RLRep

RLRep is a project to automatically generate program repair recommendation in the field of smart contracts for given code snippets with their contexts. The source code and dataset are opened.

Introduction

multistep_RLRep.py: the model framework for reinforcement learning and its implementation

src/, config/, utils/ and smartBugs.py: integrated the smartBugs tool (Ferreira et al.).

similarity_compute.py and FastText/(download from Zenodo because of Github upload size limit): by FastText library, compute the similarity (proposed by Gao et al.) between the generated contract and the buggy contract. (one of the modules that make up the reward function)

entropy_compute.py and entropy_compute/(download from Zenodo because of Github upload size limit): by the concept of entropy proposed by Ray et al., compute the entropy of the generated contract. (one of the modules that make up the reward function as well)

utils2.py: some useful methods of reward function and reinforcement learning.

top300_identifier_dict.txt: top-300 most frequent tokens in the source code.

solidityparser/: the Solidity lexer and parser built on top of ANTLR.

code2ast.js and node_modules/: convert source code to a preorder traversal sequence of AST.

genetic.py: the implement of the search-and-genetic-algorithm-based smart contract repair approach proposed by Yu et al.

dataset_vul.tar.gz: unzip it to get the folder dataset_vul/. It includes full_contract_dataset/ (853 vulnerable smart contracts), contract/(the source code of the buggy contract labeled with fault location), ast/(the preorder sequence of the abstract syntax tree of the buggy function), threelines-tokenseq/(the previous line, the next line and the buggy line) and repair_contract/ (the correct generated patches).

requirements.txt: a file listing all the dependencies for RLRepair

Usage

1. Install packages needed using pip:

pip install -r requirements.txt

2. Unzip dataset_vul.tar.gz

3. make sure that all input files are ready: (you can refer to the format of our input files in dataset_vul/newALLBUGS/)

• mapping (map source token to index): code_w2i.pkl, code_i2w.pkl, ast_w2i.pkl and ast_i2w.pkl
• first input: threelines-tokenseq/
• second input: ast/
• data for pretraining: pretrain/ and pretrain_label/
• data for validation: validation/

4. training and validation

python main.py [model_name] [dataset_path]
# for example:
python main.py multistep_RLRep dataset_vul/newALLBUGS
# or
python main.py mutation dataset_vul/newALLBUGS

5. result

At last, result can be got in dataset_vul/newALLBUGS/validation/result/.