUsageRecord:
>   * It's abstract as you note; so how's it made specific?  (the 
> following questions may be due to my ignorance of XSD's).  I guess I 
> could extend the schema but how then do we make it so that people don't 
> go off and arbitrarily extend it?  How do we capture this if they do?   
> For example if I naively defined a "grid-job"  - treat this name as 
> arbitrary ;-) - as the collection of AtomicActivity UsageRecords spawned 
> by my Resource Broker (never more than ten) what do I do?

The purpose of the CUR is to define how to build compounds, and not why.
The grounding of the compound is therefore a required step, and might be
done like this:

   <xsd:complexType name="AllGridJobsForAWeek">
      <xsd:annotation>
         <xsd:documentation>
            This is a compound UR type that is intended to indicate all
            jobs executed by a particular user on some grid in a
            particular week. Note that this is a higher-level constraint;
            no schema can capture the notion at a syntactic level. As
            such, it's just documented here. Not all compounds are this
            simple syntactically though.
         </xsd:documentation>
      </xsd:annotation>
      <xsd:complexContent>
         <xsd:extension base="usage:CompoundUsageRecord"/>
      </xsd:complexContent>
   </xsd:complexType>
   <xsd:element name="AllGridJobsForAWeek"
         type="tns:AllGridJobsForAWeek"
         substitutionGroup="usage:UsageRecord"/>

It should be noted that I might move the xsd:any from the CUR type into
the element; the processing of xsd:any in the presence of XML schema
extension is very strange indeed. (Every time I look at at it, I end up
going "what was the author on when they thought of that?" and that's not
a good sign.)

> Should I even attempt this: my aim is to extend the schema to allow only 
> compound records of 0 to 10 Atomic Activity records only?
[...]

Maybe. I'm not a big enough schema whizz to say (I've never tried doing
restriction of complex types). But it certainly looks close.

> Or should I be thinking of CompoundUsageRecord as plumbing for as yet 
> undefined parts of the UR-WG schema?

That's my real plan, with the definitions not necessarily being part of
the UR-WG schemas. Private extensions are acceptable.

Donal.