View De nition and DTD Inference for XML
View De nition and DTD Inference for XML
复制标题
XML 的视图定义和 DTD 推断
DOI:
--
复制
发表时间:
1999
期刊:
影响因子:
--
通讯作者:
V. Vianu
中科院分区:
文献类型:
--
作者:
Bertram Ludäscher;Y. Papakonstantinou;P. Velikhov;V. Vianu
We present a query language and mediator architecture for XML data [XML98b]. The query language, calledXmas (XMLMatching And Structuring Language) uses vertical and horizontal navigation. The horizontal navigation provides a powerful novel mechanism for using the order in the semistructured model underlying XML. This couples nicely with the well-known regular path expressions for vertical navigation [Suc, Abi97, Bun97, AQM97, FFLS98, AM98, CDSS98, XML98a, KS95]. Xmas also provides versatile mechanisms for constructing XML documents as answers to queries, including powerful group-by and order-by constructs. We brie y describe a prototype for an XML-based information mediation system, built around Xmas. DTDs can have multiple uses in creating integrated views and querying XML data. Our prototype uses a QBE-style query interfaces driven by the view DTD. The mediator will use DTDs to optimize the queries it sends to the sources. Finally DTDs may guide the production of style sheets, such as XSL scripts [XSL98], that display XML documents as browser-friendly HTML documents. DTDs may also help in the design of the storage structures. Thus, it is clear that DTDs of integrated views will be particularly useful. However, creating a view DTD \manually" by delving into the details of the source DTDs is error-prone and may become the bottleneck of the integration software development, compromising much of the advantage of semi-automatic data integration. A central component of our mediator is a DTD inference module. After outlining the architecture of the mediator and illustrating Xmas, we present work in progress and results (submitted for publication [PV98]) on DTD inference in views de ned by Xmas queries. We point out several strong limitations of DTDs and the need for extending them with (i) a subtyping mechanism and (ii) a more powerful speci cation mechanism than regular languages, such as contextfree languages. With these extensions, we show that one can always infer tight DTDs for views de ned by selection queries, which extract a list of elements from the input. The tight DTD precisely characterizes the documents in the view on sources satisfying given DTDs. We also show interesting special cases where we can derive a tight DTD without requiring the extension with subtyping. Finally we consider related problems such as checking conformance of a view de nition to a prede ned DTD.