Fix: 8-bit WAV bias overflowed numpy int8 waveforms (regression from #14)

recode/audio.py

@@ -36,6 +36,7 @@
 
 """
 
+import operator
 import struct
 import warnings
 import wave
@@ -164,13 +165,25 @@ def _shift_samples(frames, by: int):
     [-128, 0, 127]
     >>> _shift_samples([(0, 255), (128, 64)], -128)
     [(-128, 127), (0, -64)]
+
+    Samples are widened to Python `int` before the shift, so a numpy `int8` waveform --
+    the natural dtype for 8-bit audio -- does not overflow on its way to 0..255, and a
+    non-integer sample still fails loudly (as the struct codec always made it) instead
+    of being truncated:
+
+    >>> _shift_samples([-128, 0, 127], 128)
+    [0, 128, 255]
+    >>> _shift_samples([0.5], 128)
+    Traceback (most recent call last):
+      ...
+    TypeError: 'float' object cannot be interpreted as an integer
     """
+
+    def shift(sample):
+        return operator.index(sample) + by
+
     return [
-        (
-            tuple(sample + by for sample in frame)
-            if isinstance(frame, Iterable)
-            else frame + by
-        )
+        tuple(map(shift, frame)) if isinstance(frame, Iterable) else shift(frame)
         for frame in frames
     ]
 

test_recode.py

@@ -770,3 +770,22 @@ def test_a_non_pcm_extensible_subformat_still_raises():
     assert non_pcm != raw  # guard: the fixture really did carry the PCM GUID
     with pytest.raises(_wave.Error):
         decode_wav_bytes(non_pcm)
+
+
+def test_8_bit_encode_accepts_a_numpy_int8_waveform():
+    """int8 is the natural dtype of 8-bit audio; biasing it by 128 must not overflow.
+
+    Before the unsigned-8-bit fix, an int8 array went straight to the struct codec and
+    encoded fine; the bias then added a Python 128 to each np.int8 sample, which numpy 2
+    refuses (OverflowError) and numpy 1 silently wraps.
+    """
+    np = pytest.importorskip("numpy")
+    from recode.audio import encode_wav_bytes
+
+    mono = np.array([-128, 0, 127], dtype=np.int8)
+    raw = encode_wav_bytes(mono, _SR, width_bytes=1)
+    assert decode_wav_bytes(raw) == ([-128, 0, 127], _SR)
+
+    stereo = np.array([[-128, 1], [0, 127]], dtype=np.int8)
+    raw = encode_wav_bytes(stereo, _SR, width_bytes=1, n_channels=2)
+    assert decode_wav_bytes(raw) == ([(-128, 1), (0, 127)], _SR)