Automatic generation of useful syntax error messages

Abstract Many current compilers produce in some situations wrong error messages that mislead the user and harm his confidence in the system. It is demonstrated that a reliable and efficient syntax error handling system may be produced automatically by a compiler generator from the BNF specification of the language, and without any effort by the language implementor. This result is achieved in three ways: Some errors may not be diagnosed without knowledge of the intentions of the programmer. Some compilers employ a sophisticated analysis that attempts to capture these intentions, but which is not always successful. Such an elaborate analysis is not employed here, and instead a list of all the legal corrections is displayed, so that the programmer may readily select the right one. The recovery symbols are selected by a ‘careful’ algorithm resulting in a high probability for correct error recovery. The ‘honest’ error messages show also the parts of the code which could not be analysed correctly because of errors, and where more errors may exist. Efficiency is achieved by computing the recovery sets once and for all at compiler generation time, so that a fast error recovery at program compilation time is achieved. Experiments with erroneous programs suggest that the method compares well with the best compilers that we have seen, and is specially good at the avoidance of wrong error messages.