Artifacts of the paper under review by ISSTA

Artifacts of the paper under review by ISSTA

Welcome to the online repository of the ISSTA 2024 Submission 1305. We release the source code, dataset, and outputs of our study.

Getting Started

Requirements

To run the project, please ensure that you have the following dependencies installed:

    pytorch=2.0.0;

    torchvision=0.15.1;

    torchaudio;

    datasets==1.16.1;

    transformers==4.21.1;

    tensorboard==2.12.2;

    tree-sitter==0.19.1;

    nltk=3.8.1;

    scipy=1.10.1;

You can either install these requirements manually or execute the following script:

bash scripts/setup.sh

Dataset

We perform our experiments on following datasets:

    * CodeSearchNet from [here](https://github.com/github/CodeSearchNet).

    * BFP dataset from Tufano et al.'s TOSEM'19 paper.

    * Defects4J dataset from René et al.'s ISSTA'14 paper.

    * VulRD from Fu et al.'s FSE'22 paper.

Download the pre-training dataset:

python Dataset\pre-training\download.py

Pretrain the model

bash scripts/pre-train.sh -g [GPU_ID]

Task 1: bug repair

bash scripts/finetune_d4j.sh -g [GPU_ID]

If the evaluation during fine-tuning takes too long, you can adjust the "--evaluate_sample_size" parameter. This parameter refers to the number of cases in the validation set during evaluation.

To evaluate the performance of a specific checkpoint, add the flag "-e" followed by the checkpoint path.

Task 2: vulnerability repair

bash scripts/finetune_vul.sh -g [GPU_ID]

If the evaluation during fine-tuning takes too long, you can adjust the "--evaluate_sample_size" parameter. This parameter refers to the number of cases in the validation set during evaluation.

To evaluate the performance of a specific checkpoint, add the flag "-e" followed by the checkpoint path.

Outputs

The outputs of our tool are stored in the `outputs` folder.