tests: fast default run, slow tests opt-in

[pgarrett-scripps]

Times (shared machine under load 30-45, 1 core, workspace venv)

run	before	after
default pytest tests	103.6 s (same machine, same session); 207 s in the earlier workspace baseline	41.9 s
--run-slow	n/a	79.3 s
RUN_SLOW=1 HYPOTHESIS_PROFILE=thorough (what CI runs)	n/a	87.4 s

The 20 s target is not met under this load. No single test is over 1.7 s now; what is left is ~770 tests at ~50 ms each plus ~5 s of collection/imports. On a quiet machine the default run should be roughly half that.

Shrunk (still run by default)

• iter_precursor_spectra tests (matches merged peaks, scattered order + tiny cache, decode-once): the first 60 precursors (20 frames, 39 multi-frame). The whole-file version (2519 precursors) is kept as a slow param.
• test_merge_peaks_shuffled_frame_matches_tof_scan_order[python]: the lowest-m/z 40,000 points of the largest MS1 frame (a contiguous m/z band, all scans) instead of all ~350,000. The numba variant still uses the full frame.
• TestWatershedCentroiderCall: the lowest-TOF eighth of the first MS1 frame instead of all ~340,000 points.
• test_viz default and centroid= tests, test_smooth_kwarg_threads_through_convenience_api: a PASEF MS2 frame instead of an MS1 frame. Added n > 0 checks so an empty frame cannot pass silently.
• test_v5_vocabulary._copy hard-links the 60 MB analysis.tdf_bin (falls back to a copy) since those tests only edit analysis.tdf.

Hypothesis

Profiles move to tests/conftest.py: default 30 examples, thorough 200 (at least 2x every old per-test count), deep 2000 (the old test_properties thorough). Per-test max_examples removed so the profile governs.

Marked slow (skipped unless --run-slow or RUN_SLOW=1)

• stdio MCP server launch, core-import subprocess check, validate CLI subprocess, docs collect-from-another-dir subprocess
• whole-file precursor sweeps (above), the docs example that iterates every precursor
• PandasTdf.properties (390k rows; the other 18 tables cover the same path)

CI

ci.yml, publish.yml, latest-deps.yml set RUN_SLOW=1 and HYPOTHESIS_PROFILE=thorough at workflow level, so the test matrix, lowest, coverage and mcp jobs run everything. just test-all added; test-cov, codecov-tests, test-mcp set RUN_SLOW=1. CLAUDE.md / CONTRIBUTING updated.

No tests deleted, no assertions weakened, no src/ changes.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!