[RClaudiuM]

Fixes #485.

Writing a record where one channel is entirely NaN fails:

import numpy as np, wfdb
wfdb.wrsamp("xxx", fs=500, units=["mV"], sig_name=["I"],
            p_signal=np.array([[np.nan], [np.nan]]), fmt=["16"])
# RuntimeWarning: All-NaN slice encountered (x2)
# ValueError: cannot convert float NaN to integer

Both causes are the ones described in the issue:

• calc_adc_gain_baseline checked pmin == np.nan, which is always False, so the all-NaN branch never ran. It now uses np.isnan(pmin).
• calc_adc_params used np.nanmin/np.nanmax, which warn on all-NaN channels. It now uses np.fmin.reduce/np.fmax.reduce (as suggested in the issue), which give the same result without the warning, for both p_signal and e_p_signal.

An all-NaN channel now gets adc_gain=1, baseline=1 (the existing intent of that branch), is written as the NaN sample value, and reads back as NaN.

Testing:

• New TestSignal.test_all_nan_channel (runs with RuntimeWarning as an error): checks calc_adc_params for p_signal and e_p_signal, then a wrsamp/rdrecord round trip with one all-NaN channel and one normal channel. It fails on main and passes with this change.
• uv run pytest -m "not network": 335 passed.
• ruff format --check, ruff check and towncrier build --draft pass. Changelog entry added in changelog.d/.

[tompollard]

Thanks for this, and for the clear write-up. The pmin == np.nan check could never be true, so it's good to have that fixed. Switching to np.fmin.reduce / np.fmax.reduce as suggested in the issue removes the warnings neatly.

I checked out the branch and everything passes for me: the offline suite, ruff and your new test, which fails on main as expected. I also tried a few other inputs, such as an all-NaN float32 channel and an all-NaN channel passed as a plain list to e_p_signal, and they work too. Running the test with RuntimeWarning as an error is a nice touch.

Happy to merge!