[aaronkanzer]

pynumpress is optional, and neither pip install pyteomics nor packages that depend on pyteomics (matchms, for example) install it. Without it, pyteomics 5.0.1 reads a numpress-encoded array as if it were uncompressed:

from pyteomics import mzml
s = next(mzml.read("lin_mult8.mzML"))   # m/z array: MS-Numpress linear prediction compression, 72 bytes
s["m/z array"][:4]
# array([ 2.04753471e-312, nan, 9.76797317e-190, -6.03570075e+030])   no warning

When the encoded length is not a multiple of the dtype size, you get ValueError: buffer size must be a multiple of element size instead. All five real numpress files I tried (msconvert and OpenMS outputs) raise that error, which does not mention pynumpress.

Why: the numpress names are only added to _default_compression_map when pynumpress imports. Without it, _determine_compression finds no known compression term, _decompress gets None, and it returns the raw bytes unchanged.

Fix: when pynumpress is missing, still register the six numpress terms, mapped to a decoder that raises PyteomicsError("Decoding '<term>' requires pynumpress."). Nothing changes when pynumpress is installed. #198 uses the same pattern for the zstd codecs.

Tested:

• New test_numpress_without_pynumpress runs in a subprocess with sys.modules['pynumpress'] = None and checks that all six terms raise PyteomicsError. It fails on master (TypeError: 'NoneType' object is not callable) and passes with the fix.
• Full suite (all tests/test_*.py except test_usi and test_unimod, which need the network): 467 passed, 10 failed before. 468 passed, 10 failed after. The 10 failures are the same both times: the process-based map() tests fail with OSError: [Errno 22] in my sandbox.
• Without pynumpress, the non-numpress mzML files (pyteomics, pymzml, psims and matchms test data) still read in full.

Side note for the maintainers, not part of this PR: in draft #198, the "MS-Numpress linear prediction compression followed by zstd compression" entry is mapped to _zlibNumpress(...) instead of _zstdNumpress(...).