Details

Autor(en) / Beteiligte

• Blasi, Arianna
• Stulova, Nataliia
• Gorla, Alessandra
• Nierstrasz, Oscar

Titel

RepliComment: Identifying clones in code comments

Ist Teil von

• The Journal of systems and software, 2021-12, Vol.182, p.111069, Article 111069

Ort / Verlag

Elsevier Inc

Erscheinungsjahr

2021

Link zum Volltext

Quelle

Elsevier ScienceDirect Journals

Beschreibungen/Notizen

• Code comments are the primary means to document implementation and facilitate program comprehension. Thus, their quality should be a primary concern to improve program maintenance. While much effort has been dedicated to detecting bad smells, such as clones in code, little work has focused on comments. In this paper we present our solution to detect clones in comments that developers should fix. RepliComment can automatically analyze Java projects and report instances of copy-and-paste errors in comments, and can point developers to which comments should be fixed. Moreover, it can report when clones are signs of poorly written comments. Developers should fix these instances too in order to improve the quality of the code documentation. Our evaluation of 10 well-known open source Java projects identified over 11K instances of comment clones, and over 1,300 of them are potentially critical. We improve on our own previous work Blasi and Gorla (2018), which could only find 36 issues in the same dataset. Our manual inspection of 412 issues reported by RepliComment reveals that it achieves a precision of 79% in reporting critical comment clones. The manual inspection of 200 additional comment clones that RepliComment filters out as being legitimate, could not evince any false negative. •Comment clones can be a bad practice just as code clones•They may represent shallow documentation, or copy-and-paste mistakes•A code entity may end up being documented with a comment that does not actually document its role and behavior•Across 10 popular open-source Java projects, we found thousands of potential issues•Comment clones do not seem to correlate with code clones: The two problems must currently be tackled by different techniques