Skip to content
#graph neural networks Dataset Open access

DjangoDefects: A Multi-Granularity SZZ-Labeled Defect Dataset for Python

Oct 2026 · Zenodo (CERN European Organization for Nuclear Research)
Software Engineering Research

Abstract

DjangoDefects is an inter-release defect prediction benchmark for Python, built from the full version history of the Django web framework. It covers 24 consecutive releases (1.0 to 5.2, September 2008 to April 2025). For every release it provides defect labels at four granularities, all derived from a single SZZ run:- File level: 20,442 file-release instances, 28.4% buggy- Method level: 169,744 function and method instances with their source code, 8.3% buggy- Line level: 14,917 buggy source lines- AST-node level: 5.38 M nodes from a full DFS traversal of each affected file's AST, with typed parent edges (ready to load as a graph), 1.29% buggyA source snapshot of every release is included. All file paths, method spans, line numbers and node spans refer to the code exactly as it shipped in that release.Construction. 5,585 closed, fixed Bug tickets from Django's Trac issue tracker were linked to 4,801 fix commits through Fixed # references. Each fix was traced back to the commits that introduced the bug using SZZ Unleashed, with whitespace-insensitive blame and whitespace and comment filters, giving 6,306 bug records. Releases are assigned by git ancestry rather than by date, which accounts for Django forking each stable branch months before the release tag. Buggy lines are located in each release snapshot by blame identity.Contents.- file_level.zip, method_level.zip, line_level.zip, ast_level.zip: one CSV per release- source_code.zip: every file under django/ at each release tag- code.zip: the pipeline that generated the dataset, the cached ticket-to-commit links, and instructions for reproducing it- README.md: full documentation, including schemas, statistics and threats to validityIntended uses.- Cross-version defect prediction at file and method level- Fine-grained defect localisation at line and AST-node level- Training models that use code structure, such as graph neural networks on the ASTs or code language models on method sourceKnown limitations.- Labels for the most recent releases (about 5.0 to 5.2) are incomplete, because many of their bugs have not been reported or fixed yet.- File and method labels cover every release in which a bug is present. Line and node labels cover only the release that introduced it.- See README.md §7 for the other threats to validity. Licence. The labels, documentation and code are released under CC BY 4.0. The included Django source code is © Django Software Foundation and individual contributors, redistributed under its BSD-3-Clause licence (LICENSE-DJANGO).

View source

Similar papers

#computer vision Conference Aug 2008

Scrum in a Multiproject Environment: An Ethnographically-Inspired Case Study on the Adoption Challenges

Agile methods continue to gain popularity. In particular, the Scrum method appears to be on the verge of becoming a de-facto standard in the industry, leading the so called Agile movement. While there are success stories and recommendations, there is little scientifically valid evidence of the challenges in the adoptio...

A. Marchenko, P. Abrahamsson · 59 citations · ⚡11
#computer vision Open access Sep 2012

Making the leap to a software platform strategy: Issues and challenges

A comprehensive taxonomy of the challenges faced when a medium-scale organization decided to adopt software platforms is provided, namely: business challenges, organizational challenges, technical challenges, and people challenges.

Yaser Ghanam, F. Maurer, P. Abrahamsson · 41 citations · ⚡3
#machine learning Open access Mar 2024

Integration of molecular coarse-grained model into geometric representation learning framework for protein-protein complex property prediction

MCGLPPI, a novel geometric representation learning framework that combines graph neural networks (GNNs) with the MARTINI molecular coarse-grained (CG) model to predict overall PPI properties accurately and efficiently, offers an effective and efficient solution for PPI overall property predictions.

Yang Yue, Shu Li, Yihua Cheng et al. · 15 citations

PepPCBench is a Comprehensive Benchmarking Framework for Protein-Peptide Complex Structure Prediction

PepPCBench enables a robust evaluation of PFNN-based methods and supports their continued development for peptide-protein structure prediction, and highlights the influence of peptide length, conformational flexibility, and training set similarity on prediction accuracy.

Si-Long Zhai, Huifeng Zhao, Ji-Ke Wang et al. · 13 citations · ⚡1
#data science Open access Oct 2026

Less can be more for predicting properties with large language models

Abstract Predicting properties from coordinate–category data—sets of vectors paired with categorical information—is fundamental to computational science, manifesting in materials science as predicting formation energies or elastic moduli from crystal structures. While large language models (LLMs) have increasingly been...

Nawaf Alampara, Santiago Miret, Kevin Maik Jablonka · 13 citations

Related blog posts

Microsoft Research Blog Jul 13, 2026

Verifying Rust cryptography in SymCrypt, from standards to code

Cryptographic code supports vital protections in modern computing systems. Learn how a new method helps verify code as developers write it while preserving speed and adaptability as it gets implemented and evolves. The post Verifying Rust cryptography in SymCrypt, from standards to code appeared first on Microsoft Research.

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.