Grammatica Reference Manual

Error Recovery

Grammatica supports automatic recovery of both parse and analysis errors. Measures are also taken to reduce the number of dependency errors, i.e. errors caused by previously reported errors. The automatic recovery is not perfect, however, so in some circumstances errors are omitted or included, although they shouldn't have been.

Unexpected character errors (i.e. lexical errors) are handled by reporting the error and ignoring the character. The parser will also enter recovery mode, which means that no further errors will be reported until three tokens have been read correctly. In practice, this means that only the first character in an erroneous sequence will be reported, although all are ignored. In the figure below this is illustrated for a grammar file.

[email protected]#!

Figure 1. An erroneous character sequence for a grammar file. If this sequence occurs in a grammar file (uncommented), only the '$' character will be reported. The remaining characters are also erroneous, but the error messages for them will be supressed.

Unexpected token errors (i.e. syntax errors) are handled by reporting the error and ignoring tokens until a valid token is found. The parser also enters error recovery mode, as explained above. In the figure below the parser error recovery behaviour is illustrated.

Prod = "one" <<.>> "two" ;

Figure 1. An erroneous token in a grammar production. The regular expression token <<.>> is not allowed inside a production, and will cause an error. The parser will recover by ignoring all tokens until the ';' character.

Analysis errors are only reported as long as the parser has not encountered a lexical or syntactical error. In fact, after such an error has been encountered, no further calls to the analyzer are made at all. All analyzer errors encountered during parsing are reported, but when analyzing an existing parse tree some filtering is attempted to avoid dependency errors.

Due to the error recovery, care must be taken when implementing an analyzer. The callback methods will be called even though previous callbacks may have failed. In particular, this means that child nodes may be absent or reordered, and node values may be missing, making null pointer errors more probable. Helper methods that perform the necessary null checking have therefore been created in the base Analyzer class, and it is strongly recommended to use them.