fuzz4db
收藏DataCite Commons2026-04-23 更新2026-05-04 收录
下载链接:
https://data.mendeley.com/datasets/mjcvj7fc2m
下载链接
链接失效反馈官方服务:
资源简介:
Replication package for the paper:
"Fuzz4DB: A Practice of LLM-Agent-Guided Fuzzing for Databas Feature-Level Delta Testing", Chunling Qin, Yong Hu, Xiao Zhang,
Jinchuan Chen, Fangtao Gu, Baoxun Wang, Yuxing Chen, Anqun Pan,
Lixiong Zheng, ASE '26, October 12–16, 2026, Munich, Germany.
Contents:
- sql/ SQL test corpora (a subset of the generated test
cases) for the 9 open-source commits
- patches-generated/ Commit diff patches used for diff-cover analysis
- final_results/ Pre-computed coverage artifacts (LCOV traces,
LLVM profdata, diff-cover reports) per commit,
with summary manifest
(final_results.json / final_results.tsv)
- bug/ 63 open-source bug reports (DBMS, bug ID,
upstream link, root cause, trigger condition,
status)
- commits.json Commit metadata (hash, DBMS, feature, delta size)
- README.md Step-by-step instructions to reproduce
Table 1, Table 2, Table 3, and Table 4 in the paper
- *.sh Reproduction scripts (unpack, reproduce, verify,
show results)
Docker images are hosted on GitHub Container Registry (ghcr.io)
and pulled automatically by the scripts. No local image files
are included in this package.
Note: The Fuzz4DB source code is not included due to commercial
licensing restrictions.
Environment:
- OS: Linux (Docker host)
- Docker >= 20.10
- python3 >= 3.8
- Internet access to pull Docker images from ghcr.io
- For full execution-based reproduction (Mode 3):
- mysql client (for c3, c4, c10-c12)
- psql client (for c1, c2)
- mariadb client (for c5, c6)
- ≥16 GB RAM, ≥4 cores (≥8 cores recommended)
- ~500 MB disk space for this package; Docker images are
pulled on demand (~22 GB total)
Reproduced results:
- Table 1: Tested commits (commits.json)
- Table 2: Per-commit incremental line coverage (show_final_results.sh)
- Table 3: Bug counts by DBMS (bug/Fuzz4db_bug.csv)
- Table 4: Open-source bug summary (bug/Fuzz4db_bug.csv)
Estimated reproduction time:
- ~5 minutes to view packaged results
(Mode 1: show_final_results.sh)
- ~15 minutes to re-validate from bundled artifacts
(Mode 2: verify_final_results_from_artifacts.sh)
- ~2–3 hours for full execution-based rerun
(Mode 3: reproduce_coverage.sh, all 9 commits)
on a machine with ≥16 GB RAM and ≥8 cores.
提供机构:
Mendeley Data
创建时间:
2026-04-23



