testing

name	Testing
description	Run and evaluate the comprehensive test suite (Rust unit + Python integration).

Testing Skill

This skill guides you through the process of verifying the correctness and performance stability of the Locus library.

1. Rust Unit & Property Tests

Use cargo-nextest for fast, parallel execution of the core library tests.

cargo nextest run --workspace --all-targets --all-features

2. Python Integration Tests

Verify the zero-copy Python bindings and high-level API.

uv run pytest

3. Regression & Performance Validation

Strict evaluation against standard datasets (ICRA 2020) to ensure no regressions in accuracy or latency.

To run the suite:

TRACY_NO_INVARIANT_CHECK=1 cargo test --release --test regression_icra2020 -- --test-threads=1

Updating Snapshots: If intentional changes have been made that alter the metrics, you must update the snapshots. Always use the release profile when updating snapshots to avoid extreme execution times:

# ICRA 2020 suite
TRACY_NO_INVARIANT_CHECK=1 INSTA_UPDATE=always cargo test --release --test regression_icra2020 -- --test-threads=1

# Render-Tag suite (if applicable)
TRACY_NO_INVARIANT_CHECK=1 LOCUS_HUB_DATASET_DIR=../../tests/data/hub_cache INSTA_UPDATE=always cargo test --release --test regression_render_tag -- --test-threads=1

# Distortion-aware suite (Brown-Conrady + Kannala-Brandt)
# One-time setup: sync the two configs into tests/data/hub_cache/ via
#   uv run python tools/bench/sync_hub.py --configs \
#     aprilgrid_distortion_brown_conrady_v1_1920x1080 aprilgrid_distortion_kannala_brandt_v1_1920x1080
# If sync_hub.py fails due to upstream dataset schema drift, fall back to
# direct parquet download via `huggingface_hub.hf_hub_download`.
TRACY_NO_INVARIANT_CHECK=1 LOCUS_HUB_DATASET_DIR=tests/data/hub_cache \
  cargo insta test --release --all-features --features bench-internals \
  --test regression_distortion_hub --review

Success Criteria:

• Pass/Fail: All tests must pass (exit code 0).
• Correctness: 0% failure rate in proptest suites.
• Accuracy: RMSE of corner detection must not increase by more than 5% compared to baseline.
• Latency: Must remain within 10% of targets defined in performance_benchmark skill.
• Safety: No new unwrap() or panic!() in code paths identified as "hot loops".