Testing algorithms¶

A new algorithm needs three tests:

C unit test — direct encode/decode round trip on a known buffer.
C integration test — compress and decompress a small mzML fixture through mscompress end to end.
Python/TS round-trip test — open an mzML, compress with the new --mz-lossy / --int-lossy setting, read back, compare.

C tests¶

ctest --test-dir cli
ctest --test-dir cli -R MyAlgo        # filter by name

C tests live in cli/test/. Each test is a small executable registered with add_test() in cli/CMakeLists.txt.

Python tests¶

cd python && uv run pytest
cd python && uv run pytest test/test_transforms.py::test_my_algo -v

Tolerance for lossy algorithms¶

A lossless algorithm round-trips bit-exact. Lossy algorithms need an explicit tolerance:

import numpy as np

# For delta32 on m/z arrays, error should be under ~4e-6 Da
np.testing.assert_allclose(roundtripped_mz, original_mz, atol=4e-6, rtol=0)

Document the expected tolerance on the algorithm's catalog page so users know what they're trading away.

Cross-language consistency¶

The same MSZ file produced by the CLI must read identically in Python and TypeScript. The CI matrix builds all three and runs a round-trip test across them; if you add an algorithm, make sure the CI fixture set covers it.