Abstract: Semantics-preserving program transformations (SPTs) are widely used to generate adversarial example attacks against deep learning-based models for code analysis tasks. This work studies 34 ...