Architecture Decision: Choice of Poetry for Dependency Management - surapuramakhil-org/Job_search_agent GitHub Wiki
Context
When developing the job-hunt-assistant project, we needed a robust and reliable tool for managing dependencies and packaging. The current pip-based setup has two critical issues:
- Requirements.txt inconsistencies across different systems
- Slow installation times (120+ seconds) impacting CI/CD pipeline efficiency
Decision
We chose Poetry as our dependency management solution primarily because of its platform-independent lockfile generation. This is crucial for our open-source project where contributors work across different operating systems.
Key Advantages of Poetry:
- Platform-Agnostic Dependency Resolution: Generates consistent lockfiles that work across all operating systems
- Deterministic Builds: Ensures all contributors and CI environments use identical dependencies
- Project Metadata Management: Single
pyproject.tomlfile handles all project configurations - Built-in Publishing Tools: Streamlined package publishing to PyPI
Alternatives Considered
uv
While uv offers impressive features, it has critical limitations for our use case:
Advantages
- 10-20x faster installation speeds compared to pip
- Built-in Python version management
- Simplified toolchain replacing multiple tools
Deal-breaking Limitations
- Platform-dependent lockfile generation
- Different lockfiles needed for different operating systems
- Additional configuration required for cross-platform compatibility
- Requires building packages to determine metadata on each platform