Skip to content

RELEASE_GATES_CHECKLIST.md

Latest commit

 

History

History
125 lines (95 loc) · 4.91 KB

RELEASE_GATES_CHECKLIST.md

File metadata and controls

125 lines (95 loc) · 4.91 KB
Raw

Release Gates & Regression Tests — v2.0

Date: May 20, 2026
Branch: main
Engine: Airframe (wgpu/WebGPU — no CUDA, no llama.cpp)
Status: Active

Release Gate Pipeline

All 7 gates defined in .github/workflows/release.yml under the preflight job. Every gate must pass or the entire release stops.

Gate Name Covers
1/7 Core Build cargo build --features airframe,huggingface — GPU engine via submodule
2 CUDA Timeout Removed v2.0 — Airframe uses wgpu; no CUDA dependency
3/7 Template Packaging Docker templates included in crates.io package (Issue #60 regression)
4/7 Binary Size Constitutional 20 MB limit on shimmy binary
5/7 Test Suite cargo test --lib --no-default-features --features huggingface
5.1/7 Airframe Compile Check cargo check --features airframe — integration builds without errors
5.5/7 Issue Regression Per-issue regression tests (see list below)
6/7 Documentation cargo doc --no-deps --features huggingface
7/7 crates.io Dry-Run cargo publish --dry-run --features huggingface validates crates.io package

Regression Test Files

Active Test Files

File Status
tests/regression_tests.rs ✅ Active — main issue regression suite
tests/release_gate_integration.rs ✅ Active — validates gate system itself
tests/template_compilation_regression_test.rs ✅ Active — template file inclusion
tests/compilation_regression_test.rs ✅ Active — compilation sanity
tests/apple_silicon_detection_test.rs ✅ Active — GPU detection (Issue #87)
tests/integration_tests.rs ✅ Active — API surface integration
tests/cli_integration_tests.rs ✅ Active — CLI surface
tests/gpu_backend_tests.rs ✅ Active — GPU backend selection

Gate 5.5 Issue Regression Tests (in regression_tests.rs)

Test Issue
test_issue_111_gpu_metrics_endpoint #111 — GPU metrics endpoint
test_issue_112_safetensors_engine_selection #112 — SafeTensors engine selection
test_issue_113_openai_api_frontend_compatibility #113 — OAI /v1/models endpoint
test_issue_113_ollama_api_tags_response_structure #113 — Ollama /api/tags endpoint
test_issue_114_mlx_distribution_features #114 — MLX distribution feature flags
test_issue_191_multi_part_content_array_deserialization #191 — Multi-part content array (422 fix)
test_qwen_model_template_detection #13 — Qwen ChatML template
test_custom_model_directory_environment_variables #12 — Custom model directories

Test Execution Commands

# Gate 5 — Core lib tests
cargo test --lib --no-default-features --features huggingface -- --test-threads=1

# Gate 5.1 — Airframe compile check
cargo check --features airframe

# Gate 5.5 — Issue regression tests
cargo test --test regression_tests --no-default-features --features huggingface

# Gate validation system tests
cargo test --test release_gate_integration

# Full regression pass
cargo test --no-default-features --features huggingface

Pre-Release Checklist

Code Quality

  • cargo fmt -- --check
  • cargo clippy --no-default-features --features huggingface (no warnings)
  • cargo deny check (license check)
  • No compilation warnings on --features airframe,huggingface

Documentation

  • CHANGELOG.md updated with release version entry
  • README.md reflects current feature set
  • New features documented

Version Management

  • Cargo.toml version bumped
  • Git tag created (format: vX.Y.Z)
  • Release notes prepared

Workflow Validation

  • release.yml preflight job passes on a test branch
  • ci.yml test job passes on current branch

Feature Flag Reference

Feature Purpose crates.io safe
huggingface Model downloading, tokenizer — default ✅ Yes
airframe Airframe GPU engine via submodule path dep ❌ No (path dep)
full huggingface + airframe convenience alias ❌ No
apple Apple Silicon convenience alias platform-only

crates.io publish uses --no-default-features --features huggingface.
GitHub Release binaries use --features airframe,huggingface (built in CI matrix).

What's Gone (v2.0)

Removed Why
CUDA Gate (Gate 2) Airframe uses wgpu; zero CUDA dependency
shimmy-llama-cpp-2 dependency Replaced by Airframe GPU engine
--features llama build path No llama.cpp in this codebase
MLX-specific release artifacts Not part of v2.0 shipping target
Binary size inflation from llama.cpp Airframe is pure Rust; binary stays well under 20 MB