#12586: add provisional email policy with new header parsing and folding.

When the new policies are used (and only when the new policies are explicitly
used) headers turn into objects that have attributes based on their parsed
values, and can be set using objects that encapsulate the values, as well as
set directly from unicode strings.  The folding algorithm then takes care of
encoding unicode where needed, and folding according to the highest level
syntactic objects.

With this patch only date and time headers are parsed as anything other than
unstructured, but that is all the helper methods in the existing API handle.
I do plan to add more parsers, and complete the set specified in the RFC
before the package becomes stable.

8 files changed, 3047 insertions, 11 deletions

diff --git a/Lib/email/_encoded_words.py b/Lib/email/_encoded_words.py
new file mode 100644
index 0000000..01fe42f
--- /dev/null
+++ b/Lib/email/_encoded_words.py
@@ -0,0 +1,211 @@
+""" Routines for manipulating RFC2047 encoded words.
+
+This is currently a package-private API, but will be considered for promotion
+to a public API if there is demand.
+
+"""
+
+# An ecoded word looks like this:
+#
+#        =?charset[*lang]?cte?encoded_string?=
+#
+# for more information about charset see the charset module.  Here it is one
+# of the preferred MIME charset names (hopefully; you never know when parsing).
+# cte (Content Transfer Encoding) is either 'q' or 'b' (ignoring case).  In
+# theory other letters could be used for other encodings, but in practice this
+# (almost?) never happens.  There could be a public API for adding entries
+# to to the CTE tables, but YAGNI for now.  'q' is Quoted Printable, 'b' is
+# Base64.  The meaning of encoded_string should be obvious.  'lang' is optional
+# as indicated by the brackets (they are not part of the syntax) but is almost
+# never encountered in practice.
+#
+# The general interface for a CTE decoder is that it takes the encoded_string
+# as its argument, and returns a tuple (cte_decoded_string, defects).  The
+# cte_decoded_string is the original binary that was encoded using the
+# specified cte.  'defects' is a list of MessageDefect instances indicating any
+# problems encountered during conversion.  'charset' and 'lang' are the
+# corresponding strings extracted from the EW, case preserved.
+#
+# The general interface for a CTE encoder is that it takes a binary sequence
+# as input and returns the cte_encoded_string, which is an ascii-only string.
+#
+# Each decoder must also supply a length function that takes the binary
+# sequence as its argument and returns the length of the resulting encoded
+# string.
+#
+# The main API functions for the module are decode, which calls the decoder
+# referenced by the cte specifier, and encode, which adds the appropriate
+# RFC 2047 "chrome" to the encoded string, and can optionally automatically
+# select the shortest possible encoding.  See their docstrings below for
+# details.
+
+import re
+import base64
+import binascii
+import functools
+from string import ascii_letters, digits
+from email import errors
+
+#
+# Quoted Printable
+#
+
+# regex based decoder.
+_q_byte_subber = functools.partial(re.compile(br'=([a-fA-F0-9]{2})').sub,
+        lambda m: bytes([int(m.group(1), 16)]))
+
+def decode_q(encoded):
+    encoded = encoded.replace(b'_', b' ')
+    return _q_byte_subber(encoded), []
+
+
+# dict mapping bytes to their encoded form
+class QByteMap(dict):
+
+    safe = b'-!*+/' + ascii_letters.encode('ascii') + digits.encode('ascii')
+
+    def __missing__(self, key):
+        if key in self.safe:
+            self[key] = chr(key)
+        else:
+            self[key] = "={:02X}".format(key)
+        return self[key]
+
+_q_byte_map = QByteMap()
+
+# In headers spaces are mapped to '_'.
+_q_byte_map[ord(' ')] = '_'
+
+def encode_q(bstring):
+    return ''.join(_q_byte_map[x] for x in bstring)
+
+def len_q(bstring):
+    return sum(len(_q_byte_map[x]) for x in bstring)
+
+
+#
+# Base64
+#
+
+def decode_b(encoded):
+    defects = []
+    pad_err = len(encoded) % 4
+    if pad_err:
+        defects.append(errors.InvalidBase64PaddingDefect())
+        padded_encoded = encoded + b'==='[:4-pad_err]
+    else:
+        padded_encoded = encoded
+    try:
+        return base64.b64decode(padded_encoded, validate=True), defects
+    except binascii.Error:
+        # Since we had correct padding, this must an invalid char error.
+        defects = [errors.InvalidBase64CharactersDefect()]
+        # The non-alphabet characters are ignored as far as padding
+        # goes, but we don't know how many there are.  So we'll just
+        # try various padding lengths until something works.
+        for i in 0, 1, 2, 3:
+            try:
+                return base64.b64decode(encoded+b'='*i, validate=False), defects
+            except binascii.Error:
+                if i==0:
+                    defects.append(errors.InvalidBase64PaddingDefect())
+        else:
+            # This should never happen.
+            raise AssertionError("unexpected binascii.Error")
+
+def encode_b(bstring):
+    return base64.b64encode(bstring).decode('ascii')
+
+def len_b(bstring):
+    groups_of_3, leftover = divmod(len(bstring), 3)
+    # 4 bytes out for each 3 bytes (or nonzero fraction thereof) in.
+    return groups_of_3 * 4 + (4 if leftover else 0)
+
+
+_cte_decoders = {
+    'q': decode_q,
+    'b': decode_b,
+    }
+
+def decode(ew):
+    """Decode encoded word and return (string, charset, lang, defects) tuple.
+
+    An RFC 2047/2243 encoded word has the form:
+
+        =?charset*lang?cte?encoded_string?=
+
+    where '*lang' may be omitted but the other parts may not be.
+
+    This function expects exactly such a string (that is, it does not check the
+    syntax and may raise errors if the string is not well formed), and returns
+    the encoded_string decoded first from its Content Transfer Encoding and
+    then from the resulting bytes into unicode using the specified charset.  If
+    the cte-decoded string does not successfully decode using the specified
+    character set, a defect is added to the defects list and the unknown octets
+    are replaced by the unicode 'unknown' character \uFDFF.
+
+    The specified charset and language are returned.  The default for language,
+    which is rarely if ever encountered, is the empty string.
+
+    """
+    _, charset, cte, cte_string, _ = ew.split('?')
+    charset, _, lang = charset.partition('*')
+    cte = cte.lower()
+    # Recover the original bytes and do CTE decoding.
+    bstring = cte_string.encode('ascii', 'surrogateescape')
+    bstring, defects = _cte_decoders[cte](bstring)
+    # Turn the CTE decoded bytes into unicode.
+    try:
+        string = bstring.decode(charset)
+    except UnicodeError:
+        defects.append(errors.UndecodableBytesDefect("Encoded word "
+            "contains bytes not decodable using {} charset".format(charset)))
+        string = bstring.decode(charset, 'surrogateescape')
+    except LookupError:
+        string = bstring.decode('ascii', 'surrogateescape')
+        if charset.lower() != 'unknown-8bit':
+            defects.append(errors.CharsetError("Unknown charset {} "
+                "in encoded word; decoded as unknown bytes".format(charset)))
+    return string, charset, lang, defects
+
+
+_cte_encoders = {
+    'q': encode_q,
+    'b': encode_b,
+    }
+
+_cte_encode_length = {
+    'q': len_q,
+    'b': len_b,
+    }
+
+def encode(string, charset='utf-8', encoding=None, lang=''):
+    """Encode string using the CTE encoding that produces the shorter result.
+
+    Produces an RFC 2047/2243 encoded word of the form:
+
+        =?charset*lang?cte?encoded_string?=
+
+    where '*lang' is omitted unless the 'lang' parameter is given a value.
+    Optional argument charset (defaults to utf-8) specifies the charset to use
+    to encode the string to binary before CTE encoding it.  Optional argument
+    'encoding' is the cte specifier for the encoding that should be used ('q'
+    or 'b'); if it is None (the default) the encoding which produces the
+    shortest encoded sequence is used, except that 'q' is preferred if it is up
+    to five characters longer.  Optional argument 'lang' (default '') gives the
+    RFC 2243 language string to specify in the encoded word.
+
+    """
+    if charset == 'unknown-8bit':
+        bstring = string.encode('ascii', 'surrogateescape')
+    else:
+        bstring = string.encode(charset)
+    if encoding is None:
+        qlen = _cte_encode_length['q'](bstring)
+        blen = _cte_encode_length['b'](bstring)
+        # Bias toward q.  5 is arbitrary.
+        encoding = 'q' if qlen - blen < 5 else 'b'
+    encoded = _cte_encoders[encoding](bstring)
+    if lang:
+        lang = '*' + lang
+    return "=?{}{}?{}?{}?=".format(charset, lang, encoding, encoded)
diff --git a/Lib/email/_header_value_parser.py b/Lib/email/_header_value_parser.py
new file mode 100644
index 0000000..87d8f68
--- /dev/null
+++ b/Lib/email/_header_value_parser.py
@@ -0,0 +1,2145 @@
+"""Header value parser implementing various email-related RFC parsing rules.
+
+The parsing methods defined in this module implement various email related
+parsing rules.  Principal among them is RFC 5322, which is the followon
+to RFC 2822 and primarily a clarification of the former.  It also implements
+RFC 2047 encoded word decoding.
+
+RFC 5322 goes to considerable trouble to maintain backward compatibility with
+RFC 822 in the parse phase, while cleaning up the structure on the generation
+phase.  This parser supports correct RFC 5322 generation by tagging white space
+as folding white space only when folding is allowed in the non-obsolete rule
+sets.  Actually, the parser is even more generous when accepting input than RFC
+5322 mandates, following the spirit of Postel's Law, which RFC 5322 encourages.
+Where possible deviations from the standard are annotated on the 'defects'
+attribute of tokens that deviate.
+
+The general structure of the parser follows RFC 5322, and uses its terminology
+where there is a direct correspondence.  Where the implementation requires a
+somewhat different structure than that used by the formal grammar, new terms
+that mimic the closest existing terms are used.  Thus, it really helps to have
+a copy of RFC 5322 handy when studying this code.
+
+Input to the parser is a string that has already been unfolded according to
+RFC 5322 rules.  According to the RFC this unfolding is the very first step, and
+this parser leaves the unfolding step to a higher level message parser, which
+will have already detected the line breaks that need unfolding while
+determining the beginning and end of each header.
+
+The output of the parser is a TokenList object, which is a list subclass.  A
+TokenList is a recursive data structure.  The terminal nodes of the structure
+are Terminal objects, which are subclasses of str.  These do not correspond
+directly to terminal objects in the formal grammar, but are instead more
+practical higher level combinations of true terminals.
+
+All TokenList and Terminal objects have a 'value' attribute, which produces the
+semantically meaningful value of that part of the parse subtree.  The value of
+all whitespace tokens (no matter how many sub-tokens they may contain) is a
+single space, as per the RFC rules.  This includes 'CFWS', which is herein
+included in the general class of whitespace tokens.  There is one exception to
+the rule that whitespace tokens are collapsed into single spaces in values: in
+the value of a 'bare-quoted-string' (a quoted-string with no leading or
+trailing whitespace), any whitespace that appeared between the quotation marks
+is preserved in the returned value.  Note that in all Terminal strings quoted
+pairs are turned into their unquoted values.
+
+All TokenList and Terminal objects also have a string value, which attempts to
+be a "canonical" representation of the RFC-compliant form of the substring that
+produced the parsed subtree, including minimal use of quoted pair quoting.
+Whitespace runs are not collapsed.
+
+Comment tokens also have a 'content' attribute providing the string found
+between the parens (including any nested comments) with whitespace preserved.
+
+All TokenList and Terminal objects have a 'defects' attribute which is a
+possibly empty list all of the defects found while creating the token.  Defects
+may appear on any token in the tree, and a composite list of all defects in the
+subtree is available through the 'all_defects' attribute of any node.  (For
+Terminal notes x.defects == x.all_defects.)
+
+Each object in a parse tree is called a 'token', and each has a 'token_type'
+attribute that gives the name from the RFC 5322 grammar that it represents.
+Not all RFC 5322 nodes are produced, and there is one non-RFC 5322 node that
+may be produced: 'ptext'.  A 'ptext' is a string of printable ascii characters.
+It is returned in place of lists of (ctext/quoted-pair) and
+(qtext/quoted-pair).
+
+XXX: provide complete list of token types.
+"""
+
+import re
+from email import _encoded_words as _ew
+from email import errors
+from email import utils
+
+#
+# Useful constants and functions
+#
+
+WSP = set(' \t')
+CFWS_LEADER = WSP | set('(')
+SPECIALS = set(r'()<>@,:;.\"[]')
+ATOM_ENDS = SPECIALS | WSP
+DOT_ATOM_ENDS = ATOM_ENDS - set('.')
+# '.', '"', and '(' do not end phrases in order to support obs-phrase
+PHRASE_ENDS = SPECIALS - set('."(')
+
+def quote_string(value):
+    return '"'+str(value).replace('\\', '\\\\').replace('"', r'\"')+'"'
+
+#
+# Accumulator for header folding
+#
+
+class _Folded:
+
+    def __init__(self, maxlen, policy):
+        self.maxlen = maxlen
+        self.policy = policy
+        self.lastlen = 0
+        self.stickyspace = None
+        self.firstline = True
+        self.done = []
+        self.current = []
+
+    def newline(self):
+        self.done.extend(self.current)
+        self.done.append(self.policy.linesep)
+        self.current.clear()
+        self.lastlen = 0
+
+    def finalize(self):
+        if self.current:
+            self.newline()
+
+    def __str__(self):
+        return ''.join(self.done)
+
+    def append(self, stoken):
+        self.current.append(stoken)
+
+    def append_if_fits(self, token, stoken=None):
+        if stoken is None:
+            stoken = str(token)
+        l = len(stoken)
+        if self.stickyspace is not None:
+            stickyspace_len = len(self.stickyspace)
