Cross-language generalization of reverse-transformation bug synthesis

Establish whether the SWE-at-Scale learned reverse-transformation paradigm generalizes to programming languages beyond Python that have different syntax, semantics, and software development practices.

Background

SWE-at-Scale is evaluated exclusively on Python repositories and learns bug-inducing transformations by reversing historical software fixes. The paper notes that programming languages differ in syntax, semantics, and software development practices, which may affect whether transformations learned from Python repositories can be applied reliably elsewhere.

Determining cross-language generalization would establish whether bug-inducing models based on the reverse-transformation paradigm can support scalable synthesis of realistic software-engineering tasks across multiple programming languages, rather than remaining limited to Python.

References

First, our current study focuses exclusively on Python repositories, and it remains unclear whether the learned reverse-transformation paradigm generalizes to other programming languages with different syntax, semantics, and software development practices.

— SWE-at-Scale: Scalable Issue Synthesis from Historical Bug Fixes  (2610.21359 - Chen et al., 15 Oct 2026) in Section Discussion, paragraph “Limitations and future directions”