Most of the recent parsers have the error recovery facility to give more informations to programmers. Fischer et al. proposed the locally least-cost error repair models for LL and LR parsing methods. The formal error recovery methods are independent of any programming languages or grammars, so that can be driven by the error repair tables. The locally least-cost error repair models also use the error repair table storing informations to compute right contexts. More informations in the error repair tables can lead the error repair parsers to faster repairing porcesses.
In fact, practical error repair parsers tend to backtrack or to use more than two lookahead symbols to find more suitable insertion strings. For such parsers, the time for computing insertion strings is very important. So, information to be stored in the error repair table becomes critical. In this thesis, we study the appropriate extent of informations to be stored in the error repair tables considering the tradeoff between time and space.
We classify the informations of the tables into several categories, that is, the informations for LR(0) items, for the paths between two nonterminals in L-relation, and for the sates to traverse to find insertion strings. Furthermore, we construct various error repair tables with the KAIST Parser Generating System and test their practicality by processing errors with the error repair parsers adopting them. According to the results, we can decide the extent of informations in error repair tables, for the error repair parsers to be more efficient.