+            if self.lastlen + stickyspace_len + l <= self.maxlen:
+                self.current.append(self.stickyspace)
+                self.lastlen += stickyspace_len
+                self.current.append(stoken)
+                self.lastlen += l
+                self.stickyspace = None
+                self.firstline = False
+                return True
+            if token.has_fws:
+                ws = token.pop_leading_fws()
+                if ws is not None:
+                    self.stickyspace += str(ws)
+                    stickyspace_len += len(ws)
+                token._fold(self)
+                return True
+            if stickyspace_len and l + 1 <= self.maxlen:
+                margin = self.maxlen - l
+                if 0 < margin < stickyspace_len:
+                    trim = stickyspace_len - margin
+                    self.current.append(self.stickyspace[:trim])
+                    self.stickyspace = self.stickyspace[trim:]
+                    stickyspace_len = trim
+                self.newline()
+                self.current.append(self.stickyspace)
+                self.current.append(stoken)
+                self.lastlen = l + stickyspace_len
+                self.stickyspace = None
+                self.firstline = False
+                return True
+            if not self.firstline:
+                self.newline()
+            self.current.append(self.stickyspace)
+            self.current.append(stoken)
+            self.stickyspace = None
+            self.firstline = False
+            return True
+        if self.lastlen + l <= self.maxlen:
+            self.current.append(stoken)
+            self.lastlen += l
+            return True
+        if l < self.maxlen:
+            self.newline()
+            self.current.append(stoken)
+            self.lastlen = l
+            return True
+        return False
+
+#
+# TokenList and its subclasses
+#
+
+class TokenList(list):
+
+    token_type = None
+
+    def __init__(self, *args, **kw):
+        super().__init__(*args, **kw)
+        self.defects = []
+
+    def __str__(self):
+        return ''.join(str(x) for x in self)
+
+    def __repr__(self):
+        return '{}({})'.format(self.__class__.__name__,
+                             super().__repr__())
+
+    @property
+    def value(self):
+        return ''.join(x.value for x in self if x.value)
+
+    @property
+    def all_defects(self):
+        return sum((x.all_defects for x in self), self.defects)
+
+    #
+    # Folding API
+    #
+    # parts():
+    #
+    # return a list of objects that constitute the "higher level syntactic
+    # objects" specified by the RFC as the best places to fold a header line.
+    # The returned objects must include leading folding white space, even if
+    # this means mutating the underlying parse tree of the object.  Each object
+    # is only responsible for returning *its* parts, and should not drill down
+    # to any lower level except as required to meet the leading folding white
+    # space constraint.
+    #
+    # _fold(folded):
+    #
+    #   folded: the result accumulator.  This is an instance of _Folded.
+    #       (XXX: I haven't finished factoring this out yet, the folding code
+    #       pretty much uses this as a state object.) When the folded.current
+    #       contains as much text as will fit, the _fold method should call
+    #       folded.newline.
+    #  folded.lastlen: the current length of the test stored in folded.current.
+    #  folded.maxlen: The maximum number of characters that may appear on a
+    #       folded line.  Differs from the policy setting in that "no limit" is
+    #       represented by +inf, which means it can be used in the trivially
+    #       logical fashion in comparisons.
+    #
+    # Currently no subclasses implement parts, and I think this will remain
+    # true.  A subclass only needs to implement _fold when the generic version
+    # isn't sufficient.  _fold will need to be implemented primarily when it is
+    # possible for encoded words to appear in the specialized token-list, since
+    # there is no generic algorithm that can know where exactly the encoded
+    # words are allowed.  A _fold implementation is responsible for filling
+    # lines in the same general way that the top level _fold does. It may, and
+    # should, call the _fold method of sub-objects in a similar fashion to that
+    # of the top level _fold.
+    #
+    # XXX: I'm hoping it will be possible to factor the existing code further
+    # to reduce redundancy and make the logic clearer.
+
+    @property
+    def parts(self):
+        klass = self.__class__
+        this = []
+        for token in self:
+            if token.startswith_fws():
+                if this:
+                    yield this[0] if len(this)==1 else klass(this)
+                    this.clear()
+            end_ws = token.pop_trailing_ws()
+            this.append(token)
+            if end_ws:
+                yield klass(this)
+                this = [end_ws]
+        if this:
+            yield this[0] if len(this)==1 else klass(this)
+
+    def startswith_fws(self):
+        return self[0].startswith_fws()
+
+    def pop_leading_fws(self):
+        if self[0].token_type == 'fws':
+            return self.pop(0)
+        return self[0].pop_leading_fws()
+
+    def pop_trailing_ws(self):
+        if self[-1].token_type == 'cfws':
+            return self.pop(-1)
+        return self[-1].pop_trailing_ws()
+
+    @property
+    def has_fws(self):
+        for part in self:
+            if part.has_fws:
+                return True
+        return False
+
+    def has_leading_comment(self):
+        return self[0].has_leading_comment()
+
+    @property
+    def comments(self):
+        comments = []
+        for token in self:
+            comments.extend(token.comments)
+        return comments
+
+    def fold(self, *, policy):
+        # max_line_length 0/None means no limit, ie: infinitely long.
+        maxlen = policy.max_line_length or float("+inf")
+        folded = _Folded(maxlen, policy)
+        self._fold(folded)
+        folded.finalize()
+        return str(folded)
+
+    def as_encoded_word(self, charset):
+        # This works only for things returned by 'parts', which include
+        # the leading fws, if any, that should be used.
+        res = []
+        ws = self.pop_leading_fws()
+        if ws:
+            res.append(ws)
+        trailer = self.pop(-1) if self[-1].token_type=='fws' else ''
+        res.append(_ew.encode(str(self), charset))
+        res.append(trailer)
+        return ''.join(res)
+
+    def cte_encode(self, charset, policy):
+        res = []
+        for part in self:
+            res.append(part.cte_encode(charset, policy))
+        return ''.join(res)
+
+    def _fold(self, folded):
+        for part in self.parts:
+            tstr = str(part)
+            tlen = len(tstr)
+            try:
+                str(part).encode('us-ascii')
+            except UnicodeEncodeError:
+                if any(isinstance(x, errors.UndecodableBytesDefect)
+                        for x in part.all_defects):
+                    charset = 'unknown-8bit'
+                else:
+                    # XXX: this should be a policy setting
+                    charset = 'utf-8'
+                tstr = part.cte_encode(charset, folded.policy)
+                tlen = len(tstr)
+            if folded.append_if_fits(part, tstr):
+                continue
+            # Peel off the leading whitespace if any and make it sticky, to
+            # avoid infinite recursion.
+            ws = part.pop_leading_fws()
+            if ws is not None:
+                # Peel off the leading whitespace and make it sticky, to
+                # avoid infinite recursion.
+                folded.stickyspace = str(part.pop(0))
+                if folded.append_if_fits(part):
+                    continue
+            if part.has_fws:
+                part._fold(folded)
+                continue
+            # There are no fold points in this one; it is too long for a single
+            # line and can't be split...we just have to put it on its own line.
+            folded.append(tstr)
+            folded.newline()
+
+    def pprint(self, indent=''):
+        print('\n'.join(self._pp(indent='')))
+
+    def ppstr(self, indent=''):
+        return '\n'.join(self._pp(indent=''))
+
+    def _pp(self, indent=''):
+        yield '{}{}/{}('.format(
+            indent,
+            self.__class__.__name__,
+            self.token_type)
+        for token in self:
+            for line in token._pp(indent+'    '):
+                yield line
+        if self.defects:
+            extra = ' Defects: {}'.format(self.defects)
+        else:
+            extra = ''
+        yield '{}){}'.format(indent, extra)
+
+
+class WhiteSpaceTokenList(TokenList):
+
+    @property
+    def value(self):
+        return ' '
+
+    @property
+    def comments(self):
+        return [x.content for x in self if x.token_type=='comment']
+
+
+class UnstructuredTokenList(TokenList):
+
+    token_type = 'unstructured'
+
+    def _fold(self, folded):
+        if any(x.token_type=='encoded-word' for x in self):
+            return self._fold_encoded(folded)
+        # Here we can have either a pure ASCII string that may or may not
+        # have surrogateescape encoded bytes, or a unicode string.
+        last_ew = None
+        for part in self.parts:
+            tstr = str(part)
+            is_ew = False
+            try:
+                str(part).encode('us-ascii')
+            except UnicodeEncodeError:
+                if any(isinstance(x, errors.UndecodableBytesDefect)
+                       for x in part.all_defects):
+                    charset = 'unknown-8bit'
+                else:
+                    charset = 'utf-8'
+                if last_ew is not None:
+                    # We've already done an EW, combine this one with it
+                    # if there's room.
+                    chunk = get_unstructured(
+                        ''.join(folded.current[last_ew:]+[tstr])).as_encoded_word(charset)
+                    oldlastlen = sum(len(x) for x in folded.current[:last_ew])
+                    schunk = str(chunk)
+                    lchunk = len(schunk)
+                    if oldlastlen + lchunk <= folded.maxlen:
+                        del folded.current[last_ew:]
+                        folded.append(schunk)
+                        folded.lastlen = oldlastlen + lchunk
+                        continue
+                tstr = part.as_encoded_word(charset)
+                is_ew = True
+            if folded.append_if_fits(part, tstr):
+                if is_ew:
+                    last_ew = len(folded.current) - 1
+                continue
+            if is_ew or last_ew:
+                # It's too big to fit on the line, but since we've
+                # got encoded words we can use encoded word folding.
+                part._fold_as_ew(folded)
+                continue
+            # Peel off the leading whitespace if any and make it sticky, to
+            # avoid infinite recursion.
+            ws = part.pop_leading_fws()
+            if ws is not None:
+                folded.stickyspace = str(ws)
+                if folded.append_if_fits(part):
+                    continue
+            if part.has_fws:
+                part.fold(folded)
+                continue
+            # It can't be split...we just have to put it on its own line.
+            folded.append(tstr)
+            folded.newline()
+            last_ew = None
+
+    def cte_encode(self, charset, policy):
+        res = []
+        last_ew = None
+        for part in self:
+            spart = str(part)
+            try:
+                spart.encode('us-ascii')
+                res.append(spart)
+            except UnicodeEncodeError:
+                if last_ew is None:
+                    res.append(part.cte_encode(charset, policy))
+                    last_ew = len(res)
+                else:
+                    tl = get_unstructured(''.join(res[last_ew:] + [spart]))
+                    res.append(tl.as_encoded_word())
+        return ''.join(res)
+
+
+class Phrase(TokenList):
+
+    token_type = 'phrase'
+
+    def _fold(self, folded):
+        # As with Unstructured, we can have pure ASCII with or without
+        # surrogateescape encoded bytes, or we could have unicode.  But this
+        # case is more complicated, since we have to deal with the various
+        # sub-token types and how they can be composed in the face of
+        # unicode-that-needs-CTE-encoding, and the fact that if a token a
+        # comment that becomes a barrier across which we can't compose encoded
+        # words.
+        last_ew = None
+        for part in self.parts:
+            tstr = str(part)
+            tlen = len(tstr)
+            has_ew = False
+            try:
+                str(part).encode('us-ascii')
+            except UnicodeEncodeError:
+                if any(isinstance(x, errors.UndecodableBytesDefect)
+                        for x in part.all_defects):
+                    charset = 'unknown-8bit'
+                else:
+                    charset = 'utf-8'
+                if last_ew is not None and not part.has_leading_comment():
+                    # We've already done an EW, let's see if we can combine
+                    # this one with it.  The last_ew logic ensures that all we
+                    # have at this point is atoms, no comments or quoted
+                    # strings.  So we can treat the text between the last
+                    # encoded word and the content of this token as
+                    # unstructured text, and things will work correctly.  But
+                    # we have to strip off any trailing comment on this token
+                    # first, and if it is a quoted string we have to pull out
+                    # the content (we're encoding it, so it no longer needs to
+                    # be quoted).
+                    if part[-1].token_type == 'cfws' and part.comments:
+                        remainder = part.pop(-1)
+                    else:
+                        remainder = ''
+                    for i, token in enumerate(part):
+                        if token.token_type == 'bare-quoted-string':
+                            part[i] = UnstructuredTokenList(token[:])
+                    chunk = get_unstructured(
