Syntax Error Handling in Scannerless Generalized LR Parsers

This thesis is about a master's project as part of the one year master study 'Software-engineering'. This project is about methods for improving the quality of reporting and handling of syntax errors that are produced by a scannerless generalized left-to-right rightmost (SGLR) parser, and is done at Centrum voor Wiskunde en Informatica (CWI) in Amsterdam. SGLR is a parsing algorithm developed as part of Generic Language Technol- ogy Project at SEN1, one of the themes at CWI. SGLR is based on the GLR algorithm developed by Tomita. SGLR parsers are able to recognize arbitrary context-free grammars, which enables grammar modularization. Because SGLR does not use a separate scan- ner, also layout and comments are incorporated into the parse tree. This makes SGLR a powerful tool for code analysis and code transformations. A drawback is the way SGLR handles syntax errors. When a syntax error is detected, the current implementation of SGLR halts the parsing process and reports back to the user the point of error detection only. The text at the point of error detection is not necessarily the text that has to be changed to repair the error. This thesis describes three kinds of information that could be reported to the user, and how they could be derived from the parse process when an error is detected. These are: - The structure of the already parsed part of the input in the form of a partial parse tree. - A listing of expected symbols; those tokens or token sequences that are accept- able instead of the erroneous text. - The current parser state which could be translated into language dependent informative messages. Also two ways of recovering from an error condition are described. These are non-correcting recovery methods that enable SGLR to always return a parse tree that can be unparsed into the original input sentence. - A method that halts parsing but incorporates the remainder of the input into the parse tree. - A method that resumes parsing by means of substring parsing. During the course of the project the described approaches have been imple- mented and incorporated in the implementation of SGLR as used by the Meta- Environment, some fully, some more or less prototyped.