Caching for XML Generators: An Application of Two-Level Techniques

Caching of documents is an important concern on the Web. It is a major win in all situations where bandwidth is limited. Unfortunately, the increasing spread of dynamically generated documents seriously hampers traditional caching techniques in browsers and on proxy servers. WASH/CGI is a Haskell-based domain specific language for creating interactive Web applications. The Web pages generated by a WASH/CGI application are highly dynamic and cannot be cached with traditional means. We show how to implement the dynamic caching scheme of the BigWig language [2] in the context of WASH/CGI. The main issue in BigWig’s caching scheme is the distinction between fixed parts (that should be cached) and variable parts (that need not be cached) of a document. Since BigWig is a standalone domain-specific language, its compiler can perform the distinction as part of its static analysis. Hence, the challenge in our implementation is to obtain the same information without involving the compiler. To this end, we extend WASH/CGI’s document language by mode annotations and define the translation of the resulting annotated document language into JavaScript. To alleviate the awkwardness of programming directly in annotated language, we have defined a surface syntax in the style of HSP (Haskell Server Pages) [11].