[TheNetherWatcher]

The units in a signal line are matched with [\w\^\-\?%\/]*, which leaves out characters such as . and *. Units like a.u. are cut off at the first such character, and the rest of the line ends up in the signal name:

wfdb.wrsamp("h", fs=100, units=["a.u."], sig_name=["ECG II"], d_signal=np.array([[1], [2]]),
            fmt=["16"], adc_gain=[1.5], baseline=[0])
h = wfdb.rdheader("h")
h.units     # ['a']
h.sig_name  # ['.u. 16 0 1 3 0 ECG II']
h.adc_res   # [None]

WFDB only excludes whitespace from units, and wrsamp already allows any units without whitespace, so the regex now matches \S*.

I parsed every signal line in sample-data/ with the old and new regex. The only differences are in test_generator_2.hea, where two signals have units ° and the old regex put the whole rest of the line into the signal name. (rdheader currently drops non-ASCII characters before parsing, so this doesn't show up when reading that record.)

Tests: wrsamp then rdheader round trip for mV, a.u., uV*s, mmHg/s and 10^-3, checking units, signal name and ADC resolution. The a.u. and uV*s cases fail on main. pytest -m "not network" passes (101 tests), and the docs build passes.

[tompollard]

Thanks, this looks good! \S* matches what the WFDB C library does: readheader in lib/signal.c takes everything after the first / in the gain field, up to the next whitespace, as the units.

I also read all 88 headers in sample-data with rdheader on main and on this branch, and every field is identical.