Checking C++ codes for compatibility with operator overloading

Operator overloading allows the semantic extension of existing code without the need for sweeping code changes. For example, automatic differentiation tools in C++ commonly use this feature to enhance the code with additional derivative computation. To this end, a floating point data type is changed to a complex user-defined type. While conceptually straightforward, this type change often leads to compilation errors that can be tedious to decipher and resolve. This is due to the fact that the built-in floating point types in C++ are treated differently than user-defined types, and code constructs that are legal for floating point types can be a violation of the C++ standard for complex user-defined types. We identify and classify such problematic code constructs and suggest how the code can be changed to avoid these errors, while still allowing the use of operator overloading. To automatically flag such occurrences, we developed a Clang-based tool for the static analysis of C++ code based on our assessment of constructs problematic in operator overloading for numeric types. It automatically finds instances of problematic code locations and prints Lint-like warning messages. To showcase the relevance of this topic and the usefulness of our tool, we consider the basic routines of the OpenFOAM CFD software package, consisting of 1,476 C++ source and header files, for a total of over 150,000 lines of code. Altogether, we found 74 distinct occurrences of problematic code constructs in 21 files. As some of these files are included in over 400 different locations in the OpenFOAM base, errors in these files create a torrent of error messages that often are difficult to comprehend. In summary, the classification of problematic instances aids developers in writing numerical code that is fit for operator overloading and the tool helps programmers that augment legacy code in spotting problematic code constructs.