PLLM+
A hybrid Python dependency repair pipeline that employs a layered, cost-ascending strategy: AST static analysis, historical configuration replay, live PyPI validation, and LLM fallback. It resolves Python dependency conflicts through a Proposer/Critic dual-agent collaboration architecture.
Timeline (last 90 days)
PLLM+ is a hybrid dependency repair pipeline designed to resolve Python dependency conflict issues.
PLLM+ was proposed in an arXiv paper (arXiv:2609.26952v1).
PLLM+ adopts a cheap-first, expensive-last tiered repair strategy, prioritizing low-cost deterministic steps and using LLMs only as a last resort.
The stages of the PLLM+ pipeline, ordered from lowest to highest cost, include static AST interpreter inference, historical successful configuration replay, real-time PyPI validation, and finally falling back to an LLM structured repair loop.
The LLM fallback component of PLLM+ introduces typed error classification and performs repairs through a Proposer/Critic dual-agent collaboration.
On the HG2.9K benchmark, PLLM+ resolved 1,500 snippets while the baseline method PLLM resolved only 1,169, representing an improvement of approximately 28%.
PLLM+ reduced the average processing time per snippet from 368.7 seconds to 71.8 seconds, a speedup of more than 5×.
Of the 1,500 successful repairs by PLLM+, 1,495 came from configuration replay from the solution database, with the LLM fallback contributing only 5 additional repairs.
The configuration replay strategy works well in scenarios with pre-existing answers; in real-world open-source projects without a ready-made solution library, the relative contribution of LLM-based repair may rise significantly.
All Facts (9)
PLLM+ is a hybrid dependency repair pipeline designed to resolve Python dependency conflict issues.
50%UnverifiedPLLM+ was proposed in an arXiv paper (arXiv:2609.26952v1).
50%UnverifiedPLLM+ adopts a cheap-first, expensive-last tiered repair strategy, prioritizing low-cost deterministic steps and using LLMs only as a last resort.
50%UnverifiedThe stages of the PLLM+ pipeline, ordered from lowest to highest cost, include static AST interpreter inference, historical successful configuration replay, real-time PyPI validation, and finally falling back to an LLM structured repair loop.
50%UnverifiedThe LLM fallback component of PLLM+ introduces typed error classification and performs repairs through a Proposer/Critic dual-agent collaboration.
50%UnverifiedOn the HG2.9K benchmark, PLLM+ resolved 1,500 snippets while the baseline method PLLM resolved only 1,169, representing an improvement of approximately 28%.
50%UnverifiedPLLM+ reduced the average processing time per snippet from 368.7 seconds to 71.8 seconds, a speedup of more than 5×.
50%UnverifiedOf the 1,500 successful repairs by PLLM+, 1,495 came from configuration replay from the solution database, with the LLM fallback contributing only 5 additional repairs.
50%UnverifiedThe configuration replay strategy works well in scenarios with pre-existing answers; in real-world open-source projects without a ready-made solution library, the relative contribution of LLM-based repair may rise significantly.
50%