GAMMA: Revisiting Template-based Automated Program Repair via Mask Prediction

Running environment

The Python libraries transformers and pytorch are required to run the code.

pip install transformers
pip install pytorch

Fix templates

Fix tempaltes are defined in the code in the repository fixer-demo. To get the buggy lines which are masked according to different templates, run Main.java under the path fixer-demo/src/main/java.

Mask prediction

The code related to mask prediction task is in the repository patchGeneration. To generate patches for the bugs, first download the datasets Defects4J and QuixBugs from here and here.

Then run getContext.py to access the context of the buggy code.

Finally run unixcoder_repair.py, codebert_repair.py, and gpt_repair.py respectively to fill the masks with UniXcoder, CodeBERT and ChatGPT.