Matching Automata for Regular Patterns

Pattern mat hing me hanisms based on regular expressions are featured in a number of re ent languages for pro essing tree-stru tured data su h as XML. A ompiler for su h a language must address all the familiar problems of pattern optimization in fun tional languages with ML-style algebrai datatypes and pattern mat hing, plus some new ones arising prin ipally from the use of re ursion in patterns. We propose mat hing automata as a uniform framework for designing regular pattern ompilers and omparing different ompilation strategies. Using this framework, we de ne two pattern ompilers|one generating ba ktra king ode and the other a non-ba ktra king variant|by spe ifying two sets of stru tural restri tions on mat hing automata and giving algorithms (with sket hes of orre tness proofs) that onstru t automata of these forms from sour elanguage patterns.