Streamlining Functional XML Processing∗

Since an XML document has tree structure, XML transformations are ordinarily defined as recursive functions over the tree. Their direct implementation often causes inefficient memory usage because the input XML tree needs to be completely stored in memory. In contrast, XML stream processing can minimize the memory usage and execution time since it begins to output the transformation result before reading the whole input. However, it is much harder to write the XML transformation program in stream processing style than in functional style because stream processing requires stateful programming. In this paper, we propose a method for automatic derivation of XML stream processor from XML tree transformation written in functional style. We use an extension of macro forest transducers as a model of functional XML processing. Since an XML parser is represented by (infinitary) top-down tree transducer, the automatic derivation of XML stream processor is based on the composition of the top-down tree transducer and the extension of macro forest transducers.