DeepMutation is a mutation tool that learns how to generate mutants that are similar to real bugs.

At its core, DeepMutation relies on a Neural Machine Translation model that was trained to translate fixed code into buggy code. DeepMutation has learned from thousands of real world bug-fixes performed by developers, how to generate mutants that resemble real buggy code.

See Tufano, Michele, et al. "Learning how to mutate source code from bug-fixes." 2019 IEEE International Conference on Software Maintenance and Evolution (ICSME). IEEE, 2019.

1. Installation
   a. Requirements
   b. Setting up the environment
   c. Building
2. Running
3. How it works
4. Credits
5. Paper References

• java 1.8
• python 3.4-3.6
• pip
• seq2seq
  • tensorflow

For the study, we used versions:

• python 3.6
• seq2seq 0.1
  • tensorflow 1.3

We also used defects4j for testing. We provide defects4j_checkout.py for easier checking out following a valid defects4j installation:

$ ./defects4j_checkout.py [-f] [proj_name]...

The recommended way to setup the environment is with venv:

$ python -m venv venv
$ source venv/bin/activate
$ pip install -r requirements.txt

We include the script check_env.sh for verification.

$ ./check_env.sh

Note: at the time of writing, a bug is present in seq2seq; this script provides the instructions for the related bugfix.

DeepMutation can be compiled into DeepMutation.jar with:

$ ant

DeepMutation runs using a configuration file:

$ java -jar DeepMutation.jar config_file

An example is provided.

This will generate and test mutants.

DeepMutation is a mutation infrastructure that allows to generate mutants that resemble real bugs. At its core, DeepMutation relies on a Neural Machine Translation model that learned from more than 700 thousands real bug-fixes mined on GitHub, how to translate fixed code into buggy code. Thus, the trained model is used to translate a method into a mutant.

DeepMutation starts by receiving the system to mutate and a configuration file. Next, the following modules are executed:

• The Extractor, extracts the raw methods from the system;
• The Abstractor, abstracts these raw methods into abstract methods, using the idioms specified in an input file. A mapping file to reconstruct the raw methods is stored.
• The Mutator invokes the NMT model to generate abstract mutants;
• The Translator uses the mapping file to translate abstract mutants in concrete mutants;
• Finally, the mutants are injected in the system, compiled and tested.

The infrastructure of DeepMutation was created by Jason Kimko (@jskimko) and Shiya Wang (@sywang15). The underlying model was created and trained by Michele Tufano and Cody Watson, based on seq2seq.

TODO ICSME + ICSE tool demo submission