Typing Legacy COBOL Code

Abstract

Maintenance of COBOL applications that still exist and work today is an open issue for many companies that have not yet undertaken the crucial decision of migrating to a modern development platform. And even those who did, most likely had to face a major challenge: understanding what those million lines of code do and what business processes they originally implemented. Automating the task of reconstructing the business logic of programs somehow is hard nonetheless: COBOL code is difficult to even get parsed - let alone applying Program Understanding techniques based on sophisticate deductions over information inferred from the code. Here we propose a system based on the translation of COBOL into a simpler and cleaner language and a type system capable of inferring types of program variables. Being COBOL a language in which variable reuse has been a widespread practice for decades, strong typing rules in the usual sense wouldn’t simply fit, therefore our system provides special flow types for tracking multiple types a variable may assume and follows the program control-flow until no more type-changes occur for variables in the typing context. Alongside it detects also a number of error-prone situations, type mismatches and data corruptions due to misalignment or misfit in variable reuse when types have incompatible in-memory representations, while still guaranteing the typing process does not halt.