Scripts and data for quantitative analysis in the above paper. The goal of this work is to enable topological analysis of PR-Issue networks on GitHub via graph database queries, an image generation module, and many statistics scripts. This repository includes:
- Data scraped from over 50 large open source projects on GitHub and converted into a Neo4j data dump.
- Reusable Cypher (Neo4j query language) queries to surface instances of workflow types.
- Image generation modules written in Python to visualize workflow types and clusters of Issues and PRs, used during the paper's qualitative coding process.
- Interactive web-based project explorer to aggregate workflow types found in a project, used during developer interviews.
These were primarily used to support the qualitative aspects of this paper.
The raw data scraped from GitHub in Summer 2022, including the issues, the PRs, their links, and their comment data, is available as Pickle files, downloadable here. The OSF dataset also includes the Neo4J database dump as well as all generated workflow type instance images. The raw dataset is 639MB zipped, and the Neo4j dump is 51MB unzipped.
The Cypher queries can be used with the Neo4j query browser to explore the data. Each query specifies the layout of Issues and PRs in the graph structure associated with a workflow type and any constraints (e.g. authorship, creation timestamp) on the nodes involved. These queries can be run by importing the Neo4j data dump into the Neo4j browser and copying over the queries. More information is available in INSTALL.md.
The image generation modules will generate visualizations of the workflow type instances (highlighted in yellow) in context in their larger cluster of Issues and PRs. Images to visualize instances of a particular workflow type can be generated by running python -m generate_neo4j_images.generate_from_neo4j --cypher=cypher_scripts/[query].cypher --name=[folder name]. The results will be saved to the generate_neo4j_images/images/ folder. More information is available in INSTALL.md and an example of a generated image is available below.
The web-based project explorer shows similar information as the image generation modules, but allows users to drag around Issues and PRs and click on them to open their associated GitHub pages. An explanation of the selected workflow type match and its implications is also shown. A pre-built version of the project explorer can be found in interactive_html/. Open the folder with a browser and navigate to index.html to choose from a list of projects. More information is available in INSTALL.md and an example of a project page from the mlflow/mlflow project is shown below.
As well, the scripts used to generate the graphs in the paper (Figures 2, 3, and 4) are here. They can be run via python -m data_scripts.[script name] and their output images will be generated in misc_images/. See this documentation for more information on how to run them.
For documentation, see the docs/ folder. To install the required Python dependencies, run pip install -r requirements.txt.
- To get started with the graph database data, see this documentation.
- To learn more about the image generation module, see this documentation.
- The full database dump is available here. To import the database in your Neo4J instance, see the installation guide or follow the Neo4J documentation here.
- The Cypher queries are here. To run the queries and return the issues and PRs that are part of each workflow type, drag and drop the query over the query GUI.
- The scripts for component sampling are here and additional documentation is here.