DAppFL

A basic implementation of "DAppFL: Just-in-Time Fault Localization for Decentralized Applications in Web3", ISSTA 2024.

Installation

Please ensure that your Python version is above 3.9 :) And use following commands to install the dependency of DAppFL:

pip install -r requirements.txt
npm install

Quickly start

Let's try to locate the SushiSwap fault. Perform fault localization with DAppFL using the following command:

$ python locate.py \
  --net=Ethereum \
  --fault_txhash=0x90fb0c9976361f537330a5617a404045ffb3fef5972cf67b531386014eeae7a9 \
  --faultless_txhash=0x7df39084b561ee2e7809e690f11e8e258dc65b6128399acbacf1f2433308de6a,0xddd734c1f3e097d3d1cdd7d4c0ffae166b39992a1d055008bf6660b8c0b7582e,0x5c1d151599bbacc19a09dfee888d3be2ccf3e2fa781679b9e0970e18b3300e44

Params description

• net: indicate the blockchain network where dapp is deployed, supporting Ethereum and BNBChain
• fault_txhash: transaction hash that results in the fault, please , for join multiple hash if available
• faultless_txhash: transaction hash that has NOT resulting in the fault, please , for join multiple hash if available

You can get the output like this (where Top-0 is the true fault location):

loading model params...
collecting transaction executing data...
The following code snippets are most likely to cause faults, and the ranking is in descending order of faulty suspiciousness:
Top-0: fault function at contracts/SushiMaker.sol -> 0xe11fc0b43ab98eb91e9836129d1ee7c3bc95df50, offset is 1689:185
Top-1: fault function at contracts/SushiMaker.sol -> 0xe11fc0b43ab98eb91e9836129d1ee7c3bc95df50, offset is 3574:738
Top-2: fault function at contracts/SushiMaker.sol -> 0xe11fc0b43ab98eb91e9836129d1ee7c3bc95df50, offset is 4426:2288
Top-3: fault function at contracts/SushiMaker.sol -> 0xe11fc0b43ab98eb91e9836129d1ee7c3bc95df50, offset is 7944:173
Top-4: fault function at contracts/SushiMaker.sol -> 0xe11fc0b43ab98eb91e9836129d1ee7c3bc95df50, offset is 2988:109

File description

• algos: Contains some simple algorithms, such as graph diffusion.
• compiler: Contains various versions of the Solc compiler.
• daos: The data access object, for data processing.
• data: The contributed dataset.
• dataset: The code for data modelling.
• downloader: Crawl data from the blockchain clients.
• misc: Necessary JavaScript files, such as scripts for injecting RPC interfaces.
• models: The Graph neural network model.
• utils: Some small tools used in the project.
• main.py: The main file for starting and evaluating our method, and we save the output to file named main_output.

Train your own DAppFL

You can use main.py to train your DAppFL. There are some parameters for main.py:

• --data_path: a path for the dataset.
• --hidden_channels: the number of neurons for each layer of HGT, default is 32.
• --num_layers: the number of layers for HGT, default is 4.
• --num_heads: the number of attention heads for HGT, default is 1.
• --gpu: Enable GPU for training, default is False.
• --k_folds: Number of repeated experiments, default is 5.
• --epoch: Number of training rounds, default is 10.
• --batch_size: Batch size during training, default is 4.
• --lr: Learning rate of optimizer, default is 0.001.
• --weight_decay: Weight decay of optimizer, default is 5e-4.
• --p_norm: The Power of Graph Diffusion, default is 6.