Learning a model of Git using AALpy.

Setup

This section describes how to setup Git, in order to be learnable with AALpy. Instead of using a remote connection/repository, such as on GitHub or GitLab, we will create a bare repository locally. Experiments are performed locally for efficienty, and would need to be slightly adapted to learn directly with the remote git server.

Prerequisites: An installation of Git and following Python dependencies:

pip install -r requirements.txt

Usage

python3 git_learning.py creates two repos (one bare, one normal) in the ./tmp directory, and learns the git model with the chosen input alphabet (see source code). That script can be parametarized to learn certain configuration of Git, and learned model will be displayed and saved to a file. You can either learn models of GitCLI or GitPython, with or without the --allow-empty flag, and with any input alphabet found in the corresponding file.

Experiment runner

To execute all experiments, run:

python experiment_runner.py

All learned models and experiment statistics can be found in the models/ folder.

Differences between learned models

Following are some of the found differences with the interface_comparison script:

Inputs        : ('create_f0', 'add_f0', 'delete_f0', 'add_f0')
CLI           : ['PASS', 'PASS', 'PASS', 'PASS']
GitPython     : ['PASS', 'PASS', 'PASS', 'FAIL']

Inputs        : ('checkout_master')
CLI           : ['FAIL']
GitPython     :  ['PASS']

Inputs    : ('create_f0', 'add_f0', 'delete_f0', 'commit', 'create_branch', 'checkout_branch', 'delete_f0')
CLI       : ['PASS', 'PASS', 'PASS', 'PASS', 'PASS', 'PASS', 'FAIL']
GitPython :['PASS', 'PASS', 'PASS', 'PASS', 'PASS', 'PASS', 'PASS']

Inputs    : ('create_f0', 'commit', 'add_f0', 'create_branch', 'commit', 'checkout_branch', 'create_f0', 'checkout_master')
CLI       : ['PASS', 'PASS', 'PASS', 'PASS', 'PASS', 'PASS', 'PASS', 'FAIL']
GitPython : ['PASS', 'PASS', 'PASS', 'PASS', 'PASS', 'PASS', 'PASS', 'PASS']