[divya-git03]

Fixes #618

Problem

Record.wrsamp() raises a TypeError when a record contains physical signal data (p_signal) but has no digital signal data (d_signal). This occurs because checksum calculation expects digital samples.

Changes

Convert p_signal to digital samples using the existing adc() method when d_signal is None.

Apply the conversion only when expanded=False and physical signal data is available.

Preserve the existing behavior for records that already contain digital signals.

Add a regression test using sample-data/100 to verify that a physical-only record can be written and read back with the expected digital samples.

Add an entry to the unreleased changelog.

Testing

python -m pytest -q tests/test_record.py

90 passed, 1 skipped.

The skipped test was network-dependent and encountered a remote connection error.

git diff --check completed successfully.

Fix writing physical-only and expanded physical signals without retaining converted digital samples on the record.

[tompollard]

Thanks for picking this up! I checked that physical-only records from sample-data now write the same digital samples as when they're read with physical=False.

One problem: self.d_signal = self.adc() stays on the record after writing, so if p_signal is changed and the record is written again, the old data is written:

record = wfdb.rdrecord("sample-data/100", sampto=1000)
record.wrsamp()
record.p_signal = record.p_signal * 2
record.wrsamp()  # writes the original signal, not the doubled one

Could you convert only for the write and put d_signal back afterwards? The same works for expanded=True (with e_d_signal and self.adc(expanded=True)), which still fails at the moment. Something like:

converted = None
if not expanded and self.d_signal is None and self.p_signal is not None:
    self.d_signal = self.adc()
    converted = "d_signal"
elif expanded and self.e_d_signal is None and self.e_p_signal is not None:
    self.e_d_signal = self.adc(expanded=True)
    converted = "e_d_signal"
try:
    ...  # the existing checksum update and writing
finally:
    if converted:
        setattr(self, converted, None)

For the tests:

• Compare with the digital samples from the original file (wfdb.rdrecord("sample-data/100", sampto=1000, physical=False).d_signal) rather than record.adc(), which is the function being tested.
• Add a second write after changing p_signal, and check that record.d_signal is still None afterwards.
• Add an expanded=True case (for example sample-data/test01_00s_frame read with smooth_frames=False).

Could you also add "Fixes #618" to the description, and put the changelog entry on one line like the others? We're planning to merge #620, which moves changelog entries into separate files in changelog.d/, so once that's in, the entry will need to move to a file there (for example changelog.d/wrsamp-physical.rst).

[divya-git03]

Thanks for the detailed review!

I've addressed the requested changes:

• Made the digital conversion temporary and restored d_signal / e_d_signal after writing.
• Added support for expanded=True using adc(expanded=True).
• Updated the regression test to compare against the original file's digital samples.
• Added a second-write test after modifying p_signal.
• Added an expanded physical-only record test.
• Moved the changelog entry to changelog.d/wrsamp-physical.rst in the new one-line format.

I've pushed the updates in commit 4b17cf7. The targeted regression tests pass (3 passed), and git diff --check is clean.

I've also updated the PR description with Fixes #618.

Thanks again for the guidance!