+                        ''.join(folded.current[last_ew:]+[tstr])).as_encoded_word(charset)
+                    schunk = str(chunk)
+                    lchunk = len(schunk)
+                    if last_ew + lchunk <= folded.maxlen:
+                        del folded.current[last_ew:]
+                        folded.append(schunk)
+                        folded.lastlen = sum(len(x) for x in folded.current)
+                        continue
+                tstr = part.as_encoded_word(charset)
+                tlen = len(tstr)
+                has_ew = True
+            if folded.append_if_fits(part, tstr):
+                if has_ew and not part.comments:
+                    last_ew = len(folded.current) - 1
+                elif part.comments or part.token_type == 'quoted-string':
+                    # If a comment is involved we can't combine EWs.  And if a
+                    # quoted string is involved, it's not worth the effort to
+                    # try to combine them.
+                    last_ew = None
+                continue
+            part._fold(folded)
+
+    def cte_encode(self, charset, policy):
+        res = []
+        last_ew = None
+        is_ew = False
+        for part in self:
+            spart = str(part)
+            try:
+                spart.encode('us-ascii')
+                res.append(spart)
+            except UnicodeEncodeError:
+                is_ew = True
+                if last_ew is None:
+                    if not part.comments:
+                        last_ew = len(res)
+                    res.append(part.cte_encode(charset, policy))
+                elif not part.has_leading_comment():
+                    if part[-1].token_type == 'cfws' and part.comments:
+                        remainder = part.pop(-1)
+                    else:
+                        remainder = ''
+                    for i, token in enumerate(part):
+                        if token.token_type == 'bare-quoted-string':
+                            part[i] = UnstructuredTokenList(token[:])
+                    tl = get_unstructured(''.join(res[last_ew:] + [spart]))
+                    res[last_ew:] = [tl.as_encoded_word(charset)]
+            if part.comments or (not is_ew and part.token_type == 'quoted-string'):
+                last_ew = None
+        return ''.join(res)
+
+class Word(TokenList):
+
+    token_type = 'word'
+
+
+class CFWSList(WhiteSpaceTokenList):
+
+    token_type = 'cfws'
+
+    def has_leading_comment(self):
+        return bool(self.comments)
+
+
+class Atom(TokenList):
+
+    token_type = 'atom'
+
+
+class EncodedWord(TokenList):
+
+    token_type = 'encoded-word'
+    cte = None
+    charset = None
+    lang = None
+
+    @property
+    def encoded(self):
+        if self.cte is not None:
+            return self.cte
+        _ew.encode(str(self), self.charset)
+
+
+
+class QuotedString(TokenList):
+
+    token_type = 'quoted-string'
+
+    @property
+    def content(self):
+        for x in self:
+            if x.token_type == 'bare-quoted-string':
+                return x.value
+
+    @property
+    def quoted_value(self):
+        res = []
+        for x in self:
+            if x.token_type == 'bare-quoted-string':
+                res.append(str(x))
+            else:
+                res.append(x.value)
+        return ''.join(res)
+
+
+class BareQuotedString(QuotedString):
+
+    token_type = 'bare-quoted-string'
+
+    def __str__(self):
+        return quote_string(''.join(self))
+
+    @property
+    def value(self):
+        return ''.join(str(x) for x in self)
+
+
+class Comment(WhiteSpaceTokenList):
+
+    token_type = 'comment'
+
+    def __str__(self):
+        return ''.join(sum([
+                            ["("],
+                            [self.quote(x) for x in self],
+                            [")"],
+                            ], []))
+
+    def quote(self, value):
+        if value.token_type == 'comment':
+            return str(value)
+        return str(value).replace('\\', '\\\\').replace(
+                                  '(', '\(').replace(
+                                  ')', '\)')
+
+    @property
+    def content(self):
+        return ''.join(str(x) for x in self)
+
+    @property
+    def comments(self):
+        return [self.content]
+
+class AddressList(TokenList):
+
+    token_type = 'address-list'
+
+    @property
+    def addresses(self):
+        return [x for x in self if x.token_type=='address']
+
+    @property
+    def mailboxes(self):
+        return sum((x.mailboxes
+                    for x in self if x.token_type=='address'), [])
+
+    @property
+    def all_mailboxes(self):
+        return sum((x.all_mailboxes
+                    for x in self if x.token_type=='address'), [])
+
+
+class Address(TokenList):
+
+    token_type = 'address'
+
+    @property
+    def display_name(self):
+        if self[0].token_type == 'group':
+            return self[0].display_name
+
+    @property
+    def mailboxes(self):
+        if self[0].token_type == 'mailbox':
+            return [self[0]]
+        elif self[0].token_type == 'invalid-mailbox':
+            return []
+        return self[0].mailboxes
+
+    @property
+    def all_mailboxes(self):
+        if self[0].token_type == 'mailbox':
+            return [self[0]]
+        elif self[0].token_type == 'invalid-mailbox':
+            return [self[0]]
+        return self[0].all_mailboxes
+
+class MailboxList(TokenList):
+
+    token_type = 'mailbox-list'
+
+    @property
+    def mailboxes(self):
+        return [x for x in self if x.token_type=='mailbox']
+
+    @property
+    def all_mailboxes(self):
+        return [x for x in self
+            if x.token_type in ('mailbox', 'invalid-mailbox')]
+
+
+class GroupList(TokenList):
+
+    token_type = 'group-list'
+
+    @property
+    def mailboxes(self):
+        if not self or self[0].token_type != 'mailbox-list':
+            return []
+        return self[0].mailboxes
+
+    @property
+    def all_mailboxes(self):
+        if not self or self[0].token_type != 'mailbox-list':
+            return []
+        return self[0].all_mailboxes
+
+
+class Group(TokenList):
+
+    token_type = "group"
+
+    @property
+    def mailboxes(self):
+        if self[2].token_type != 'group-list':
+            return []
+        return self[2].mailboxes
+
+    @property
+    def all_mailboxes(self):
+        if self[2].token_type != 'group-list':
+            return []
+        return self[2].all_mailboxes
+
+    @property
+    def display_name(self):
+        return self[0].display_name
+
+
+class NameAddr(TokenList):
+
+    token_type = 'name-addr'
+
+    @property
+    def display_name(self):
+        if len(self) == 1:
+            return None
+        return self[0].display_name
+
+    @property
+    def local_part(self):
+        return self[-1].local_part
+
+    @property
+    def domain(self):
+        return self[-1].domain
+
+    @property
+    def route(self):
+        return self[-1].route
+
+    @property
+    def addr_spec(self):
+        return self[-1].addr_spec
+
+
+class AngleAddr(TokenList):
+
+    token_type = 'angle-addr'
+
+    @property
+    def local_part(self):
+        for x in self:
+            if x.token_type == 'addr-spec':
+                return x.local_part
+
+    @property
+    def domain(self):
+        for x in self:
+            if x.token_type == 'addr-spec':
+                return x.domain
+
+    @property
+    def route(self):
+        for x in self:
+            if x.token_type == 'obs-route':
+                return x.domains
+
+    @property
+    def addr_spec(self):
+        for x in self:
+            if x.token_type == 'addr-spec':
+                return x.addr_spec
+
+
+class ObsRoute(TokenList):
+
+    token_type = 'obs-route'
+
+    @property
+    def domains(self):
+        return [x.domain for x in self if x.token_type == 'domain']
+
+
+class Mailbox(TokenList):
+
+    token_type = 'mailbox'
+
+    @property
+    def display_name(self):
+        if self[0].token_type == 'name-addr':
+            return self[0].display_name
+
+    @property
+    def local_part(self):
+        return self[0].local_part
+
+    @property
+    def domain(self):
+        return self[0].domain
+
+    @property
+    def route(self):
+        if self[0].token_type == 'name-addr':
+            return self[0].route
+
+    @property
+    def addr_spec(self):
+        return self[0].addr_spec
+
+
+class InvalidMailbox(TokenList):
+
+    token_type = 'invalid-mailbox'
+
+    @property
+    def display_name(self):
+        return None
+
+    local_part = domain = route = addr_spec = display_name
+
+
+class Domain(TokenList):
+
+    token_type = 'domain'
+
+    @property
+    def domain(self):
+        return ''.join(super().value.split())
+
+
+class DotAtom(TokenList):
+
+    token_type = 'dot-atom'
+
+
+class DotAtomText(TokenList):
+
+    token_type = 'dot-atom-text'
+
+
+class AddrSpec(TokenList):
+
+    token_type = 'addr-spec'
+
+    @property
+    def local_part(self):
+        return self[0].local_part
+
+    @property
+    def domain(self):
+        if len(self) < 3:
+            return None
+        return self[-1].domain
+
+    @property
+    def value(self):
+        if len(self) < 3:
+            return self[0].value
+        return self[0].value.rstrip()+self[1].value+self[2].value.lstrip()
+
+    @property
+    def addr_spec(self):
+        nameset = set(self.local_part)
+        if len(nameset) > len(nameset-DOT_ATOM_ENDS):
+            lp = quote_string(self.local_part)
+        else:
+            lp = self.local_part
+        if self.domain is not None:
+            return lp + '@' + self.domain
+        return lp
+
+
+class ObsLocalPart(TokenList):
+
+    token_type = 'obs-local-part'
+
+
+class DisplayName(Phrase):
+
+    token_type = 'display-name'
+
+    @property
+    def display_name(self):
+        res = TokenList(self)
+        if res[0].token_type == 'cfws':
+            res.pop(0)
+        else:
+            if res[0][0].token_type == 'cfws':
+                res[0] = TokenList(res[0][1:])
+        if res[-1].token_type == 'cfws':
+            res.pop()
+        else:
+            if res[-1][-1].token_type == 'cfws':
+                res[-1] = TokenList(res[-1][:-1])
+        return res.value
+
+    @property
+    def value(self):
+        quote = False
+        if self.defects:
+            quote = True
+        else:
+            for x in self:
+                if x.token_type == 'quoted-string':
+                    quote = True
+        if quote:
+            pre = post = ''
+            if self[0].token_type=='cfws' or self[0][0].token_type=='cfws':
+                pre = ' '
+            if self[-1].token_type=='cfws' or self[-1][-1].token_type=='cfws':
+                post = ' '
+            return pre+quote_string(self.display_name)+post
+        else:
+            return super().value
+
+
+class LocalPart(TokenList):
+
+    token_type = 'local-part'
+
+    @property
+    def value(self):
+        if self[0].token_type == "quoted-string":
+            return self[0].quoted_value
+        else:
+            return self[0].value
+
+    @property
+    def local_part(self):
+        # Strip whitespace from front, back, and around dots.
+        res = [DOT]
+        last = DOT
+        last_is_tl = False
+        for tok in self[0] + [DOT]:
+            if tok.token_type == 'cfws':
+                continue
+            if (last_is_tl and tok.token_type == 'dot' and
+                    last[-1].token_type == 'cfws'):
+                res[-1] = TokenList(last[:-1])
+            is_tl = isinstance(tok, TokenList)
+            if (is_tl and last.token_type == 'dot' and
+                    tok[0].token_type == 'cfws'):
+                res.append(TokenList(tok[1:]))
+            else:
+                res.append(tok)
+            last = res[-1]
+            last_is_tl = is_tl
+        res = TokenList(res[1:-1])
+        return res.value
+
+
+class DomainLiteral(TokenList):
+
+    token_type = 'domain-literal'
+
+    @property
+    def domain(self):
+        return ''.join(super().value.split())
+
+    @property
+    def ip(self):
+        for x in self:
+            if x.token_type == 'ptext':
+                return x.value
+
+
+class HeaderLabel(TokenList):
+
+    token_type = 'header-label'
+
+
+class Header(TokenList):
+
+    token_type = 'header'
+
+    def _fold(self, folded):
+        folded.append(str(self.pop(0)))
+        folded.lastlen = len(folded.current[0])
+        # The first line of the header is different from all others: we don't
+        # want to start a new object on a new line if it has any fold points in
+        # it that would allow part of it to be on the first header line.
+        # Further, if the first fold point would fit on the new line, we want
+        # to do that, but if it doesn't we want to put it on the first line.
+        # Folded supports this via the stickyspace attribute.  If this
+        # attribute is not None, it does the special handling.
+        folded.stickyspace = str(self.pop(0)) if self[0].token_type == 'cfws' else ''
+        rest = self.pop(0)
+        if self:
+            raise ValueError("Malformed Header token list")
+        rest._fold(folded)
+
+
+#
+# Terminal classes and instances
+#
+
+class Terminal(str):
+
+    def __new__(cls, value, token_type):
+        self = super().__new__(cls, value)
+        self.token_type = token_type
+        self.defects = []
+        return self
+
+    def __repr__(self):
+        return "{}({})".format(self.__class__.__name__, super().__repr__())
+
+    @property
+    def all_defects(self):
+        return list(self.defects)
+
+    def _pp(self, indent=''):
+        return ["{}{}/{}({}){}".format(
+            indent,
+            self.__class__.__name__,
+            self.token_type,
+            super().__repr__(),
+            '' if not self.defects else ' {}'.format(self.defects),
+            )]
+
+    def cte_encode(self, charset, policy):
+        value = str(self)
+        try:
+            value.encode('us-ascii')
+            return value
+        except UnicodeEncodeError:
+            return _ew.encode(value, charset)
+
+    def pop_trailing_ws(self):
+        # This terminates the recursion.
+        return None
+
+    def pop_leading_fws(self):
+        # This terminates the recursion.
+        return None
+
+    @property
+    def comments(self):
+        return []
+
+    def has_leading_comment(self):
+        return False
+
+    def __getnewargs__(self):
