5543 SML URI seems overconstrained

Hi Henry,
The SML WG had reached a resolution on this issue on 7/24. The resolution was to require support for XML Schema determined IDs and keep support for DTD determined IDs implementation defined. Your response on 7/26 indicates that you are not satisfied with this resolution and that you want to require support for DTD determined IDs in addition to XML Schema determined IDs.

We have debated this issue extensively inside the WG. It is quite possible that our brief response on 7/25 did not fully explain our reasoning behind the resolution. We would like to present more data on our discussion in the hope that it may persuade you to reconsider your position.

Details of the SML WG's discussion:

The WG evaluated 4 cases when discussing this issue.

1.    DTD and XML schema are both absent

2.    DTD and XML schema are both present

3.    Only XML schema is present

4.    Only DTD is present

The summary of the discussion of each of the 4 cases is presented below.

DTD and XML schema are both absent
In this case, there can be neither schema-determined nor DTD-determined ID to process therefore this case is not relevant to our discussion.

DTD and XML schema are both present
There is variability in the behavior depending upon the underlying XML data model used. For example, if DOM is used, one can get either DTD type info or Schema type info (but not both) depending upon whether DTD validation or Schema validation was performed. Further, XML schema processors operate on the infoset provided by the underlying XML processor but the XML schema spec does not require XML schema processors to make available all of the input infoset to its invoker. More specifically, XML schema processors are not required by the schema spec to expose DTD ID information to their invoker. SML validators are layered on top of XML schema processors. They do not interface directly with the XML processor used by the schema processor, therefore they do not always have access to DTD ID information. Consequently, if the SML spec requires support for DTD IDs some schema processors cannot be used by for SML validation without additional infrastructure. This can potentially limit the implementation base for SML validators.

Only XML schema is present
In this case, only schema-determined IDs can be present. The SML WG has already agreed to support them.

Only DTD is present
The SML group's charter is to standardize extensions to the XML Schema, therefore the group's focus is on supporting scenarios where XML schema combined with Schematron rules is used to validate a model. However, the group recognizes that there may be some cases where XML schema is not used at all. The group does not want to go out of the way and forbid such cases.

SML constraints can only be specified as a part of XML schema. When model has only DTDs, there cannot be SML constraints present. This is not the mainstream use-case for SML models. We expect majority of SML models to include XML schemas (with or without SML constraints) because XML schema validation is an integral part of SML model validity assessment and XML schemas are the only means of defining SML constraints. Due to implementation and schedule constraints, we would like to allow but not require DTD ID support for this case. We believe this will have minimal impact on the users of SML because most SML models will use XML Schemas for validation.


Summary:
The SML spec currently forbids the use of barenames in SML references. The WG has agreed to remove this restriction. Further, the group has agreed to require support for XML Schema determined IDs. We believe that this covers a majority of SML models. In addition, we have agreed to not forbid DTD-determined IDs. We would like to keep that support application defined for reasons described earlier.

Received on Thursday, 7 August 2008 18:45:17 UTC