According to documentation (http://docs.python.org/library/audioop.html), adpcm2lin, alaw2lin and ulaw2lin are using 'width' argument to represent output frames width. However, in audioop.c module there are checks that are raising exceptions if input frames length is not multiple of 'width'. I have replaced checking of 'len' to match 'size' with checking of 'len*size' to match 'size' in order to retain only basic length validity checks.

Well, you should just replace these calls with audioop_check_size() instead. Apparently the checks were blindly added in issue7673.

By the way, I'm surprised audioop accepts unicode strings... :/

Yep, you're right. Didn't noticed audioop_check_size() function at first.

The fact that audioop accepts unicode strings seems weird to me too. I've replaced strings with bytes in tests. However, I'm afraid to add restrictions to library itself because of backward compatibility.

Do you think it would be possible to add a test for valid compressed data that used to trigger the check?

Yes, that tests really should be there too. Added them to patch.

New changeset 77188bc37c74 by Antoine Pitrou in branch '3.2':
Issue #13806: The size check in audioop decompression functions was too strict and could reject valid compressed data.
http://hg.python.org/cpython/rev/77188bc37c74

New changeset ce89dcba008f by Antoine Pitrou in branch 'default':
Issue #13806: The size check in audioop decompression functions was too strict and could reject valid compressed data.
http://hg.python.org/cpython/rev/ce89dcba008f

New changeset f1ee3bb6ba64 by Antoine Pitrou in branch '2.7':
Issue #13806: The size check in audioop decompression functions was too strict and could reject valid compressed data.
http://hg.python.org/cpython/rev/f1ee3bb6ba64

Thank you for your work Oleg! The patch is now committed.

This broke a sparc buildbot:
"""

======================================================================
FAIL: test_alaw2lin (test.test_audioop.TestAudioop)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home2/buildbot/slave/2.7.loewis-sun/build/Lib/test/test_audioop.py", line 114, in test_alaw2lin
    self.assertEqual(audioop.alaw2lin(d, 2), b'\x08\x00\x08\x01\x10\x02')
AssertionError: '\x00\x08\x01\x08\x02\x10' != '\x08\x00\x08\x01\x10\x02'

======================================================================
FAIL: test_ulaw2lin (test.test_audioop.TestAudioop)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home2/buildbot/slave/2.7.loewis-sun/build/Lib/test/test_audioop.py", line 127, in test_ulaw2lin
    self.assertEqual(audioop.ulaw2lin(d, 2), b'\x00\x00\x04\x01\x0c\x02')
AssertionError: '\x00\x00\x01\x04\x02\x0c' != '\x00\x00\x04\x01\x0c\x02'
"""

It's obviously an endianess issue (sparc is big endian).

New changeset bc6e768de2cc by Antoine Pitrou in branch '2.7':
Fix failing test on big-endian machines (issue #13806).
http://hg.python.org/cpython/rev/bc6e768de2cc

New changeset cfbd800af3a4 by Antoine Pitrou in branch '3.2':
Fix failing test on big-endian machines (issue #13806).
http://hg.python.org/cpython/rev/cfbd800af3a4

New changeset 05b40b006565 by Antoine Pitrou in branch 'default':
Fix failing test on big-endian machines (issue #13806).
http://hg.python.org/cpython/rev/05b40b006565

Will wait for the buildbot to clear.

The fix I committed seems ok. Unfortunately the buildbot doesn't compile 3.2 or 3.3, see #13843.