+        return(str(self), self.token_type)
+
+
+class WhiteSpaceTerminal(Terminal):
+
+    @property
+    def value(self):
+        return ' '
+
+    def startswith_fws(self):
+        return True
+
+    has_fws = True
+
+
+class ValueTerminal(Terminal):
+
+    @property
+    def value(self):
+        return self
+
+    def startswith_fws(self):
+        return False
+
+    has_fws = False
+
+    def as_encoded_word(self, charset):
+        return _ew.encode(str(self), charset)
+
+
+class EWWhiteSpaceTerminal(WhiteSpaceTerminal):
+
+    @property
+    def value(self):
+        return ''
+
+    @property
+    def encoded(self):
+        return self[:]
+
+    def __str__(self):
+        return ''
+
+    has_fws = True
+
+
+# XXX these need to become classes and used as instances so
+# that a program can't change them in a parse tree and screw
+# up other parse trees.  Maybe should have  tests for that, too.
+DOT = ValueTerminal('.', 'dot')
+ListSeparator = ValueTerminal(',', 'list-separator')
+RouteComponentMarker = ValueTerminal('@', 'route-component-marker')
+
+#
+# Parser
+#
+
+"""Parse strings according to RFC822/2047/2822/5322 rules.
+
+This is a stateless parser.  Each get_XXX function accepts a string and
+returns either a Terminal or a TokenList representing the RFC object named
+by the method and a string containing the remaining unparsed characters
+from the input.  Thus a parser method consumes the next syntactic construct
+of a given type and returns a token representing the construct plus the
+unparsed remainder of the input string.
+
+For example, if the first element of a structured header is a 'phrase',
+then:
+
+    phrase, value = get_phrase(value)
+
+returns the complete phrase from the start of the string value, plus any
+characters left in the string after the phrase is removed.
+
+"""
+
+_wsp_splitter = re.compile(r'([{}]+)'.format(''.join(WSP))).split
+_non_atom_end_matcher = re.compile(r"[^{}]+".format(
+    ''.join(ATOM_ENDS).replace('\\','\\\\').replace(']','\]'))).match
+_non_printable_finder = re.compile(r"[\x00-\x20\x7F]").findall
+
+def _validate_xtext(xtext):
+    """If input token contains ASCII non-printables, register a defect."""
+
+    non_printables = _non_printable_finder(xtext)
+    if non_printables:
+        xtext.defects.append(errors.NonPrintableDefect(non_printables))
+    if utils._has_surrogates(xtext):
+        xtext.defects.append(errors.UndecodableBytesDefect(
+            "Non-ASCII characters found in header token"))
+
+def _get_ptext_to_endchars(value, endchars):
+    """Scan printables/quoted-pairs until endchars and return unquoted ptext.
+
+    This function turns a run of qcontent, ccontent-without-comments, or
+    dtext-with-quoted-printables into a single string by unquoting any
+    quoted printables.  It returns the string, the remaining value, and
+    a flag that is True iff there were any quoted printables decoded.
+
+    """
+    fragment, *remainder = _wsp_splitter(value, 1)
+    vchars = []
+    escape = False
+    had_qp = False
+    for pos in range(len(fragment)):
+        if fragment[pos] == '\\':
+            if escape:
+                escape = False
+                had_qp = True
+            else:
+                escape = True
+                continue
+        if escape:
+            escape = False
+        elif fragment[pos] in endchars:
+            break
+        vchars.append(fragment[pos])
+    else:
+        pos = pos + 1
+    return ''.join(vchars), ''.join([fragment[pos:]] + remainder), had_qp
+
+def _decode_ew_run(value):
+    """ Decode a run of RFC2047 encoded words.
+
+        _decode_ew_run(value) -> (text, value, defects)
+
+    Scans the supplied value for a run of tokens that look like they are RFC
+    2047 encoded words, decodes those words into text according to RFC 2047
+    rules (whitespace between encoded words is discarded), and returns the text
+    and the remaining value (including any leading whitespace on the remaining
+    value), as well as a list of any defects encountered while decoding.  The
+    input value may not have any leading whitespace.
+
+    """
+    res = []
+    defects = []
+    last_ws = ''
+    while value:
+        try:
+            tok, ws, value = _wsp_splitter(value, 1)
+        except ValueError:
+            tok, ws, value = value, '', ''
+        if not (tok.startswith('=?') and tok.endswith('?=')):
+            return ''.join(res), last_ws + tok + ws + value, defects
+        text, charset, lang, new_defects = _ew.decode(tok)
+        res.append(text)
+        defects.extend(new_defects)
+        last_ws = ws
+    return ''.join(res), last_ws, defects
+
+def get_fws(value):
+    """FWS = 1*WSP
+
+    This isn't the RFC definition.  We're using fws to represent tokens where
+    folding can be done, but when we are parsing the *un*folding has already
+    been done so we don't need to watch out for CRLF.
+
+    """
+    newvalue = value.lstrip()
+    fws = WhiteSpaceTerminal(value[:len(value)-len(newvalue)], 'fws')
+    return fws, newvalue
+
+def get_encoded_word(value):
+    """ encoded-word = "=?" charset "?" encoding "?" encoded-text "?="
+
+    """
+    ew = EncodedWord()
+    if not value.startswith('=?'):
+        raise errors.HeaderParseError(
+            "expected encoded word but found {}".format(value))
+    tok, *remainder = value[2:].split('?=', 1)
+    if tok == value[2:]:
+        raise errors.HeaderParseError(
+            "expected encoded word but found {}".format(value))
+    remstr = ''.join(remainder)
+    if remstr[:2].isdigit():
+        rest, *remainder = remstr.split('?=', 1)
+        tok = tok + '?=' + rest
+    if len(tok.split()) > 1:
+        ew.defects.append(errors.InvalidHeaderDefect(
+            "whitespace inside encoded word"))
+    ew.cte = value
+    value = ''.join(remainder)
+    try:
+        text, charset, lang, defects = _ew.decode('=?' + tok + '?=')
+    except ValueError:
+        raise errors.HeaderParseError(
+            "encoded word format invalid: '{}'".format(ew.cte))
+    ew.charset = charset
+    ew.lang = lang
+    ew.defects.extend(defects)
+    while text:
+        if text[0] in WSP:
+            token, text = get_fws(text)
+            ew.append(token)
+            continue
+        chars, *remainder = _wsp_splitter(text, 1)
+        vtext = ValueTerminal(chars, 'vtext')
+        _validate_xtext(vtext)
+        ew.append(vtext)
+        text = ''.join(remainder)
+    return ew, value
+
+def get_unstructured(value):
+    """unstructured = (*([FWS] vchar) *WSP) / obs-unstruct
+       obs-unstruct = *((*LF *CR *(obs-utext) *LF *CR)) / FWS)
+       obs-utext = %d0 / obs-NO-WS-CTL / LF / CR
+
+       obs-NO-WS-CTL is control characters except WSP/CR/LF.
+
+    So, basically, we have printable runs, plus control characters or nulls in
+    the obsolete syntax, separated by whitespace.  Since RFC 2047 uses the
+    obsolete syntax in its specification, but requires whitespace on either
+    side of the encoded words, I can see no reason to need to separate the
+    non-printable-non-whitespace from the printable runs if they occur, so we
+    parse this into xtext tokens separated by WSP tokens.
+
+    Because an 'unstructured' value must by definition constitute the entire
+    value, this 'get' routine does not return a remaining value, only the
+    parsed TokenList.
+
+    """
+    # XXX: but what about bare CR and LF?  They might signal the start or
+    # end of an encoded word.  YAGNI for now, since out current parsers
+    # will never send us strings with bard CR or LF.
+
+    unstructured = UnstructuredTokenList()
+    while value:
+        if value[0] in WSP:
+            token, value = get_fws(value)
+            unstructured.append(token)
+            continue
+        if value.startswith('=?'):
+            try:
+                token, value = get_encoded_word(value)
+            except errors.HeaderParseError:
+                pass
+            else:
+                have_ws = True
+                if len(unstructured) > 0:
+                    if unstructured[-1].token_type != 'fws':
+                        unstructured.defects.append(errors.InvalidHeaderDefect(
+                            "missing whitespace before encoded word"))
+                        have_ws = False
+                if have_ws and len(unstructured) > 1:
+                    if unstructured[-2].token_type == 'encoded-word':
+                        unstructured[-1] = EWWhiteSpaceTerminal(
+                            unstructured[-1], 'fws')
+                unstructured.append(token)
+                continue
+        tok, *remainder = _wsp_splitter(value, 1)
+        vtext = ValueTerminal(tok, 'vtext')
+        _validate_xtext(vtext)
+        unstructured.append(vtext)
+        value = ''.join(remainder)
+    return unstructured
+
+def get_qp_ctext(value):
+    """ctext = <printable ascii except \ ( )>
+
+    This is not the RFC ctext, since we are handling nested comments in comment
+    and unquoting quoted-pairs here.  We allow anything except the '()'
+    characters, but if we find any ASCII other than the RFC defined printable
+    ASCII an NonPrintableDefect is added to the token's defects list.  Since
+    quoted pairs are converted to their unquoted values, what is returned is
+    a 'ptext' token.  In this case it is a WhiteSpaceTerminal, so it's value
+    is ' '.
+
+    """
+    ptext, value, _ = _get_ptext_to_endchars(value, '()')
+    ptext = WhiteSpaceTerminal(ptext, 'ptext')
+    _validate_xtext(ptext)
+    return ptext, value
+
+def get_qcontent(value):
+    """qcontent = qtext / quoted-pair
+
+    We allow anything except the DQUOTE character, but if we find any ASCII
+    other than the RFC defined printable ASCII an NonPrintableDefect is
+    added to the token's defects list.  Any quoted pairs are converted to their
+    unquoted values, so what is returned is a 'ptext' token.  In this case it
+    is a ValueTerminal.
+
+    """
+    ptext, value, _ = _get_ptext_to_endchars(value, '"')
+    ptext = ValueTerminal(ptext, 'ptext')
+    _validate_xtext(ptext)
+    return ptext, value
+
+def get_atext(value):
+    """atext = <matches _atext_matcher>
+
+    We allow any non-ATOM_ENDS in atext, but add an InvalidATextDefect to
+    the token's defects list if we find non-atext characters.
+    """
+    m = _non_atom_end_matcher(value)
+    if not m:
+        raise errors.HeaderParseError(
+            "expected atext but found '{}'".format(value))
+    atext = m.group()
+    value = value[len(atext):]
+    atext = ValueTerminal(atext, 'atext')
+    _validate_xtext(atext)
+    return atext, value
+
+def get_bare_quoted_string(value):
+    """bare-quoted-string = DQUOTE *([FWS] qcontent) [FWS] DQUOTE
+
+    A quoted-string without the leading or trailing white space.  Its
+    value is the text between the quote marks, with whitespace
+    preserved and quoted pairs decoded.
+    """
+    if value[0] != '"':
+        raise errors.HeaderParseError(
+            "expected '\"' but found '{}'".format(value))
+    bare_quoted_string = BareQuotedString()
+    value = value[1:]
+    while value and value[0] != '"':
+        if value[0] in WSP:
+            token, value = get_fws(value)
+        else:
+            token, value = get_qcontent(value)
+        bare_quoted_string.append(token)
+    if not value:
+        bare_quoted_string.defects.append(errors.InvalidHeaderDefect(
+            "end of header inside quoted string"))
+        return bare_quoted_string, value
+    return bare_quoted_string, value[1:]
+
+def get_comment(value):
+    """comment = "(" *([FWS] ccontent) [FWS] ")"
+       ccontent = ctext / quoted-pair / comment
+
+    We handle nested comments here, and quoted-pair in our qp-ctext routine.
+    """
+    if value and value[0] != '(':
+        raise errors.HeaderParseError(
+            "expected '(' but found '{}'".format(value))
+    comment = Comment()
+    value = value[1:]
+    while value and value[0] != ")":
+        if value[0] in WSP:
+            token, value = get_fws(value)
+        elif value[0] == '(':
+            token, value = get_comment(value)
+        else:
+            token, value = get_qp_ctext(value)
+        comment.append(token)
+    if not value:
+        comment.defects.append(errors.InvalidHeaderDefect(
+            "end of header inside comment"))
+        return comment, value
+    return comment, value[1:]
+
+def get_cfws(value):
+    """CFWS = (1*([FWS] comment) [FWS]) / FWS
+
+    """
+    cfws = CFWSList()
+    while value and value[0] in CFWS_LEADER:
+        if value[0] in WSP:
+            token, value = get_fws(value)
+        else:
+            token, value = get_comment(value)
+        cfws.append(token)
+    return cfws, value
+
+def get_quoted_string(value):
+    """quoted-string = [CFWS] <bare-quoted-string> [CFWS]
+
+    'bare-quoted-string' is an intermediate class defined by this
+    parser and not by the RFC grammar.  It is the quoted string
+    without any attached CFWS.
+    """
+    quoted_string = QuotedString()
+    if value and value[0] in CFWS_LEADER:
+        token, value = get_cfws(value)
+        quoted_string.append(token)
+    token, value = get_bare_quoted_string(value)
+    quoted_string.append(token)
+    if value and value[0] in CFWS_LEADER:
+        token, value = get_cfws(value)
+        quoted_string.append(token)
+    return quoted_string, value
+
+def get_atom(value):
+    """atom = [CFWS] 1*atext [CFWS]
+
+    """
+    atom = Atom()
+    if value and value[0] in CFWS_LEADER:
+        token, value = get_cfws(value)
+        atom.append(token)
+    if value and value[0] in ATOM_ENDS:
+        raise errors.HeaderParseError(
+            "expected atom but found '{}'".format(value))
+    token, value = get_atext(value)
+    atom.append(token)
+    if value and value[0] in CFWS_LEADER:
+        token, value = get_cfws(value)
+        atom.append(token)
+    return atom, value
+
+def get_dot_atom_text(value):
+    """ dot-text = 1*atext *("." 1*atext)
+
+    """
+    dot_atom_text = DotAtomText()
+    if not value or value[0] in ATOM_ENDS:
+        raise errors.HeaderParseError("expected atom at a start of "
+            "dot-atom-text but found '{}'".format(value))
+    while value and value[0] not in ATOM_ENDS:
+        token, value = get_atext(value)
+        dot_atom_text.append(token)
+        if value and value[0] == '.':
+            dot_atom_text.append(DOT)
+            value = value[1:]
+    if dot_atom_text[-1] is DOT:
