- ensure that option names in interpolations are handled by
  self.optionxform in the same way that other references to option
  names
- add tests, documentation
(closes SF bug #857881, patch #865455)

converting it into assertRaises() form.  Restored the 2.3 code, and
explained why assertRaises() cannot be used instead.

still 7-bit.

parser must recognize outer boundaries in inner parts.  So cruise through the
EOF stack backwards testing each predicate against the current line.

There's still some discussion about whether this is (always) the best thing to
do.  Anthony would rather parse these messages as if the outer boundaries were
ignored.  I think that's counter to the RFC, but might be practically more
useful.  Can you say behavior flag?  (ug).

Two evil samples from Anthony's MIME torture tests.

same boundary as the outer part.

separating boundary for an outer part inside an inner part.

optimization for short input; beef up tests for fix_sentence_endings
feature.

Minor wording fix.

same method that implements __setitem__ also implements __delitem__.
Also, there were several good use cases (removing items from a queue
and implementing Forth style stack ops).

(closes SF patch #870287)

(closes SF patch #952047)

(closes SF bug #952212)

message/delivery-status clause, and genericize it to handle all (other)
message/* content types.  This lets us correctly parse 2 more of Anthony's
MIME torture tests (specifically, the message/external-body examples).

copyright years.

Already backported to release23-maint.

sources

binascii_a2b_qp() and binascii_b2a_qp() with calls to PyMem_Malloc() and
PyMem_Free().  These won't return NULL unless the allocations actually fail,
so it won't trigger a bogus memory error on some platforms <cough>AIX</cough>
when passed a length of zero.

_EXPORT_INT calls in #ifdef's, to avoid breaking the build on MkLinux
(Linux 2.0).