Overview
Field: Machine Learning Authors: Veronica Poweska, Ariana Oyanguren, Jessica Pourleyli, Sourena Khanzadeh, Manar Alalfi Published: 2026-09-25 arXiv: 2609.26952
Summary
Dependency conflicts in Python ecosystems arise from incompatible version constraints, missing packages, and undocumented compatibility relationships, causing many real-world code snippets to fail at execution. This paper presents PLLM+, a hybrid dependency-repair pipeline evaluated on the HG2.9K benchmark of 2,891 dependency-failing snippets.
PLLM+ prioritizes inexpensive deterministic steps before invoking LLM-based repair:
- Static AST-based interpreter inference to determine the required Python environment.
- Replay of historically successful dependency configurations from the competition-provided solutions database.
- Live PyPI validation of candidate package versions.
- On HG2.9K, PLLM+ resolved 1,500 of 2,891 snippets, compared with 1,169 for the PLLM baseline.
- Average runtime per snippet dropped from 368.7 seconds to 71.8 seconds.
- 1,495 of the 1,500 successful repairs were produced by replaying the solutions database; the LLM fallback contributed only 5.
When these steps do not resolve a case, the system falls back to a structured LLM-based repair loop featuring typed error classification and Proposer/Critic agents.
Key Results
Takeaway
Under this benchmark's setting, deterministically reusing verified dependency configurations is a simple and highly effective strategy, while LLM-based repair serves mainly as a secondary fallback for cases outside the known-configuration coverage.
--- *Auto-collected on 2026-09-25*