v0.2.0

What's Changed

Added

• Python API (edgefirst-ara2 on PyPI) — complete PyO3 bindings with:
  • Session, Endpoint, and Model wrappers with full API parity
  • numpy tensor I/O (set_input_tensor, get_output_tensor, dequantize)
  • DMA-BUF file descriptor access (input_tensor_fd, output_tensor_fd) for
    zero-copy GPU preprocessing with edgefirst-hal.import_image()
  • DVM metadata API (read_metadata, read_labels, has_metadata)
  • Python exception hierarchy (Ara2Error → LibraryError, HardwareError,
    ProxyError, ModelError, TensorError, MetadataError)
  • Bounds checking on all tensor index accessors (raises IndexError)
  • Allocation guards (run() before allocate_tensors() raises TensorError)
  • Context manager support (with statement) on Session and Model
  • os.PathLike support on all path parameters
  • Comprehensive .pyi type stubs with docstrings
• Python YOLOv8 example (examples/yolov8.py) with DMA-BUF pipeline,
  HAL decoder integration, and --benchmark mode
• Rust YOLOv8 --benchmark mode with matching output format
• PyPI publishing via GitHub Actions with OIDC trusted publishing
• python.yml CI workflow for building manylinux2014 wheels (x86_64, aarch64)

Changed

• Upgraded edgefirst-hal from 0.11.0 to 0.13.0
• Migrated Rust YOLOv8 example to HAL 0.13 import_image / PlaneDescriptor API
• Upgraded pyo3 from 0.23 to 0.24, added numpy 0.24
• Release workflow now builds and publishes Python wheels alongside Rust crates
• Python version derived from Cargo.toml via dynamic = ["version"]
• Updated all GitHub Action hashes to latest versions (checkout v6.0.2,
  upload-artifact v7.0.0, rust-cache v2.9.1, install-action v2.69.10)
• Updated examples/endpoints.py to use edgefirst_ara2 module name

Fixed

• Release workflow tag patterns (glob-style v[0-9]* instead of regex v[0-9]+)
• Release workflow SBOM collection (recursive find for nested artifact paths)
• Rust YOLOv8 args parsing bounds check (prevents panic on missing flag value)

---

crates.io: ara2 | ara2-sys
PyPI: edgefirst-ara2