LHDiff

Setup

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Run

python3 lh-diff/lh-diff.py

Output goes to output/ as XML files.

File Structure

Project Structure

├── lh-diff/
│   └── lh-diff.py      # LHDiff implementation
├── datasets/
│   └── provided/       # Test dataset + ground truth
│   └── new/            # Our dataset
│   └── custom/         # add new dataset here...
├── output/             # Generated XML mappings
└── requirements.txt

Tunable Parameters

In lh-diff.py:

Variable	Default	Description
alpha	0.6	Weight: content (higher) vs context (lower)
threshold	0.6	Minimum similarity to match
k	15	SimHash candidates per line
window	4	Context lines for similarity

Adding a New Dataset

1. Add files to datasets/your_dataset/ as Name_1.java, Name_2.java, etc.

2. In lh-diff.py, add your dataset config:

custom_path = "./datasets/your_dataset/"
custom_files = ["Name1", "Name2", ...]
custom_dataset = [
    (i + 1, f, sorted(glob.glob(f"{custom_path}{f}_*")))
    for i, f in enumerate(custom_files)
]

3. Set selected_dataset = custom_dataset

Bonus: Bug-Introducing Change Detection

Uses the SZZ algorithm to identify which commits introduced bugs that were later fixed.

Run

python3 bonus/bonus.py /path/to/git/repo

How it works

1. Scans commit messages for bug-fix patterns (fix, bug, closes #, etc.)
2. Shows recent fix commits and lets you choose one
3. Uses git blame to trace deleted/modified lines back to their original commit
4. Reports which commits introduced the buggy code