LR(k) sparse-parsers and their optimisation

• Main Author: Rushby, John Martin
• Published: University of Newcastle Upon Tyne 1977
• Subjects: 410
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.471188

A method of syntactic analysis is developed which is believed to surpass all known competitors in all major respects. I The method is based upon that associated with the LR(k) grammars but is faster because it bypasses all reduction steps concerned with 'chain' productions. These are freely selected productions which are considered semantically irrelevant and whose right parts consist of just a single symbol. The parses produced by the method are 'sparse' in that they contain no references to chain productions - they are termed 'chain-free' parses. The CFLR(k) grammars are introduced as the largest class which can be -Chain-F-ree parsed from -Le-ft to Right while looking ~ symbols ahead of the current point of the parse. The properties of these grammars are examined in detail and their relationship to the conventional LR(k) grammars is explored. Techniques are presented for testing grammars for the CFLR(k) property and for constructing chain-free parsers for those grammars possessing the property. Methods are also presented for. converting ordinary LR(k) parsers into chain-free parsers. CFLR(k) parsers are more widely applicable than their LR(k) counterparts, are faster 'and provide the same excellent detection of syntactic errors. Unfortunately they also tend to be rather larger. A 'simple optimization is presented which completely'overcomes this single disadvantage without sacrificing any of the advantages of the method. These theoretical techniques are adapted to provide truly practical chain-free parsers based on the conventional SLR and,LALR parsing methods. Detailed consideration is given to use of 'default reductions' and related techniques for achd.evfng compact representations of these parsers. The resulting chain-free parsers are not only faster than their ordinary counterparts, but probably smaller too. We believe their advantages are such that they should substantially replace other parsing methods currently used in programming language compilers.