An approach to safely evolve preprocessor-based C program families.

Abstract

Since the 70s, the C preprocessor is still widely used in practice in a numbers of projects, including Apache,Linux ,and Libssh, totail or systems to different platforms and application scenarios. In academia,however, the preprocess or has received strong critic is msinceatl east the early 90s. Researchers have criticized its lack of separation of concerns, its proneness to introduce subtle errors, and its obfuscation of the source code. To better understand the problems of using the C preprocessor, taking the perception of developers into account, we conducted 40 interviewsandasurveyamong 202 developers. We found that developers deal with three common problems in practice: configuration-related bugs, combinatorial testing, and code comprehension. Developers aggravate these problems when using undisciplined directives (i.e., bad smells regarding preprocessor use), which are preprocessor directives thatdo notrespect thesyntactic structureof thesource code. To safely evolve preprocessor based program families, we proposed strategies to detect configuration-relatedbugs and bad smells, and a set of 14 refactorings to remove bad smells. To better deal with exponential configuration spaces, our strategies uses variability-aware analysis that considers the entire set of possible configurations, and sampling, which allows to reuse C tools that consider only one configuration at a time to detect bugs. To propose a suitable sampling algorithm, we compared 10 algorithms with respect to effort (i.e., number of configurations to test) andbug-detection capabilities (i.e.,numberofbugs detected in the sampled configurations). Based on the results, we proposed a sampling algorithm with an useful balance between effort and bug-detection capability. We performed empirical studies using a corpus of 40 C real-world systems. We detected 128 configuration-related bugs, submitted 43 patches to fix bugs not fixed yet, and developers accepted 65% of the patches. The results of our survey show that most developers prefer to use the refactored (i.e., disciplined) version of the code instead of the original code with undisciplined directives. Furthermore, developers accepted 21 (75%) out of 28 patches submitted to refactor undisciplined into disciplined directives. Our work presents useful findings for C developers during their development tasks, contributing to minimize the chances of introducing configuration-related bugs and bad smells, improve code comprehension, and guide developers to perform combinatorial testing.