Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

0.3.0 - 2026-02-06

vLLM public API surface generation via SnakeBridge (module_mode: :docs, driven by committed priv/snakebridge/vllm.docs.json)
Class method guardrail (max_class_methods) to prevent extremely large wrappers from inheritance-heavy internal classes
Coverage reports in .snakebridge/coverage/ for tracking API binding completeness
Credo configuration (.credo.exs) to exclude generated wrappers from static analysis
Python test script (basic_test.py) for direct vLLM validation
vLLM v1 multiprocessing configuration option (:auto, :on, :off)
Auto-set VLLM_WORKER_MULTIPROC_METHOD=spawn when v1 multiprocessing is enabled (avoids forking multi-threaded gRPC server)
Documentation for mix snakebridge.regen command with --clean option

Updated to SnakeBridge 0.15.0
Enhanced configuration guide with vLLM v1 multiprocessing documentation
Updated direct_api.exs to be wrapper-only while still demonstrating runtime attribute access for Python refs
Enhanced run_all.sh to compile once and run examples with --no-compile (avoids repeated codegen), plus better process cleanup and interrupt handling

Dialyzer no longer hangs due to an excessively-large generated wrapper surface (docs-manifest generation + class method guardrails keep modules small).
Examples now create wrapper refs in the same runtime session as the LLM and pass __runtime__ opts, preventing cross-session reference errors.
Example docs and scripts now derive runtime_opts from the LLM ref consistently for generation, chat, and embeddings calls.