A slicing-based approach for locating type errors

Abstract

The effectiveness of a type-checking tool strongly depends on the accuracy of the positional information that is associated with type errors. We present an approach where the location associated with an error message e is defined as a slice P e of the program P being type-checked. We show that this approach yields highly accurate positional information: P e is a program that contains precisely those program constructs in P that caused error e . Semantically, we have the interesting property that type-checking P e is guaranteed to produce the same error e . Our approach is completely language-independent and has been implemented for a significant subset of Pascal. We also report on experiments with object-oriented type systems, and with a subset of ML.