(An) efficient computation of right context for LR based error repair

Abstract

The left context in LR-based parsing can be defined as the sequence of states in the parsing stack. The right contexts are the expected vocabulary strings that will appear for a given left context. One of the important applications of right contexts is syntax error repair. If a syntax error occurs during a parsing, right contexts may be used to get an insertion string that corrects the syntax error. An LR-based parser, however, requires additional space and time in computing right contexts since an LR parsing stack contains the vocabulary strings that have been seen until now, whereas an LL parsing stack contains the expected vocabulary strings from this point. For the practical programming languages, however, the LR parser is required for deterministic parsing. Hence the methods of computing right contexts for an LR-based parser have been studied by many researchers. One of the principle sources of the inefficiencies of the previous methods, in our view, comes from the fact that there are multiple kernel items in an LR(0) state. In general, there are many LR(0) states that contain multiple kernel items. In previous works, the right contexts for an LR(0) state are considered as the union of the right contexts for each kernel item in the state. Our investigation has been devoted to the relationships between the kernel items in the same state. We have found that every kernel item in the same state has the common suffixes in computing right contexts in most cases. By means of factoring out the common suffixes at parser generation time, it is possible to avoid some redundant computations of right contexts at parsing time. Also, through defining the right contexts for LR(0) states, not for LR(0) kernel items, we develop an efficient method of computing right contexts such that our method simplifies the previous diverging algorithm into a linear algorithm.