Pattern matching for the masses using custom notations

Abstract

For many programmers, the notion of “pattern matching” evokes nothing more than regular expressions for matching unstructured text, or technologies such as XPath to match semi-structured data in XML. This common perception of pattern matching is partly due to the success of regular expressions and XPath, which are supported in many popular programming languages today, either as standard libraries or as part of the language. But it is also due to the fact that many programmers never used another elegant form of pattern matching—on structured data, i.e., the native data structures of a programming language. This form of matching is common in functional or logic languages used in the research, but unfortunately much less used in the software industry. It is indeed very surprising that none of the popular languages in use today support, in their standard form, a nearly general form of structured data matching, decades after this technology has been discovered and continuously improved.This paper shows that programmers do not have to wait for next generation languages to integrate pattern matching, neither need they use non-standard pre-processors, thereby losing some advantages that are most important in an industrial setting: official support, compatibility, standardization, etc. Instead, pattern matching of native data in custom notations can be implemented as a minimalist library in popular object languages. Thus, some of the comfortable existing notations from logic languages can be reused, existing standard notations for structured data such as JSON (JavaScript Object Notation) can be smoothly extended to support pattern matching, and new notations can be designed.As in most library implementations of regular expressions, custom notation patterns are simply represented as strings. They can be used in two different modes: interpreted and compiled. This paper presents two open-source implementations of custom matching notations, for Java and JavaScript, exhibiting a reasonable overhead compared to other forms of pattern matching.