+        raise errors.HeaderParseError("expected atom at end of dot-atom-text "
+            "but found '{}'".format('.'+value))
+    return dot_atom_text, value
+
+def get_dot_atom(value):
+    """ dot-atom = [CFWS] dot-atom-text [CFWS]
+
+    """
+    dot_atom = DotAtom()
+    if value[0] in CFWS_LEADER:
+        token, value = get_cfws(value)
+        dot_atom.append(token)
+    token, value = get_dot_atom_text(value)
+    dot_atom.append(token)
+    if value and value[0] in CFWS_LEADER:
+        token, value = get_cfws(value)
+        dot_atom.append(token)
+    return dot_atom, value
+
+def get_word(value):
+    """word = atom / quoted-string
+
+    Either atom or quoted-string may start with CFWS.  We have to peel off this
+    CFWS first to determine which type of word to parse.  Afterward we splice
+    the leading CFWS, if any, into the parsed sub-token.
+
+    If neither an atom or a quoted-string is found before the next special, a
+    HeaderParseError is raised.
+
+    The token returned is either an Atom or a QuotedString, as appropriate.
+    This means the 'word' level of the formal grammar is not represented in the
+    parse tree; this is because having that extra layer when manipulating the
+    parse tree is more confusing than it is helpful.
+
+    """
+    if value[0] in CFWS_LEADER:
+        leader, value = get_cfws(value)
+    else:
+        leader = None
+    if value[0]=='"':
+        token, value = get_quoted_string(value)
+    elif value[0] in SPECIALS:
+        raise errors.HeaderParseError("Expected 'atom' or 'quoted-string' "
+                                      "but found '{}'".format(value))
+    else:
+        token, value = get_atom(value)
+    if leader is not None:
+        token[:0] = [leader]
+    return token, value
+
+def get_phrase(value):
+    """ phrase = 1*word / obs-phrase
+        obs-phrase = word *(word / "." / CFWS)
+
+    This means a phrase can be a sequence of words, periods, and CFWS in any
+    order as long as it starts with at least one word.  If anything other than
+    words is detected, an ObsoleteHeaderDefect is added to the token's defect
+    list.  We also accept a phrase that starts with CFWS followed by a dot;
+    this is registered as an InvalidHeaderDefect, since it is not supported by
+    even the obsolete grammar.
+
+    """
+    phrase = Phrase()
+    try:
+        token, value = get_word(value)
+        phrase.append(token)
+    except errors.HeaderParseError:
+        phrase.defects.append(errors.InvalidHeaderDefect(
+            "phrase does not start with word"))
+    while value and value[0] not in PHRASE_ENDS:
+        if value[0]=='.':
+            phrase.append(DOT)
+            phrase.defects.append(errors.ObsoleteHeaderDefect(
+                "period in 'phrase'"))
+            value = value[1:]
+        else:
+            try:
+                token, value = get_word(value)
+            except errors.HeaderParseError:
+                if value[0] in CFWS_LEADER:
+                    token, value = get_cfws(value)
+                    phrase.defects.append(errors.ObsoleteHeaderDefect(
+                        "comment found without atom"))
+                else:
+                    raise
+            phrase.append(token)
+    return phrase, value
+
+def get_local_part(value):
+    """ local-part = dot-atom / quoted-string / obs-local-part
+
+    """
+    local_part = LocalPart()
+    leader = None
+    if value[0] in CFWS_LEADER:
+        leader, value = get_cfws(value)
+    if not value:
+        raise errors.HeaderParseError(
+            "expected local-part but found '{}'".format(value))
+    try:
+        token, value = get_dot_atom(value)
+    except errors.HeaderParseError:
+        try:
+            token, value = get_word(value)
+        except errors.HeaderParseError:
+            if value[0] != '\\' and value[0] in PHRASE_ENDS:
+                raise
+            token = TokenList()
+    if leader is not None:
+        token[:0] = [leader]
+    local_part.append(token)
+    if value and (value[0]=='\\' or value[0] not in PHRASE_ENDS):
+        obs_local_part, value = get_obs_local_part(str(local_part) + value)
+        if obs_local_part.token_type == 'invalid-obs-local-part':
+            local_part.defects.append(errors.InvalidHeaderDefect(
+                "local-part is not dot-atom, quoted-string, or obs-local-part"))
+        else:
+            local_part.defects.append(errors.ObsoleteHeaderDefect(
+                "local-part is not a dot-atom (contains CFWS)"))
+        local_part[0] = obs_local_part
+    try:
+        local_part.value.encode('ascii')
+    except UnicodeEncodeError:
+        local_part.defects.append(errors.NonASCIILocalPartDefect(
+                "local-part contains non-ASCII characters)"))
+    return local_part, value
+
+def get_obs_local_part(value):
+    """ obs-local-part = word *("." word)
+    """
+    obs_local_part = ObsLocalPart()
+    last_non_ws_was_dot = False
+    while value and (value[0]=='\\' or value[0] not in PHRASE_ENDS):
+        if value[0] == '.':
+            if last_non_ws_was_dot:
+                obs_local_part.defects.append(errors.InvalidHeaderDefect(
+                    "invalid repeated '.'"))
+            obs_local_part.append(DOT)
+            last_non_ws_was_dot = True
+            value = value[1:]
+            continue
+        elif value[0]=='\\':
+            obs_local_part.append(ValueTerminal(value[0],
+                                                'misplaced-special'))
+            value = value[1:]
+            obs_local_part.defects.append(errors.InvalidHeaderDefect(
+                "'\\' character outside of quoted-string/ccontent"))
+            last_non_ws_was_dot = False
+            continue
+        if obs_local_part and obs_local_part[-1].token_type != 'dot':
+            obs_local_part.defects.append(errors.InvalidHeaderDefect(
+                "missing '.' between words"))
+        try:
+            token, value = get_word(value)
+            last_non_ws_was_dot = False
+        except errors.HeaderParseError:
+            if value[0] not in CFWS_LEADER:
+                raise
+            token, value = get_cfws(value)
+        obs_local_part.append(token)
+    if (obs_local_part[0].token_type == 'dot' or
+            obs_local_part[0].token_type=='cfws' and
+            obs_local_part[1].token_type=='dot'):
+        obs_local_part.defects.append(errors.InvalidHeaderDefect(
+            "Invalid leading '.' in local part"))
+    if (obs_local_part[-1].token_type == 'dot' or
+            obs_local_part[-1].token_type=='cfws' and
+            obs_local_part[-2].token_type=='dot'):
+        obs_local_part.defects.append(errors.InvalidHeaderDefect(
+            "Invalid trailing '.' in local part"))
+    if obs_local_part.defects:
+        obs_local_part.token_type = 'invalid-obs-local-part'
+    return obs_local_part, value
+
+def get_dtext(value):
+    """ dtext = <printable ascii except \ [ ]> / obs-dtext
+        obs-dtext = obs-NO-WS-CTL / quoted-pair
+
+    We allow anything except the excluded characters, but but if we find any
+    ASCII other than the RFC defined printable ASCII an NonPrintableDefect is
+    added to the token's defects list.  Quoted pairs are converted to their
+    unquoted values, so what is returned is a ptext token, in this case a
+    ValueTerminal.  If there were quoted-printables, an ObsoleteHeaderDefect is
+    added to the returned token's defect list.
+
+    """
+    ptext, value, had_qp = _get_ptext_to_endchars(value, '[]')
+    ptext = ValueTerminal(ptext, 'ptext')
+    if had_qp:
+        ptext.defects.append(errors.ObsoleteHeaderDefect(
+            "quoted printable found in domain-literal"))
+    _validate_xtext(ptext)
+    return ptext, value
+
+def _check_for_early_dl_end(value, domain_literal):
+    if value:
+        return False
+    domain_literal.append(errors.InvalidHeaderDefect(
+        "end of input inside domain-literal"))
+    domain_literal.append(ValueTerminal(']', 'domain-literal-end'))
+    return True
+
+def get_domain_literal(value):
+    """ domain-literal = [CFWS] "[" *([FWS] dtext) [FWS] "]" [CFWS]
+
+    """
+    domain_literal = DomainLiteral()
+    if value[0] in CFWS_LEADER:
+        token, value = get_cfws(value)
+        domain_literal.append(token)
+    if not value:
+        raise errors.HeaderParseError("expected domain-literal")
+    if value[0] != '[':
+        raise errors.HeaderParseError("expected '[' at start of domain-literal "
+                "but found '{}'".format(value))
+    value = value[1:]
+    if _check_for_early_dl_end(value, domain_literal):
+        return domain_literal, value
+    domain_literal.append(ValueTerminal('[', 'domain-literal-start'))
+    if value[0] in WSP:
+        token, value = get_fws(value)
+        domain_literal.append(token)
+    token, value = get_dtext(value)
+    domain_literal.append(token)
+    if _check_for_early_dl_end(value, domain_literal):
+        return domain_literal, value
+    if value[0] in WSP:
+        token, value = get_fws(value)
+        domain_literal.append(token)
+    if _check_for_early_dl_end(value, domain_literal):
+        return domain_literal, value
+    if value[0] != ']':
+        raise errors.HeaderParseError("expected ']' at end of domain-literal "
+                "but found '{}'".format(value))
+    domain_literal.append(ValueTerminal(']', 'domain-literal-end'))
+    value = value[1:]
+    if value and value[0] in CFWS_LEADER:
+        token, value = get_cfws(value)
+        domain_literal.append(token)
+    return domain_literal, value
+
+def get_domain(value):
+    """ domain = dot-atom / domain-literal / obs-domain
+        obs-domain = atom *("." atom))
+
+    """
+    domain = Domain()
+    leader = None
+    if value[0] in CFWS_LEADER:
+        leader, value = get_cfws(value)
+    if not value:
+        raise errors.HeaderParseError(
+            "expected domain but found '{}'".format(value))
+    if value[0] == '[':
+        token, value = get_domain_literal(value)
+        if leader is not None:
+            token[:0] = [leader]
+        domain.append(token)
+        return domain, value
+    try:
+        token, value = get_dot_atom(value)
+    except errors.HeaderParseError:
+        token, value = get_atom(value)
+    if leader is not None:
+        token[:0] = [leader]
+    domain.append(token)
+    if value and value[0] == '.':
+        domain.defects.append(errors.ObsoleteHeaderDefect(
+            "domain is not a dot-atom (contains CFWS)"))
+        if domain[0].token_type == 'dot-atom':
+            domain[:] = domain[0]
+        while value and value[0] == '.':
+            domain.append(DOT)
+            token, value = get_atom(value[1:])
+            domain.append(token)
+    return domain, value
+
+def get_addr_spec(value):
+    """ addr-spec = local-part "@" domain
+
+    """
+    addr_spec = AddrSpec()
+    token, value = get_local_part(value)
+    addr_spec.append(token)
+    if not value or value[0] != '@':
+        addr_spec.defects.append(errors.InvalidHeaderDefect(
+            "add-spec local part with no domain"))
+        return addr_spec, value
+    addr_spec.append(ValueTerminal('@', 'address-at-symbol'))
+    token, value = get_domain(value[1:])
+    addr_spec.append(token)
+    return addr_spec, value
+
+def get_obs_route(value):
+    """ obs-route = obs-domain-list ":"
+        obs-domain-list = *(CFWS / ",") "@" domain *("," [CFWS] ["@" domain])
+
+        Returns an obs-route token with the appropriate sub-tokens (that is,
+        there is no obs-domain-list in the parse tree).
+    """
+    obs_route = ObsRoute()
+    while value and (value[0]==',' or value[0] in CFWS_LEADER):
+        if value[0] in CFWS_LEADER:
+            token, value = get_cfws(value)
+            obs_route.append(token)
+        elif value[0] == ',':
+            obs_route.append(ListSeparator)
+            value = value[1:]
+    if not value or value[0] != '@':
+        raise errors.HeaderParseError(
+            "expected obs-route domain but found '{}'".format(value))
+    obs_route.append(RouteComponentMarker)
+    token, value = get_domain(value[1:])
+    obs_route.append(token)
+    while value and value[0]==',':
+        obs_route.append(ListSeparator)
+        value = value[1:]
+        if not value:
+            break
+        if value[0] in CFWS_LEADER:
+            token, value = get_cfws(value)
+            obs_route.append(token)
+        if value[0] == '@':
+            obs_route.append(RouteComponentMarker)
+            token, value = get_domain(value[1:])
+            obs_route.append(token)
+    if not value:
+        raise errors.HeaderParseError("end of header while parsing obs-route")
+    if value[0] != ':':
+        raise errors.HeaderParseError( "expected ':' marking end of "
+            "obs-route but found '{}'".format(value))
+    obs_route.append(ValueTerminal(':', 'end-of-obs-route-marker'))
+    return obs_route, value[1:]
+
+def get_angle_addr(value):
+    """ angle-addr = [CFWS] "<" addr-spec ">" [CFWS] / obs-angle-addr
+        obs-angle-addr = [CFWS] "<" obs-route addr-spec ">" [CFWS]
+
+    """
+    angle_addr = AngleAddr()
+    if value[0] in CFWS_LEADER:
+        token, value = get_cfws(value)
+        angle_addr.append(token)
+    if not value or value[0] != '<':
+        raise errors.HeaderParseError(
+            "expected angle-addr but found '{}'".format(value))
+    angle_addr.append(ValueTerminal('<', 'angle-addr-start'))
+    value = value[1:]
+    try:
+        token, value = get_addr_spec(value)
+    except errors.HeaderParseError:
+        try:
+            token, value = get_obs_route(value)
+            angle_addr.defects.append(errors.ObsoleteHeaderDefect(
+                "obsolete route specification in angle-addr"))
+        except errors.HeaderParseError:
+            raise errors.HeaderParseError(
+                "expected addr-spec or but found '{}'".format(value))
+        angle_addr.append(token)
+        token, value = get_addr_spec(value)
+    angle_addr.append(token)
+    if value and value[0] == '>':
+        value = value[1:]
+    else:
+        angle_addr.defects.append(errors.InvalidHeaderDefect(
+            "missing trailing '>' on angle-addr"))
+    angle_addr.append(ValueTerminal('>', 'angle-addr-end'))
+    if value and value[0] in CFWS_LEADER:
+        token, value = get_cfws(value)
+        angle_addr.append(token)
+    return angle_addr, value
+
+def get_display_name(value):
+    """ display-name = phrase
