Abstract:

With software scaling up continuously, logging mechanism has become an indispensable part in failure diagnosis area. A pretty similar symptom may be caused by various software bugs, and the most obvious evidence is always logging messages. Meanwhile, the development of most pieces of largescale software is affected by developers' personal habits rather than being guided by certain conventional specification, so logrelated analysis suffers in largescale software. The recognition of logging function plays a precondition role in log analysis and affects the results of log analysis directly. We propose a machine learning method to fill the gap that logging function recognition has not been paid attention by most existing logrelated works. Learning from widelyused software, we summary three logging functions, extract five common features to complement automated loggingfunction recognition tool iLog based on machine learning. Evaluations show that the recognition ability of iLog is 76 times of those  using key words. Additionally, 10fold crossvalidation shows that the FScore average is 0.93.

Key words: logging function, machine learning, static analysis, code quality, failure diagnosis