A string in Python 3 is a sequence of unicode characters, right?  The
documentation of the bz2 module says:

8<------------------------------------------------------------------
class bz2.BZ2File(filename, mode='r', buffering=0, compresslevel=9)
[...]
write(data)
    Write string data to file. Note that due to buffering, close() may be
    needed before the file on disk reflects the data written.
[...]
8<------------------------------------------------------------------

So the documentation wants me to pass a "string data" to the write() method:

8<------------------------------------------------------------------
>>> import bz2
>>> with bz2.BZ2File('test.bz2', mode='w') as cfh:
...     cfh.write('Test')
...
Traceback (most recent call last):
  File "<stdin>", line 2, in <module>
TypeError: must be bytes or buffer, not str
8<------------------------------------------------------------------

So what write() really wants is a byte array or an encoded string:

8<------------------------------------------------------------------
>>> with bz2.BZ2File('test.bz2', mode='w') as cfh:
...     cfh.write(bytes('Test', encoding='iso-8859-1'))
... 
>>> with bz2.BZ2File('test.bz2', mode='w') as cfh:
...     cfh.write('Test'.encode('iso-8859-1'))
... 
8<------------------------------------------------------------------

Is this an inaccuracy of the documentation or did I get something wrong?

Fixed in r78884 (py3k) and r78885 (release31-maint), thanks for the report!