+
+    Because this is simply a name-rule, we don't return a display-name
+    token containing a phrase, but rather a display-name token with
+    the content of the phrase.
+
+    """
+    display_name = DisplayName()
+    token, value = get_phrase(value)
+    display_name.extend(token[:])
+    display_name.defects = token.defects[:]
+    return display_name, value
+
+
+def get_name_addr(value):
+    """ name-addr = [display-name] angle-addr
+
+    """
+    name_addr = NameAddr()
+    # Both the optional display name and the angle-addr can start with cfws.
+    leader = None
+    if value[0] in CFWS_LEADER:
+        leader, value = get_cfws(value)
+        if not value:
+            raise errors.HeaderParseError(
+                "expected name-addr but found '{}'".format(leader))
+    if value[0] != '<':
+        if value[0] in PHRASE_ENDS:
+            raise errors.HeaderParseError(
+                "expected name-addr but found '{}'".format(value))
+        token, value = get_display_name(value)
+        if not value:
+            raise errors.HeaderParseError(
+                "expected name-addr but found '{}'".format(token))
+        if leader is not None:
+            token[0][:0] = [leader]
+            leader = None
+        name_addr.append(token)
+    token, value = get_angle_addr(value)
+    if leader is not None:
+        token[:0] = [leader]
+    name_addr.append(token)
+    return name_addr, value
+
+def get_mailbox(value):
+    """ mailbox = name-addr / addr-spec
+
+    """
+    # The only way to figure out if we are dealing with a name-addr or an
+    # addr-spec is to try parsing each one.
+    mailbox = Mailbox()
+    try:
+        token, value = get_name_addr(value)
+    except errors.HeaderParseError:
+        try:
+            token, value = get_addr_spec(value)
+        except errors.HeaderParseError:
+            raise errors.HeaderParseError(
+                "expected mailbox but found '{}'".format(value))
+    if any(isinstance(x, errors.InvalidHeaderDefect)
+                       for x in token.all_defects):
+        mailbox.token_type = 'invalid-mailbox'
+    mailbox.append(token)
+    return mailbox, value
+
+def get_invalid_mailbox(value, endchars):
+    """ Read everything up to one of the chars in endchars.
+
+    This is outside the formal grammar.  The InvalidMailbox TokenList that is
+    returned acts like a Mailbox, but the data attributes are None.
+
+    """
+    invalid_mailbox = InvalidMailbox()
+    while value and value[0] not in endchars:
+        if value[0] in PHRASE_ENDS:
+            invalid_mailbox.append(ValueTerminal(value[0],
+                                                 'misplaced-special'))
+            value = value[1:]
+        else:
+            token, value = get_phrase(value)
+            invalid_mailbox.append(token)
+    return invalid_mailbox, value
+
+def get_mailbox_list(value):
+    """ mailbox-list = (mailbox *("," mailbox)) / obs-mbox-list
+        obs-mbox-list = *([CFWS] ",") mailbox *("," [mailbox / CFWS])
+
+    For this routine we go outside the formal grammar in order to improve error
+    handling.  We recognize the end of the mailbox list only at the end of the
+    value or at a ';' (the group terminator).  This is so that we can turn
+    invalid mailboxes into InvalidMailbox tokens and continue parsing any
+    remaining valid mailboxes.  We also allow all mailbox entries to be null,
+    and this condition is handled appropriately at a higher level.
+
+    """
+    mailbox_list = MailboxList()
+    while value and value[0] != ';':
+        try:
+            token, value = get_mailbox(value)
+            mailbox_list.append(token)
+        except errors.HeaderParseError:
+            leader = None
+            if value[0] in CFWS_LEADER:
+                leader, value = get_cfws(value)
+                if not value or value[0] in ',;':
+                    mailbox_list.append(leader)
+                    mailbox_list.defects.append(errors.ObsoleteHeaderDefect(
+                        "empty element in mailbox-list"))
+                else:
+                    token, value = get_invalid_mailbox(value, ',;')
+                    if leader is not None:
+                        token[:0] = [leader]
+                    mailbox_list.append(token)
+                    mailbox_list.defects.append(errors.InvalidHeaderDefect(
+                        "invalid mailbox in mailbox-list"))
+            elif value[0] == ',':
+                mailbox_list.defects.append(errors.ObsoleteHeaderDefect(
+                    "empty element in mailbox-list"))
+            else:
+                token, value = get_invalid_mailbox(value, ',;')
+                if leader is not None:
+                    token[:0] = [leader]
+                mailbox_list.append(token)
+                mailbox_list.defects.append(errors.InvalidHeaderDefect(
+                    "invalid mailbox in mailbox-list"))
+        if value and value[0] not in ',;':
+            # Crap after mailbox; treat it as an invalid mailbox.
+            # The mailbox info will still be available.
+            mailbox = mailbox_list[-1]
+            mailbox.token_type = 'invalid-mailbox'
+            token, value = get_invalid_mailbox(value, ',;')
+            mailbox.extend(token)
+            mailbox_list.defects.append(errors.InvalidHeaderDefect(
+                "invalid mailbox in mailbox-list"))
+        if value and value[0] == ',':
+            mailbox_list.append(ListSeparator)
+            value = value[1:]
+    return mailbox_list, value
+
+
+def get_group_list(value):
+    """ group-list = mailbox-list / CFWS / obs-group-list
+        obs-group-list = 1*([CFWS] ",") [CFWS]
+
+    """
+    group_list = GroupList()
+    if not value:
+        group_list.defects.append(errors.InvalidHeaderDefect(
+            "end of header before group-list"))
+        return group_list, value
+    leader = None
+    if value and value[0] in CFWS_LEADER:
+        leader, value = get_cfws(value)
+        if not value:
+            # This should never happen in email parsing, since CFWS-only is a
+            # legal alternative to group-list in a group, which is the only
+            # place group-list appears.
+            group_list.defects.append(errors.InvalidHeaderDefect(
+                "end of header in group-list"))
+            group_list.append(leader)
+            return group_list, value
+        if value[0] == ';':
+            group_list.append(leader)
+            return group_list, value
+    token, value = get_mailbox_list(value)
+    if len(token.all_mailboxes)==0:
+        if leader is not None:
+            group_list.append(leader)
+        group_list.extend(token)
+        group_list.defects.append(errors.ObsoleteHeaderDefect(
+            "group-list with empty entries"))
+        return group_list, value
+    if leader is not None:
+        token[:0] = [leader]
+    group_list.append(token)
+    return group_list, value
+
+def get_group(value):
+    """ group = display-name ":" [group-list] ";" [CFWS]
+
+    """
+    group = Group()
+    token, value = get_display_name(value)
+    if not value or value[0] != ':':
+        raise errors.HeaderParseError("expected ':' at end of group "
+            "display name but found '{}'".format(value))
+    group.append(token)
+    group.append(ValueTerminal(':', 'group-display-name-terminator'))
+    value = value[1:]
+    if value and value[0] == ';':
+        group.append(ValueTerminal(';', 'group-terminator'))
+        return group, value[1:]
+    token, value = get_group_list(value)
+    group.append(token)
+    if not value:
+        group.defects.append(errors.InvalidHeaderDefect(
+            "end of header in group"))
+    if value[0] != ';':
+        raise errors.HeaderParseError(
+            "expected ';' at end of group but found {}".format(value))
+    group.append(ValueTerminal(';', 'group-terminator'))
+    value = value[1:]
+    if value and value[0] in CFWS_LEADER:
+        token, value = get_cfws(value)
+        group.append(token)
+    return group, value
+
+def get_address(value):
+    """ address = mailbox / group
+
+    Note that counter-intuitively, an address can be either a single address or
+    a list of addresses (a group).  This is why the returned Address object has
+    a 'mailboxes' attribute which treats a single address as a list of length
+    one.  When you need to differentiate between to two cases, extract the single
+    element, which is either a mailbox or a group token.
+
+    """
+    # The formal grammar isn't very helpful when parsing an address.  mailbox
+    # and group, especially when allowing for obsolete forms, start off very
+    # similarly.  It is only when you reach one of @, <, or : that you know
+    # what you've got.  So, we try each one in turn, starting with the more
+    # likely of the two.  We could perhaps make this more efficient by looking
+    # for a phrase and then branching based on the next character, but that
+    # would be a premature optimization.
+    address = Address()
+    try:
+        token, value = get_group(value)
+    except errors.HeaderParseError:
+        try:
+            token, value = get_mailbox(value)
+        except errors.HeaderParseError:
+            raise errors.HeaderParseError(
+                "expected address but found '{}'".format(value))
+    address.append(token)
+    return address, value
+
+def get_address_list(value):
+    """ address_list = (address *("," address)) / obs-addr-list
+        obs-addr-list = *([CFWS] ",") address *("," [address / CFWS])
+
+    We depart from the formal grammar here by continuing to parse until the end
+    of the input, assuming the input to be entirely composed of an
+    address-list.  This is always true in email parsing, and allows us
+    to skip invalid addresses to parse additional valid ones.
+
+    """
+    address_list = AddressList()
+    while value:
+        try:
+            token, value = get_address(value)
+            address_list.append(token)
+        except errors.HeaderParseError as err:
+            leader = None
+            if value[0] in CFWS_LEADER:
+                leader, value = get_cfws(value)
+                if not value or value[0] == ',':
+                    address_list.append(leader)
+                    address_list.defects.append(errors.ObsoleteHeaderDefect(
+                        "address-list entry with no content"))
+                else:
+                    token, value = get_invalid_mailbox(value, ',')
+                    if leader is not None:
+                        token[:0] = [leader]
+                    address_list.append(Address([token]))
+                    address_list.defects.append(errors.InvalidHeaderDefect(
+                        "invalid address in address-list"))
+            elif value[0] == ',':
+                address_list.defects.append(errors.ObsoleteHeaderDefect(
+                    "empty element in address-list"))
+            else:
+                token, value = get_invalid_mailbox(value, ',')
+                if leader is not None:
+                    token[:0] = [leader]
+                address_list.append(Address([token]))
+                address_list.defects.append(errors.InvalidHeaderDefect(
+                    "invalid address in address-list"))
+        if value and value[0] != ',':
+            # Crap after address; treat it as an invalid mailbox.
+            # The mailbox info will still be available.
+            mailbox = address_list[-1][0]
+            mailbox.token_type = 'invalid-mailbox'
+            token, value = get_invalid_mailbox(value, ',')
+            mailbox.extend(token)
+            address_list.defects.append(errors.InvalidHeaderDefect(
+                "invalid address in address-list"))
+        if value:  # Must be a , at this point.
+            address_list.append(ValueTerminal(',', 'list-separator'))
+            value = value[1:]
+    return address_list, value
diff --git a/Lib/email/_headerregistry.py b/Lib/email/_headerregistry.py
new file mode 100644
index 0000000..6588546
--- /dev/null
+++ b/Lib/email/_headerregistry.py
@@ -0,0 +1,456 @@
+"""Representing and manipulating email headers via custom objects.
+
+This module provides an implementation of the HeaderRegistry API.
+The implementation is designed to flexibly follow RFC5322 rules.
+
+Eventually HeaderRegistry will be a public API, but it isn't yet,
+and will probably change some before that happens.
+
+"""
+
+from email import utils
+from email import errors
+from email import _header_value_parser as parser
+
+class Address:
+
+    def __init__(self, display_name='', username='', domain='', addr_spec=None):
+        """Create an object represeting a full email address.
+
+        An address can have a 'display_name', a 'username', and a 'domain'.  In
+        addition to specifying the username and domain separately, they may be
+        specified together by using the addr_spec keyword *instead of* the
+        username and domain keywords.  If an addr_spec string is specified it
+        must be properly quoted according to RFC 5322 rules; an error will be
+        raised if it is not.
+
+        An Address object has display_name, username, domain, and addr_spec
+        attributes, all of which are read-only.  The addr_spec and the string
+        value of the object are both quoted according to RFC5322 rules, but
+        without any Content Transfer Encoding.
+
+        """
+        # This clause with its potential 'raise' may only happen when an
+        # application program creates an Address object using an addr_spec
+        # keyword.  The email library code itself must always supply username
+        # and domain.
+        if addr_spec is not None:
+            if username or domain:
+                raise TypeError("addrspec specified when username and/or "
+                                "domain also specified")
+            a_s, rest = parser.get_addr_spec(addr_spec)
+            if rest:
+                raise ValueError("Invalid addr_spec; only '{}' "
+                                 "could be parsed from '{}'".format(
+                                    a_s, addr_spec))
+            if a_s.all_defects:
+                raise a_s.all_defects[0]
+            username = a_s.local_part
+            domain = a_s.domain
+        self._display_name = display_name
+        self._username = username
+        self._domain = domain
+
+    @property
+    def display_name(self):
