We combine LR(k)-parsing technology, data-flow analysis, and abstract interpretation to analyze, in advance of execution, the documents generated dynamically by a program. Based on the document language's context-free reference grammar and the program's control structure, the analysis predicts how the documents will be generated and parses the predicted documents. Our strategy remembers context-free structure by computing abstract LR-parse stacks. The technique is implemented in Objective Caml and has statically validated a suite of PHP programs that dynamically generate HTML documents.

Joint work with Kyung-Goo Doh and Hyunha Kim, Hanyang University, Ansan, South Korea.