Detecting and tracking code clones can ease various software development and maintenance tasks when changes in a code fragment should be propagated over all its copies. Several deep learning based clone detection models have appeared in the literature for detecting syntactic and semantic clones, and these models have widely been evaluated with the BigCloneBench dataset. However, the class imbalance and small number of semantic clones make BigCloneBench less ideal when interpreting the model performances. Sometimes researchers use a few other semantic clone datasets such as GoogleCodeJam, OJClone and SemanticCloneBench to understand a model's generalizability. To overcome the limitations of the existing datasets, recently a GPT-assisted large semantic and cross-language clone dataset GPT-CloneBench has been released, but it is not clear how all these models would compare and contrast in terms of these datasets. In this paper, we propose a multi-step evaluation approach for five state-of-the-art clone detection models leveraging existing bench-mark datasets including the recently proposed GPTCloneBench and exploiting the mutation operators to study the extent of these clone detection models' ability. More specifically, we examined the performance of three highly-performing single-language clone detection models (ASTNN, GMN, CodeBERT) that use various code representations (e.g., AST, flow augmented AST with graph matching network, and bidirectional encoder representation) for detecting semantic clones. In addition to using BigCloneBench, we tested them on SemanticCloneBench and GPTCloneBench, investigated their robustness under mutation operations, and examined them against cutting-edge cross-language clone detection tools (C4, CLCDSA) that are also known to learn semantic clones. While all single-language models showed high F1 scores for BigCloneBench, their performances varied quite differently (sometimes over 20%) when tested on SemanticCloneBench. Interestingly, the cross-language model (C4) consistently showed superior performance (around 7%) on SemanticCloneBench over other models and performed similarly for BigCloneBench and GPTCloneBench. On mutation-based datasets, C4 appeared to have a more robust performance (less than 1% difference) whereas single-language models showed high variability.

Software clones are often introduced when developers reuse code fragments to implement similar functionalities in the same or different software systems resulting in duplicated fragments or code clones in those systems. Due to the adverse effect of clones on software maintenance, a great many tools and techniques and techniques have appeared in the literature to detect clones. Many high-performing clone detection tools today are based on deep learning techniques and are mostly used for detecting clones written in the same programming language, whereas clone detection tools for detecting cross-language clones are also emerging rapidly. The popularity of deep learning-based clone detection tools creates an opportunity to investigate how known strategies that boost the performances of deep learning models could be further leveraged to improve the clone detection tools. In this paper, we investigate such a strategy, data augmentation, which has not yet been explored for cross-language clone detection as opposed to single language clone detection. We show how the existing knowledge on transcompilers (source-to-source translators) can be used for data augmentation to boost the performance of cross-language clone detection models, as well as to adapt single-language clone detection models to create cross-language clone detection pipelines. To demonstrate the performance boost for cross-language clone detection through data augmentation, we exploit Transcoder, which is a pre-trained source-to-source translator. To show how to extend single-language models for cross-language clone detection, we extend a popular single-language model, Graph Matching Network (GMN), in a combination with the transcompilers and code parsers (srcML). We evaluated our models on popular benchmark datasets. Our experimental results showed improvements in F1 scores (sometimes up to 3%) for the cutting-edge cross-language clone detection models. Even when extending GMN for cross-language clone detection, the models built leveraging data augmentation outperformed the baseline with scores of 0.90, 0.92, and 0.91 for precision, recall, and F1 score, respectively.