Units-of-Measure Correctness in Fortran Programs

Abstract

Much of mathematics' use in science revolves around measurements of physical quantities, both abstractly and concretely. Such measurements are naturally classified by their dimension, i.e. whether the measurement is of distance, energy, time, and so on. Dimensionality is further refined by the units-of-measure (or units for short) of a measurement e.g., metres, Joules, seconds; thereby, units-of-measure distinguish magnitudes from each other, giving additional meaning. Despite their extensive use in the practice of science, units-of-measure do not see widespread adoption in tools for scientific computing. Here we demonstrate how our freely available and open-source tool, named CamFort, provides a low-effort and automated way of detecting mismatched units-ofmeasure in code. This feature of CamFort is an example of a lightweight, non-binding specification and analysis tool that can help find bugs in programs before they strike. We hope that, in general, these kinds of program analysis tool will become more widely used by scientists to save time and reduce grief during the process of development, as well as increasing confidence in results of numerical models.

Making legacy Fortran code type safe through automated program transformation

“…There is also some work on the formalisation of Fortran 95 semantics using VDM [15] but there is no publication on the final outcome. Specifically with regards to the type system, the only work that we are aware of is on the extension of Fortran 90 types with an attribute reflecting the unit of measurement [4]. According to our survey, a formalisation of the FORTRAN 77 type system or an analysis of its type safety has not been reported before.…”

Guiding user annotations for units-of-measure verification

“…Checking the consistency of units-of-measure or dimensionality in equations is a method long employed by scientists [8]. However, units checking has not been widely adopted in scientific programming despite the familiarity of the technique and a multitude of proposed systems for verification of units-of-measure [2,5,6,11,12,15] (many influenced by the work of Kennedy [7] on expressing units-of-measure as a type system). One explanation for this lack of adoption is the burden of adding units annotations to existing codebases, which may be large [14].…”