Skip to content

adr-001-python-3.13.md

Latest commit

 

History

History
64 lines (47 loc) · 2.58 KB

adr-001-python-3.13.md

File metadata and controls

64 lines (47 loc) · 2.58 KB

ADR-001: Python 3.13 as Primary Language

Status: Accepted
Date: 2024-12
Deciders: Architecture Team, Ricardo Cataldi

Context

The accelerator requires a language that supports:

  • Async/await for concurrent I/O (Redis, Cosmos, Blob, Event Hubs, AI Search)
  • Rich ecosystem for AI/ML integration (Microsoft Agent Framework, Foundry SDK)
  • Fast iteration cycles for adapter/agent development
  • Strong typing for maintainability (Pydantic, type hints)
  • Enterprise support and Azure SDK maturity

Decision

Adopt Python 3.13 across all libs and apps.

Rationale

  1. Performance: Python 3.13 includes JIT improvements and faster async/await
  2. Async Native: First-class asyncio support for parallel adapter calls
  3. AI Ecosystem: Microsoft Agent Framework, Foundry SDK, Azure SDKs all Python-first
  4. Typing: Pydantic v2 + Python 3.13 type hints enable strong contracts
  5. Tooling: pytest, pylint, black, isort provide robust CI/CD gates

Consequences

Positive

  • Fast Development: Rapid prototyping for adapters and agents
  • Azure Integration: Native SDKs for all required services
  • Agent Framework: Direct Foundry integration without FFI overhead
  • Hiring: Large talent pool familiar with Python

Negative

  • Runtime Performance: Lower than compiled languages for CPU-bound tasks (mitigated by offloading ML to hosted models)
  • Packaging: Virtual env management adds setup complexity (mitigated by standardized pyproject.toml)
  • GIL: Global Interpreter Lock limits true parallelism (mitigated by async I/O focus and multi-process deployment)

Alternatives Considered

C# / .NET

  • Pros: Better runtime performance, strong typing, Azure SDK parity
  • Cons: Smaller AI/ML ecosystem; Agent Framework support lagging; slower iteration for prototypes

TypeScript / Node.js

  • Pros: Single language for UI + backend; strong async model
  • Cons: Weaker AI/ML ecosystem; no native Agent Framework support; less mature Azure AI SDKs

Go

  • Pros: Excellent concurrency, fast binaries
  • Cons: Minimal AI/ML libraries; no Agent Framework support; immature Azure AI SDKs

Implementation Notes

  • All pyproject.toml files specify requires-python = ">=3.13"
  • CI matrix tests on Python 3.13 only (no backcompat with 3.11/3.12)
  • Azure Functions / AKS base images locked to Python 3.13 slim
  • Pre-commit hooks enforce pylint, black, isort compliance

Related ADRs