We are updating..

News - We apply our approach on Flaky Test Prediction, which gets a 1% performance improvement compared with Flakify: A Black-Box, Language Model-based Predictor for Flaky Tests. You can find the details below.

News - We release the tool based on Soot to extract the graphs from Java bytecode.

Reminding that GraphCode2Vec works with Java Bytecode so that all the handled source code should be compiled.

• Extract Graph tool

  • We implement the tool based on Soot. Soot Version is in pom.xml.
  • You can freely modify it.
  • You can use it to extract graph for your data.
  • compile mvn clean package
  • run command see the example, extractGraphs.sh

• Requirements/Environment

  • recommend that you use conda
  • see requirements.txt

• Download Dataset from Zenodo

  • put all downloaded files to the source folder
  • make sure you have enough disk space, around 100G, bash untar.sh

• Pretrained Model

  1. https://drive.google.com/file/d/1PGF6e56CQ4XAfZMEU2Jl3w1TW4Q-59MG/view?usp=sharing
  2. put it into the source folder and decompress

• Run experiments

  1. cd source/graphscripts_experiments/
  2. run bash job files
  3. reminding, it contains all trial experiment jobs. We will clean it later and keep only experiments in the paper.

• We apply our approach to predict the flaky tests, which gets a 1% performance improvement compared with Flakify: A Black-Box, Language Model-based Predictor for Flaky Tests. Please see the repository, https://github.com/Marvinmw/Flaky .

  1. We compile all the code.
  2. We filter the test cases that do not have .class files or failed to extract graphs, which results in losing some flaky tests.
  3. To compare, we rerun Flakify and make sure we use the same dataset.
  4. You can find all info in https://github.com/Marvinmw/Flaky .

• paper link, GraphCode2Vec: Generic Code Embedding via Lexical and Program Dependence Analyses

• License, Boost Software License 1.0