Dear all,
I would like to propose a set of changes to the protocol trying to address the main issues that we've recently discussed here. The changes are described below and if you want to browse the corresponding schema you can find it here:
http://www.cria.org.br/~renato/tapir.xsd
The main issues are related to TapirLite (how to accomodate it without making generic clients overly complicated), and are also related to Donald's suggestion about inventory templates. All other changes proposed were a mere consequence of browsing the schema and finding parts that I thought could be improved.
Please, take a look when you have some time and let me know if you have any comments, suggestions, objections, etc. Personally I'm happy with the changes, so if you all agree, and if we don't find other issues, I'll suggest to "feature freeze" it as soon as possible so that people can (re)start implementation based on it.
The feature freeze was originally planned for yesterday (December 15th), but since the current version still has open issues and this proposal should obviously be reviewed and discussed, I'll suggest that we postpone it for at least January 5th. By the way, I'll be on vacation during the next two weeks, so only accessing e-mail sporadically.
Best Regards, -- Renato
------- TAPIR / proposed changes -------
* Change filter capabilities so that implementations either accept all filter functionality or none. Specific changes: rename element /response/capabilities/operators to "filter" and include an optional subelement called "encoding". Rename "boolean" to "booleanOperators", and make all logical and comparative operators mandatory. This way parsers can always rely on the existence of a <filter> element and then pass it to a specific handler (same technique suggested by Steve in other parts of the protocol). Such handler can just detect if there's an element "encoding" or not, which means "I understand the basic filtering mechanism" or "I don't know how to parse any filters".
TapirLite implementations will answer in the capabilities: ... <filters/> ...
Full TAPIR implementations will answer: ... <filters> <encoding> <expression> <concept/> <literal/> <parameter/> <variable/> <arithmetic/> </expression> <booleanOperators> <logical> <not/> <and/> <or/> </logical> <comparative> <equals/> <greaterThan/> <greaterThanOrEquals/> <lessThan/> <lessThanOrEquals/> <in/> <isNull/> <like/> </comparative> </booleanOperators> </encoding> </filters> ...
* Change multiCOPType (basis of the <in> operator) so that <values> can contain a sequence of "expression" elements instead of only literals. This will allow parameters with multiple values (or multiple parameters) to be used by the TCS API.
* Re-include an operation completely dedicated to the usage of queryTemples (I called it "view" but I'm happy with any other better names if people feel uncomfortable with it). This operation would only be accessible through GET whithout using XML requests. And it would not accept the "filter" and "partial" parameters. This seems to be the cleanest way to accomodate TapirLite since the _only_ change in the schema is the new optional element /response/capabilities/operations/view indicating the query templates supported. Obs: I tried other ways, but the results were not so nice and the changes were more drastic.
* Remove the possibility to use queryTemplateURI in XML search requests. If someone wants to use a queryTemplateURI in a search request, it's so much simpler to use regular GET parameters in the URL that it's looks reasonable to simplify the protocol in these situations.
* Move /response/capabilities/operations/search/queryTemplates to /response/capabilities/operations/view/queryTemplates.
* On the other hand, include a new <predefinedOutputModels> element in /response/capabilities/operations/search (issue raised in a recent discussion with Rob in this mailing list). DiGIR2 will make use of it.
* Reduce "queryTemplateGroup" to contain only <label> and <documentation>. Move the other elements to a new "searchTemplateGroup" which also references the "queryTemplateGroup". Create a new "inventoryTemplateGroup" in the same way, and make the inventory request element reference it (as it happens in the search request definition). Remove global element <queryTemplate>, including <searchTemplate> and <inventoryTemplate> to make the schema validate external template definitions. This way query templates used in view operations can point to predefined search or inventory operations.
Other changes:
* Make /response/capabilities/concepts/schema a mandatory element so that any implementation shows which concepts are present.
* Remove controlled vocabulary of element /response/metadata/relatedEntities/entity/contact/role and make it unbounded: controlled vocabularies do help with standardization, but they also reduce flexibility. In different contexts people could easily envision different uses for that attribute, so I would rather enforce vocabulary control in other levels (such as software configuration or network configuration).
* Remove "datum" attribute from geoLocationType: the corresponding element has been included so that "related entities" (part of metadata responses) could indicate their location. People can easily always put lat/long in WGS84 in this case, it's just a matter of adding a note about that in the documentation.
* Rename primaryConceptNameServer and secondaryConceptNameServer to primary and secondary, respectively (the enclosing element already provides the context). Make <primary> optional.
* Rename global element <environment> to <variable>, and move /response/capabilities/search/variables to /response/capabilities/variables since they can be used in both mappings and filters. Remove redefinition of all variables in the capabilities response by referencing the existing variable element defintion. Remove empty /response/capabilities/search/variables/environment/metadata element.
-------