Replication package for the paper:
An empirical study on test fixture usage by coding agents on open source software João Almeida, Andre Hora
FixtureDB collects test fixtures from open-source repositories. It compares fixtures written by coding agents with fixtures written by humans before LLM coding tools existed. The code covers Python, Java, JavaScript and TypeScript.
- Dataset A (agent). Fixtures from commits made by coding agents, in repositories that have agent configuration files.
- Dataset C (human, pre-LLM). Fixtures from repositories created from 2016 to 2020, read at their last commit on or before 2020-12-31.
Details are in the dataset card.
pip install -r requirements.txt
python -m collection --help
python -m collection toy --dataset a --repos 5
pytest tests/The toy run writes to toy-dataset/. It does not touch datasets/ or db/.
Some agent-detection heuristics (co-authored-by trailer parsing, config-file patterns for individual agents) were checked and improved against agent-mining, a catalog of AI coding agent detection heuristics maintained by the LaBRI research group.