Engineering a Converter Between Two Domain-Specific Languages for Sorting

摘要

Part of the ecosystem of applications running on mainframe computers is the DFSORT program. It is responsible for sorting and reformatting data (amongst other functionalities) and is configured by specifications written in a Domain-Specific Language (DSL). When migrating such sort workloads off from the mainframe, the SyncSort product is an attractive alternative. It is also configured by specifications written in a DSL but this language is structured in a radically different way. Whereas the DFSORT DSL uses an explicit fixed pipeline for processing, the SyncSort DSL does not. To allow DFSORT workloads to run on SyncSort we have therefore built a source-to-source translator from the DFSORT DSL to the SyncSort DSL. Our language converter performs abstract interpretation of the DFSORT specification, considering the different steps in the DFSORT pipeline at translation time. This is done by building a graph of objects and key to the construction of this graph is the reification of the records being sorted. In this paper we report on the design and implementation of the converter, describing how it treats the DFSORT pipeline. We also show how its design allowed for the straightforward implementation of unexpected changes in requirements for the generated output.