+        return self._display_name
+
+    @property
+    def username(self):
+        return self._username
+
+    @property
+    def domain(self):
+        return self._domain
+
+    @property
+    def addr_spec(self):
+        """The addr_spec (username@domain) portion of the address, quoted
+        according to RFC 5322 rules, but with no Content Transfer Encoding.
+        """
+        nameset = set(self.username)
+        if len(nameset) > len(nameset-parser.DOT_ATOM_ENDS):
+            lp = parser.quote_string(self.username)
+        else:
+            lp = self.username
+        if self.domain:
+            return lp + '@' + self.domain
+        if not lp:
+            return '<>'
+        return lp
+
+    def __repr__(self):
+        return "Address(display_name={!r}, username={!r}, domain={!r})".format(
+                        self.display_name, self.username, self.domain)
+
+    def __str__(self):
+        nameset = set(self.display_name)
+        if len(nameset) > len(nameset-parser.SPECIALS):
+            disp = parser.quote_string(self.display_name)
+        else:
+            disp = self.display_name
+        if disp:
+            addr_spec = '' if self.addr_spec=='<>' else self.addr_spec
+            return "{} <{}>".format(disp, addr_spec)
+        return self.addr_spec
+
+    def __eq__(self, other):
+        if type(other) != type(self):
+            return False
+        return (self.display_name == other.display_name and
+                self.username == other.username and
+                self.domain == other.domain)
+
+
+class Group:
+
+    def __init__(self, display_name=None, addresses=None):
+        """Create an object representing an address group.
+
+        An address group consists of a display_name followed by colon and an
+        list of addresses (see Address) terminated by a semi-colon.  The Group
+        is created by specifying a display_name and a possibly empty list of
+        Address objects.  A Group can also be used to represent a single
+        address that is not in a group, which is convenient when manipulating
+        lists that are a combination of Groups and individual Addresses.  In
+        this case the display_name should be set to None.  In particular, the
+        string representation of a Group whose display_name is None is the same
+        as the Address object, if there is one and only one Address object in
+        the addresses list.
+
+        """
+        self._display_name = display_name
+        self._addresses = tuple(addresses) if addresses else tuple()
+
+    @property
+    def display_name(self):
+        return self._display_name
+
+    @property
+    def addresses(self):
+        return self._addresses
+
+    def __repr__(self):
+        return "Group(display_name={!r}, addresses={!r}".format(
+                 self.display_name, self.addresses)
+
+    def __str__(self):
+        if self.display_name is None and len(self.addresses)==1:
+            return str(self.addresses[0])
+        disp = self.display_name
+        if disp is not None:
+            nameset = set(disp)
+            if len(nameset) > len(nameset-parser.SPECIALS):
+                disp = parser.quote_string(disp)
+        adrstr = ", ".join(str(x) for x in self.addresses)
+        adrstr = ' ' + adrstr if adrstr else adrstr
+        return "{}:{};".format(disp, adrstr)
+
+    def __eq__(self, other):
+        if type(other) != type(self):
+            return False
+        return (self.display_name == other.display_name and
+                self.addresses == other.addresses)
+
+
+# Header Classes #
+
+class BaseHeader(str):
+
+    """Base class for message headers.
+
+    Implements generic behavior and provides tools for subclasses.
+
+    A subclass must define a classmethod named 'parse' that takes an unfolded
+    value string and a dictionary as its arguments.  The dictionary will
+    contain one key, 'defects', initialized to an empty list.  After the call
+    the dictionary must contain two additional keys: parse_tree, set to the
+    parse tree obtained from parsing the header, and 'decoded', set to the
+    string value of the idealized representation of the data from the value.
+    (That is, encoded words are decoded, and values that have canonical
+    representations are so represented.)
+
+    The defects key is intended to collect parsing defects, which the message
+    parser will subsequently dispose of as appropriate.  The parser should not,
+    insofar as practical, raise any errors.  Defects should be added to the
+    list instead.  The standard header parsers register defects for RFC
+    compliance issues, for obsolete RFC syntax, and for unrecoverable parsing
+    errors.
+
+    The parse method may add additional keys to the dictionary.  In this case
+    the subclass must define an 'init' method, which will be passed the
+    dictionary as its keyword arguments.  The method should use (usually by
+    setting them as the value of similarly named attributes) and remove all the
+    extra keys added by its parse method, and then use super to call its parent
+    class with the remaining arguments and keywords.
+
+    The subclass should also make sure that a 'max_count' attribute is defined
+    that is either None or 1. XXX: need to better define this API.
+
+    """
+
+    def __new__(cls, name, value):
+        kwds = {'defects': []}
+        cls.parse(value, kwds)
+        if utils._has_surrogates(kwds['decoded']):
+            kwds['decoded'] = utils._sanitize(kwds['decoded'])
+        self = str.__new__(cls, kwds['decoded'])
+        del kwds['decoded']
+        self.init(name, **kwds)
+        return self
+
+    def init(self, name, *, parse_tree, defects):
+        self._name = name
+        self._parse_tree = parse_tree
+        self._defects = defects
+
+    @property
+    def name(self):
+        return self._name
+
+    @property
+    def defects(self):
+        return tuple(self._defects)
+
+    def __reduce__(self):
+        return (
+            _reconstruct_header,
+            (
+                self.__class__.__name__,
+                self.__class__.__bases__,
+                str(self),
+            ),
+            self.__dict__)
+
+    @classmethod
+    def _reconstruct(cls, value):
+        return str.__new__(cls, value)
+
+    def fold(self, *, policy):
+        """Fold header according to policy.
+
+        The parsed representation of the header is folded according to
+        RFC5322 rules, as modified by the policy.  If the parse tree
+        contains surrogateescaped bytes, the bytes are CTE encoded using
+        the charset 'unknown-8bit".
+
+        Any non-ASCII characters in the parse tree are CTE encoded using
+        charset utf-8. XXX: make this a policy setting.
+
+        The returned value is an ASCII-only string possibly containing linesep
+        characters, and ending with a linesep character.  The string includes
+        the header name and the ': ' separator.
+
+        """
+        # At some point we need to only put fws here if it was in the source.
+        header = parser.Header([
+            parser.HeaderLabel([
+                parser.ValueTerminal(self.name, 'header-name'),
+                parser.ValueTerminal(':', 'header-sep')]),
+            parser.CFWSList([parser.WhiteSpaceTerminal(' ', 'fws')]),
+                             self._parse_tree])
+        return header.fold(policy=policy)
+
+
+def _reconstruct_header(cls_name, bases, value):
+    return type(cls_name, bases, {})._reconstruct(value)
+
+
+class UnstructuredHeader:
+
+    max_count = None
+    value_parser = staticmethod(parser.get_unstructured)
+
+    @classmethod
+    def parse(cls, value, kwds):
+        kwds['parse_tree'] = cls.value_parser(value)
+        kwds['decoded'] = str(kwds['parse_tree'])
+
+
+class UniqueUnstructuredHeader(UnstructuredHeader):
+
+    max_count = 1
+
+
+class DateHeader:
+
+    """Header whose value consists of a single timestamp.
+
+    Provides an additional attribute, datetime, which is either an aware
+    datetime using a timezone, or a naive datetime if the timezone
+    in the input string is -0000.  Also accepts a datetime as input.
+    The 'value' attribute is the normalized form of the timestamp,
+    which means it is the output of format_datetime on the datetime.
+    """
+
+    max_count = None
+
+    # This is used only for folding, not for creating 'decoded'.
+    value_parser = staticmethod(parser.get_unstructured)
+
+    @classmethod
+    def parse(cls, value, kwds):
+        if not value:
+            kwds['defects'].append(errors.HeaderMissingRequiredValue())
+            kwds['datetime'] = None
+            kwds['decoded'] = ''
+            kwds['parse_tree'] = parser.TokenList()
+            return
+        if isinstance(value, str):
+            value = utils.parsedate_to_datetime(value)
+        kwds['datetime'] = value
+        kwds['decoded'] = utils.format_datetime(kwds['datetime'])
+        kwds['parse_tree'] = cls.value_parser(kwds['decoded'])
+
+    def init(self, *args, **kw):
+        self._datetime = kw.pop('datetime')
+        super().init(*args, **kw)
+
+    @property
+    def datetime(self):
+        return self._datetime
+
+
+class UniqueDateHeader(DateHeader):
+
+    max_count = 1
+
+
+class AddressHeader:
+
+    max_count = None
+
+    @staticmethod
+    def value_parser(value):
+        address_list, value = parser.get_address_list(value)
+        assert not value, 'this should not happen'
+        return address_list
+
+    @classmethod
+    def parse(cls, value, kwds):
+        if isinstance(value, str):
+            # We are translating here from the RFC language (address/mailbox)
+            # to our API language (group/address).
+            kwds['parse_tree'] = address_list = cls.value_parser(value)
+            groups = []
+            for addr in address_list.addresses:
+                groups.append(Group(addr.display_name,
+                                    [Address(mb.display_name or '',
+                                             mb.local_part or '',
+                                             mb.domain or '')
+                                     for mb in addr.all_mailboxes]))
+            defects = list(address_list.all_defects)
+        else:
+            # Assume it is Address/Group stuff
+            if not hasattr(value, '__iter__'):
+                value = [value]
+            groups = [Group(None, [item]) if not hasattr(item, 'addresses')
+                                          else item
+                                    for item in value]
+            defects = []
+        kwds['groups'] = groups
+        kwds['defects'] = defects
+        kwds['decoded'] = ', '.join([str(item) for item in groups])
+        if 'parse_tree' not in kwds:
+            kwds['parse_tree'] = cls.value_parser(kwds['decoded'])
+
+    def init(self, *args, **kw):
+        self._groups = tuple(kw.pop('groups'))
+        self._addresses = None
+        super().init(*args, **kw)
+
+    @property
+    def groups(self):
+        return self._groups
+
+    @property
+    def addresses(self):
+        if self._addresses is None:
+            self._addresses = tuple([address for group in self._groups
+                                             for address in group.addresses])
+        return self._addresses
+
+
+class UniqueAddressHeader(AddressHeader):
+
+    max_count = 1
+
+
+class SingleAddressHeader(AddressHeader):
+
+    @property
+    def address(self):
+        if len(self.addresses)!=1:
+            raise ValueError(("value of single address header {} is not "
+                "a single address").format(self.name))
+        return self.addresses[0]
+
+
+class UniqueSingleAddressHeader(SingleAddressHeader):
+
+    max_count = 1
+
+
+# The header factory #
+
+_default_header_map = {
+    'subject':          UniqueUnstructuredHeader,
+    'date':             UniqueDateHeader,
+    'resent-date':      DateHeader,
+    'orig-date':        UniqueDateHeader,
+    'sender':           UniqueSingleAddressHeader,
+    'resent-sender':    SingleAddressHeader,
+    'to':               UniqueAddressHeader,
+    'resent-to':        AddressHeader,
+    'cc':               UniqueAddressHeader,
+    'resent-cc':        AddressHeader,
+    'bcc':              UniqueAddressHeader,
+    'resent-bcc':       AddressHeader,
+    'from':             UniqueAddressHeader,
+    'resent-from':      AddressHeader,
+    'reply-to':         UniqueAddressHeader,
+    }
+
+class HeaderRegistry:
+
+    """A header_factory and header registry."""
+
+    def __init__(self, base_class=BaseHeader, default_class=UnstructuredHeader,
+                       use_default_map=True):
+        """Create a header_factory that works with the Policy API.
+
+        base_class is the class that will be the last class in the created
+        header class's __bases__ list.  default_class is the class that will be
+        used if "name" (see __call__) does not appear in the registry.
+        use_default_map controls whether or not the default mapping of names to
+        specialized classes is copied in to the registry when the factory is
+        created.  The default is True.
+
+        """
+        self.registry = {}
+        self.base_class = base_class
+        self.default_class = default_class
+        if use_default_map:
+            self.registry.update(_default_header_map)
+
+    def map_to_type(self, name, cls):
+        """Register cls as the specialized class for handling "name" headers.
+
+        """
+        self.registry[name.lower()] = cls
+
+    def __getitem__(self, name):
+        cls = self.registry.get(name.lower(), self.default_class)
+        return type('_'+cls.__name__, (cls, self.base_class), {})
+
+    def __call__(self, name, value):
+        """Create a header instance for header 'name' from 'value'.
+
+        Creates a header instance by creating a specialized class for parsing
+        and representing the specified header by combining the factory
+        base_class with a specialized class from the registry or the
+        default_class, and passing the name and value to the constructed
+        class's constructor.
+
+        """
+        return self[name](name, value)
diff --git a/Lib/email/_policybase.py b/Lib/email/_policybase.py
index 05736d0..6bc298b 100644
--- a/Lib/email/_policybase.py
+++ b/Lib/email/_policybase.py
@@ -64,10 +64,16 @@ class _PolicyBase:
         except for the changes passed in as keyword arguments.
 
