homepage

Three audio modules, aifc, sunau and wave, have similar interfaces and poor tests. The proposed patch introduces new file Lib/test/audiotests.py with common audio tests.
New testing exposes some bugs and discrepancy between different audio modules. For example aifc uses bytes for compression type and name, while sunau and wave use strings (issue8934). aifc closes underlied file object on close(), while sunau and wave don't. wave rounds a framerate, while aifc and sunau truncate it. Different modules have different behaviors when process framedata with length which is not divisible by frame size.

I love this idea! I was thinking while working on sunau/aifc/wave patches that we can do more than this, unify the entire audio modules,
getting rid of Aifc_write/read and Wave_write/read was in fact my first desire.
 One way that I thought about was to provide an abc and each module would implement that interface.
That way we could create new modules for audio operations for other formats as well. If this sounds like a good idea, I could try to provide a patch for this.

New changeset 61ca4732399b by Serhiy Storchaka in branch 'default':
Issues #18901, #18919: Fix a typo in the _sunau_params name.
http://hg.python.org/cpython/rev/61ca4732399b 

Patch updated. It now contains tests for all supported encoding formats (PCM8-32, u-Law, A-Law).

Patch updated. Now files created in external program (Audacity) are used for testing. This exposed yet one bug, in the aifc module (issue19131).

New changeset 3303d735058f by Serhiy Storchaka in branch '2.7':
Issue #18919: If the close() method of a writer in the sunau or wave module
http://hg.python.org/cpython/rev/3303d735058f
New changeset b7eae747385c by Serhiy Storchaka in branch '3.3':
Issue #18919: If the close() method of a writer in the sunau or wave module
http://hg.python.org/cpython/rev/b7eae747385c 

New changeset a9f967c8d002 by Serhiy Storchaka in branch '2.7':
Issue #18919: Added tests for the sunau module. Unified and extended tests
http://hg.python.org/cpython/rev/a9f967c8d002 

New changeset ee7563b07c6f by Serhiy Storchaka in branch '3.3':
Issue #18919: Unified and extended tests for audio modules: aifc, sunau and
http://hg.python.org/cpython/rev/ee7563b07c6f
New changeset fc418ecd8777 by Serhiy Storchaka in branch 'default':
Issue #18919: Unified and extended tests for audio modules: aifc, sunau and
http://hg.python.org/cpython/rev/fc418ecd8777 

Tests fail for 2.7 on the koobs-freebsd9 and koobs-freebsd10 buildbots since a9f967c8d002db364021a5fb1c9926a27029e849
Taking a snapshot of the output here for reference:
test_aifc
test test_aifc produced unexpected output:
**********************************************************************
Warning: bad COMM chunk size
Warning: bad COMM chunk size
Warning: MARK chunk contains only 0 markers instead of 1
**********************************************************************
<snip>
363 tests OK.
1 test failed:
 test_aifc
<snip>
Re-running failed tests in verbose mode
Re-running test 'test_aifc' in verbose mode
<snip>
test_read_comm_kludge_compname_even (test.test_aifc.AIFCLowLevelTest) ... Warning: bad COMM chunk size
ok
test_read_comm_kludge_compname_odd (test.test_aifc.AIFCLowLevelTest) ... Warning: bad COMM chunk size
ok
test_read_wrong_marks (test.test_aifc.AIFCLowLevelTest) ... Warning: MARK chunk contains only 0 markers instead of 1
ok
----------------------------------------------------------------------
Ran 80 tests in 0.102s
OK

Is unexpected output considered as test failing? The aifc module uses bare print to output these "warnings" to stdout. Here is a patch which uses captured_stdout().

In 2.7 unexpected output is a test failure, yes.
In 3.x we removed that check. But I wonder if that was wise :) (We did it for technical reasons when adding other features to regrtest).
Yes, using captured_stdout and checking for the expected warnings would be a better test, IMO.

