Abstract:
Nowadays, programmers willing to start optimising their code must undergo a lengthy interaction with dedicated profiling tools. This paper proposes as an alternative to make compilers generate feedback messages aimed at explaining how they understand the code, and how it could be improved. The study aims at foreseeing the technical integration of feedback notifications in modern compilers, as well as sketching how Integrated Development Environments (IDE) would display them.
A first comparison of three related works enables the core differentiators to be highlighted: letting the compiler inform where code is actually fine and does not need any refinement, displaying the notifications along the relevant source lines rather than in a separate interface, insisting on the absence of artificial intelligence, and evaluating the importance of each message to filter in a handful. Then, a preparatory field study is carried to observe different programmers and poll their receptiveness to a compiler feedback. The findings relate the usefulness of optimisations’ suggestions to fit where users lack expert knowledge, the existence of dormant interrogations calling for serendipitous information retrieval, and the avoidable mistakes inherent to Message of the Day windows.
Three prototypes are designed to embody three distinct approaches, using Web tools to provide an appearance close to code editors along with decent interactivity. With the help of a new user study with the prototypes, a final set of refinements is discussed so as to shape a coherent result and differentiate it further: users can create and share sets of feedback messages to supplement the ones included in their compiler, a list of rules is provided to help designers compose the messages, an emphasis is laid on transparency to help exhibit the absence of artificial intelligence, and the heuristic used to evaluate the importance of messages is sketched.
A first comparison of three related works enables the core differentiators to be highlighted: letting the compiler inform where code is actually fine and does not need any refinement, displaying the notifications along the relevant source lines rather than in a separate interface, insisting on the absence of artificial intelligence, and evaluating the importance of each message to filter in a handful. Then, a preparatory field study is carried to observe different programmers and poll their receptiveness to a compiler feedback. The findings relate the usefulness of optimisations’ suggestions to fit where users lack expert knowledge, the existence of dormant interrogations calling for serendipitous information retrieval, and the avoidable mistakes inherent to Message of the Day windows.
Three prototypes are designed to embody three distinct approaches, using Web tools to provide an appearance close to code editors along with decent interactivity. With the help of a new user study with the prototypes, a final set of refinements is discussed so as to shape a coherent result and differentiate it further: users can create and share sets of feedback messages to supplement the ones included in their compiler, a list of rules is provided to help designers compose the messages, an emphasis is laid on transparency to help exhibit the absence of artificial intelligence, and the heuristic used to evaluate the importance of messages is sketched.