[Python 2.5a0 (#9, Dec  8 2004, 12:37:01); the
behaviours go way back.]

binascii.b2a_qp has several quirks discovered when
reimplementing.

(1) It lets low bytes pass through unescaped -- e.g.
binascii.b2a_qp('\x08') == '\x08' -- unless quotetabs
is True.  This looks to be an error in program logic at
lines ~1165 and ~1234, unless this is intended
(although quopri's internal version quotes them).  This
doesn't seem RFC 1521 compatible on my reading of
section 5 (admittedly I only read it for the first time
last week. :-)

(2) It determines whether to enforce \n or \r\n by
scanning the string for the first occurrence [which
should be mentioned in the docs].  binascii.c does this
by calling strchr, which stops at the first \x00.  

This means that:

>>> s0 = "The quick " + chr(0) + "brown fox.\r\n"
>>> s1 = "The quick " + chr(1) + "brown fox.\r\n"
>>> binascii.b2a_qp(s0)
'The quick \x00brown fox.\n'
>>> binascii.b2a_qp(s1)
'The quick \x01brown fox.\r\n'

and related strangenesses.

(3) The code escapes the period "." if and only if it's
the second character in a line.  I'm not sure why; the
only mention I can find in 1521 is in appendix B of
escaping a period ALONE on a line in some situations,
which binascii.b2a_qp doesn't do in any event.  This
behaviour may be prescribed by some other spec but is
strange enough to be mentioned in the docs if it's
intentional.

A new strictly RFC1521-compliant qp encoding would be a
good thing if backwards compatibility prevents changing
some of these.

Fixed (1) and (3) in rev. 54367. I think (2) is not important enough to do anything about it.