Audience: Developers and maintainers
This page documents internal testing practices and coverage notes; it is not required for normal users.

Testing & coverage

FoodSpec relies on extensive, fast, synthetic tests to keep the protocol trustworthy.

Running tests

pytest --disable-warnings -q

With coverage (if configured):

pytest --cov=foodspec

Current status (latest run)

• Approximate coverage: ~90% (pytest --cov, Python 3.12).
• Well covered: core data models, preprocessing (baseline/smoothing/normalization/derivatives), features (peaks/bands/ratios), metrics, viz, CLI workflows, protocol validation, model registry, robustness helpers.
• Partially covered but acceptable: chemometrics deep/MLP, vendor IO loaders (SPC/OPUS/JCAMP), some stats branches (hypothesis_tests/correlations edge cases), public data loaders for vendor formats.
• Gaps / future work:
• Add mocks/fixtures to raise IO/base/core/text/vendor coverage.
• Exercise more branches in stats (alternative tails, assumption checks) and preprocess (baseline/normalization edge cases).
• Deep-learning paths beyond smoke tests only if DL becomes first-class.

Why high coverage matters

• Protocol-oriented library: small regressions can invalidate published pipelines.
• Ensures CLI workflows (oil-auth, heating, mixture, protocol benchmarks) stay stable.
• Encourages reproducibility: deterministic, synthetic datasets avoid external dependencies.

Testing practices

• Use tmp_path for filesystem; monkeypatch for loaders and external dependencies.
• No network or large datasets; keep tests CPU-light.
• Add tests for new public APIs and edge cases (errors as well as happy paths).