🤖 AI Summary
Legacy scientific codes written in Fortran-77—especially those incorporating nonstandard extensions (e.g., Cray pointers)—incur high maintenance costs, and existing compiler frameworks lack efficient support for their modernization. This paper introduces StmtTree, a lightweight and flexible Fortran code transformation toolkit. Its core innovation is the “statement tree” (StmtTree) intermediate representation, which unifies low-level APIs with a high-level, domain-specific mini-language (DSL), balancing expressiveness and usability. Built upon a custom Fortran parser, StmtTree enables precise, syntax-tree–level transformations. Evaluated on real-world legacy code, it demonstrates practical efficacy: complex refactorings—such as automatic removal of unused statements—are implemented in under 100 lines of Python, drastically reducing the development barrier and implementation cost for modernization tools. To our knowledge, StmtTree is the first Fortran transformation framework capable of performing such high-level semantic transformations within a hundred lines of code.
📝 Abstract
The Fortran programming language continues to dominate the scientific computing community, with many production codes written in the outdated Fortran-77 dialect, yet with many non-standard extensions such as Cray poiters. This creates significant maintenance burden within the community, with tremendous efforts devoted to modernization. However, despite the modern age of advanced compiler frameworks, processing and transforming old Fortran codes remains challenging. In this paper, we present StmtTree, a new Fortran code transformation toolkit to address this issue. StmtTree abstracts the Fortran grammar into statement tree, offering both a low-level representation manipulation API and a high-level, easy-to-use query and manipulation mini-language. StmtTree simplifies the creation of customized Fortran transformation tools. Experiments show that StmtTree adapts well to legacy Fortran-77 codes, and complex tools such as removing unused statements can be developed with fewer than 100 lines of python code.