Populate the benchmark, stress and property-based test tiers

[tschm]

Closes #86

Problem

The Makefile wires up three test tiers this repo never filled:

• make test passed --ignore=tests/benchmarks --ignore=tests/stress for directories that did not exist.
• make benchmark and make stress had nothing to run — and pytest-benchmark emitted a warning on every make test run about benchmarks being disabled, for benchmarks that did not exist.
• hypothesis shipped via .rhiza/requirements/tests.txt without a single import.

Change

tests/benchmarks/ measures the two hot paths: DataManager.get_chunk, which runs on every pan and zoom, and validate_input, which dominates plot() start-up. Measured here:

Benchmark	Mean
get_chunk — 5k-bar viewport slice	218 µs
validate_input — 250k bars	373 µs
DataManager construction — 250k bars	402 µs
get_chunk — full 250k-bar serialize	14.5 ms

tests/stress/ pushes the shared _data_managers registry that the unit suite only ever touches one session at a time: 100 concurrent registrations, 200 overlapping reads of one manager, and create/delete churn.

Property-based tests state validate_input's contract as invariants rather than examples — every series normalizes to a length-n ndarray, length disagreement always raises, auto-filled high never falls below low, and list and ndarray inputs agree. They carry the property marker pytest.ini already registers, so make hypothesis-test selects them.

hypothesis is added to the test dependency group so uv run pytest works without make install having run first — matching that group's stated intent in the existing comment.

A note on the stress tests

The first draft of test_many_concurrent_sessions_stay_isolated offset only close, leaving high below it. validate_input correctly rejected that with High must be >= max(Open, Close) — a good sign for the validation path. The test now shifts the whole bar.

Verification

• make test — 176 passed (was 172), coverage 100%
• make stress — 4 passed, 180 deselected
• make hypothesis-test — 4 passed, 50 examples each
• make benchmark — 4 benchmarks, numbers above
• check_test_layout.py — parity still holds; both new directories are exempt by design

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This pull request fills out the previously-empty test tiers by adding benchmark and stress suites, plus property-based tests for validate_input, and wires hypothesis into the standard test dependency group so uv run pytest works without a separate install step.

Changes:

• Add tests/benchmarks/ with pytest-benchmark benchmarks for validate_input and DataManager.get_chunk.
• Add tests/stress/ to exercise the global _data_managers registry under concurrent access patterns.
• Add Hypothesis-based property tests for validate_input and include hypothesis in the test dependency group.

Reviewed changes

Copilot reviewed 6 out of 7 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
uv.lock	Locks hypothesis and its dependency (sortedcontainers) and adds it to the test group.
pyproject.toml	Declares hypothesis under [project.optional-dependencies].test to support uv run pytest.
tests/benchmarks/init.py	Introduces the benchmarks test tier and documents exclusion from the default unit run.
tests/benchmarks/test_ingestion_benchmarks.py	Adds ingestion and slicing benchmarks for the core hot paths.
tests/stress/init.py	Introduces the stress test tier and documents exclusion from the default unit run.
tests/stress/test_session_registry_stress.py	Adds concurrency and churn stress tests around the session registry and slicing path.
tests/pycharting/data/test_ingestion.py	Adds Hypothesis property-based tests expressing validate_input invariants.

Suppressed comments (2)

tests/benchmarks/test_ingestion_benchmarks.py:44

• This benchmark test uses the ohlc fixture whose values are not all np.ndarray (the index is a pd.DatetimeIndex). The ohlc parameter type should be loosened to avoid incorrect typing.

def test_validate_input_on_dense_ohlc(benchmark, ohlc: dict[str, np.ndarray]) -> None:

tests/benchmarks/test_ingestion_benchmarks.py:63

• This benchmark test uses the ohlc fixture whose values are not all np.ndarray (the index is a pd.DatetimeIndex). The ohlc parameter type should be loosened to avoid incorrect typing.

def test_data_manager_construction(benchmark, ohlc: dict[str, np.ndarray]) -> None:

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.