Thank you Terry.

> Is whatever the case in 2.6 only (if so, drop obsolete) or since 2.6. If the latter, I would rewrite as "this is the case since Python 2.6.".
> The addition is not clear to me. Are you implying that something should be made true for memory view (in a future patch)?

In 2.6 the sentence possible was true and the code was correct. But it is wrong for currently supported versions when we pass memoryview as an argument. This is a bug (possible even a crash) and should be fixed. But on x86 or when we don't use unaligned memoryview (most peoples don't do this) it is never occurred. Due to this I left this bug unfixed yet. There are other minor bugs which I left for different issues. This patch fixes most critical bugs and creates a solid basis for testing.

I see unstable AIX buildbots failed on test_aifc. Perhaps this patch will fix it or expose an issue in audioop module (current audioop testing too poor).