Natural and flexible error recovery for generated parsers

• Parser generators are an indispensable tool for rapid language devel- opment. However, they often fall short of the finesse of a hand-crafted parser, built with the language semantics in mind. One area where generated parsers have pro- vided unsatisfactory results is that of error recovery. Good error recovery is both natural, giving recovery suggestions in line with the intention of the programmer; and flexible, allowing it to be adapted according to language insights and lan- guage changes. This paper describes a novel approach to error recovery, taking into account not only the context-free grammar, but also indentation usage. We base our approach on an extension of the SGLR parser that supports fine-grained error recovery rules and can be used to parse complex, composed languages. We take a divide-and-conquer approach to error recovery: using indentation, erro- neous regions of code are identified. These regions constrain the search space for applying recovery rules, improving performance and ensuring recovery sug- gestions local to the error. As a last resort, erroneous regions can be discarded. Our approach also integrates bridge parsing to provide more accurate suggestions for indentation-sensitive language constructs such as scopes. We evaluate our ap- proach by comparison with the JDT Java parser used in Eclipse.

Ämnesord

NATURVETENSKAP  -- Data- och informationsvetenskap -- Datavetenskap (hsv//swe)

NATURAL SCIENCES  -- Computer and Information Sciences -- Computer Sciences (hsv//eng)

Publikations- och innehållstyp

kap (ämneskategori)

ref (ämneskategori)