Some new (I assume) ResourceWarnings from test_wave:
/home/gbr/devel/python/Lib/unittest/case.py:571: ResourceWarning: unclosed file <_io.BufferedWriter name='@test_1974_tmp'>
 testMethod()
/home/gbr/devel/python/Lib/chunk.py:61: ResourceWarning: unclosed file <_io.FileIO name='/home/gbr/devel/python/Lib/test/audiodata/pluck-pcm16.wav' mode='rb'>
 self.chunkname = file.read(4)
/home/gbr/devel/python/Lib/chunk.py:61: ResourceWarning: unclosed file <_io.FileIO name='/home/gbr/devel/python/Lib/test/audiodata/pluck-pcm24.wav' mode='rb'>
 self.chunkname = file.read(4)
/home/gbr/devel/python/Lib/chunk.py:61: ResourceWarning: unclosed file <_io.FileIO name='/home/gbr/devel/python/Lib/test/audiodata/pluck-pcm32.wav' mode='rb'>
 self.chunkname = file.read(4)
/home/gbr/devel/python/Lib/chunk.py:61: ResourceWarning: unclosed file <_io.FileIO name='/home/gbr/devel/python/Lib/test/audiodata/pluck-pcm8.wav' mode='rb'>
 self.chunkname = file.read(4)

New changeset 164a60cce934 by Serhiy Storchaka in branch '2.7':
Issue #18919: Catch and check warnings in the aifc module tests.
http://hg.python.org/cpython/rev/164a60cce934
New changeset 9eecd00ffc28 by Serhiy Storchaka in branch '3.3':
Issue #18919: Check warnings messages in the aifc module tests.
http://hg.python.org/cpython/rev/9eecd00ffc28
New changeset d168f094d16d by Serhiy Storchaka in branch 'default':
Issue #18919: Check warnings messages in the aifc module tests.
http://hg.python.org/cpython/rev/d168f094d16d 

New changeset fd7752db1e54 by Serhiy Storchaka in branch '2.7':
Issue #18919: Fixed resource leaks in audio tests.
http://hg.python.org/cpython/rev/fd7752db1e54
New changeset 2850fc02f324 by Serhiy Storchaka in branch '3.3':
Issue #18919: Fixed resource leaks in audio tests.
http://hg.python.org/cpython/rev/2850fc02f324
New changeset 835c6ea487b1 by Serhiy Storchaka in branch 'default':
Issue #18919: Fixed resource leaks in audio tests.
http://hg.python.org/cpython/rev/835c6ea487b1 

> In 3.x we removed that check. But I wonder if that was wise :) (We did it for technical reasons when adding other features to regrtest).
Actually the checks left, but there was no check for warning message. Now they are added.
> Some new (I assume) ResourceWarnings from test_wave:
Thank you Greg for this report. Yes, these warnings are new. The tests only existed for the aifc module (which always closes externally opened file), and when I extended it to other modules I forgot close externally opened files. Hope it now fixed.

The new tests fail when run from an installed Python. You also need to add all of the new test directories to LIBSUBDIRS for the libinstall target in Makefile.pre.in.

New changeset de1f62a55648 by Serhiy Storchaka in branch '2.7':
Issue #18919: Add test/audiodata to LIBSUBDIRS.
http://hg.python.org/cpython/rev/de1f62a55648
New changeset 865bab7387f2 by Serhiy Storchaka in branch '3.3':
Issue #18919: Add test/audiodata to LIBSUBDIRS.
http://hg.python.org/cpython/rev/865bab7387f2
New changeset 4d5143ff58ea by Serhiy Storchaka in branch 'default':
Issue #18919: Add test/audiodata to LIBSUBDIRS.
http://hg.python.org/cpython/rev/4d5143ff58ea 

It looks like test_wave is still failing on PPC64 PowerLinux:
http://buildbot.python.org/all/builders/PPC64%20PowerLinux%203.x/builds/850/steps/test/logs/stdio 

Thank you David. Fixed in issue19276.