         """
+        newpolicy = self.__class__.__new__(self.__class__)
         for attr, value in self.__dict__.items():
-            if attr not in kw:
-                kw[attr] = value
-        return self.__class__(**kw)
+            object.__setattr__(newpolicy, attr, value)
+        for attr, value in kw.items():
+            if not hasattr(self, attr):
+                raise TypeError(
+                    "{!r} is an invalid keyword argument for {}".format(
+                        attr, self.__class__.__name__))
+            object.__setattr__(newpolicy, attr, value)
+        return newpolicy
 
     def __setattr__(self, name, value):
         if hasattr(self, name):
diff --git a/Lib/email/errors.py b/Lib/email/errors.py
index c04deb4..f916229 100644
--- a/Lib/email/errors.py
+++ b/Lib/email/errors.py
@@ -5,7 +5,6 @@
 """email package exception classes."""
 
 
-
 class MessageError(Exception):
     """Base class for errors in the email package."""
 
@@ -30,9 +29,8 @@ class CharsetError(MessageError):
     """An illegal charset was given."""
 
 
-
 # These are parsing defects which the parser was able to work around.
-class MessageDefect(Exception):
+class MessageDefect(ValueError):
     """Base class for a message defect."""
 
     def __init__(self, line=None):
@@ -58,3 +56,42 @@ class MultipartInvariantViolationDefect(MessageDefect):
 
 class InvalidMultipartContentTransferEncodingDefect(MessageDefect):
     """An invalid content transfer encoding was set on the multipart itself."""
+
+class UndecodableBytesDefect(MessageDefect):
+    """Header contained bytes that could not be decoded"""
+
+class InvalidBase64PaddingDefect(MessageDefect):
+    """base64 encoded sequence had an incorrect length"""
+
+class InvalidBase64CharactersDefect(MessageDefect):
+    """base64 encoded sequence had characters not in base64 alphabet"""
+
+# These errors are specific to header parsing.
+
+class HeaderDefect(MessageDefect):
+    """Base class for a header defect."""
+
+class InvalidHeaderDefect(HeaderDefect):
+    """Header is not valid, message gives details."""
+
+class HeaderMissingRequiredValue(HeaderDefect):
+    """A header that must have a value had none"""
+
+class NonPrintableDefect(HeaderDefect):
+    """ASCII characters outside the ascii-printable range found"""
+
+    def __init__(self, non_printables):
+        super().__init__(non_printables)
+        self.non_printables = non_printables
+
+    def __str__(self):
+        return ("the following ASCII non-printables found in header: "
+            "{}".format(self.non_printables))
+
+class ObsoleteHeaderDefect(HeaderDefect):
+    """Header uses syntax declared obsolete by RFC 5322"""
+
+class NonASCIILocalPartDefect(HeaderDefect):
+    """local_part contains non-ASCII characters"""
+    # This defect only occurs during unicode parsing, not when
+    # parsing messages decoded from binary.
diff --git a/Lib/email/generator.py b/Lib/email/generator.py
index bfa288b..fcecf93 100644
--- a/Lib/email/generator.py
+++ b/Lib/email/generator.py
@@ -95,9 +95,15 @@ class Generator:
         self._encoded_NL = self._encode(self._NL)
         self._EMPTY = ''
         self._encoded_EMTPY = self._encode('')
-        p = self.policy
+        # Because we use clone (below) when we recursively process message
+        # subparts, and because clone uses the computed policy (not None),
+        # submessages will automatically get set to the computed policy when
+        # they are processed by this code.
+        old_gen_policy = self.policy
+        old_msg_policy = msg.policy
         try:
             self.policy = policy
+            msg.policy = policy
             if unixfrom:
                 ufrom = msg.get_unixfrom()
                 if not ufrom:
@@ -105,7 +111,8 @@ class Generator:
                 self.write(ufrom + self._NL)
             self._write(msg)
         finally:
-            self.policy = p
+            self.policy = old_gen_policy
+            msg.policy = old_msg_policy
 
     def clone(self, fp):
         """Clone this generator with the exact same options."""
diff --git a/Lib/email/policy.py b/Lib/email/policy.py
index dae2dc7..ea90a8f 100644
--- a/Lib/email/policy.py
+++ b/Lib/email/policy.py
@@ -2,11 +2,178 @@
 code that adds all the email6 features.
 """
 
-from email._policybase import Policy, compat32, Compat32
+from email._policybase import Policy, Compat32, compat32
+from email.utils import _has_surrogates
+from email._headerregistry import HeaderRegistry as _HeaderRegistry
 
-# XXX: temporarily derive everything from compat32.
+__all__ = [
+    'Compat32',
+    'compat32',
+    'Policy',
+    'EmailPolicy',
+    'default',
+    'strict',
+    'SMTP',
+    'HTTP',
+    ]
 
-default = compat32
+class EmailPolicy(Policy):
+
+    """+
+    PROVISIONAL
+
+    The API extensions enabled by this this policy are currently provisional.
+    Refer to the documentation for details.
+
+    This policy adds new header parsing and folding algorithms.  Instead of
+    simple strings, headers are custom objects with custom attributes
+    depending on the type of the field.  The folding algorithm fully
+    implements RFCs 2047 and 5322.
+
+    In addition to the settable attributes listed above that apply to
+    all Policies, this policy adds the following additional attributes:
+
+    refold_source       -- if the value for a header in the Message object
+                           came from the parsing of some source, this attribute
+                           indicates whether or not a generator should refold
+                           that value when transforming the message back into
+                           stream form.  The possible values are:
+
+                           none  -- all source values use original folding
+                           long  -- source values that have any line that is
+                                    longer than max_line_length will be
+                                    refolded
+                           all  -- all values are refolded.
+
+                           The default is 'long'.
+
+    header_factory      -- a callable that takes two arguments, 'name' and
+                           'value', where 'name' is a header field name and
+                           'value' is an unfolded header field value, and
+                           returns a string-like object that represents that
+                           header.  A default header_factory is provided that
+                           understands some of the RFC5322 header field types.
+                           (Currently address fields and date fields have
+                           special treatment, while all other fields are
+                           treated as unstructured.  This list will be
+                           completed before the extension is marked stable.)
+    """
+
+    refold_source = 'long'
+    header_factory = _HeaderRegistry()
+
+    def __init__(self, **kw):
+        # Ensure that each new instance gets a unique header factory
+        # (as opposed to clones, which share the factory).
+        if 'header_factory' not in kw:
+            object.__setattr__(self, 'header_factory', _HeaderRegistry())
+        super().__init__(**kw)
+
+    # The logic of the next three methods is chosen such that it is possible to
+    # switch a Message object between a Compat32 policy and a policy derived
+    # from this class and have the results stay consistent.  This allows a
+    # Message object constructed with this policy to be passed to a library
+    # that only handles Compat32 objects, or to receive such an object and
+    # convert it to use the newer style by just changing its policy.  It is
+    # also chosen because it postpones the relatively expensive full rfc5322
+    # parse until as late as possible when parsing from source, since in many
+    # applications only a few headers will actually be inspected.
+
+    def header_source_parse(self, sourcelines):
+        """+
+        The name is parsed as everything up to the ':' and returned unmodified.
+        The value is determined by stripping leading whitespace off the
+        remainder of the first line, joining all subsequent lines together, and
+        stripping any trailing carriage return or linefeed characters.  (This
+        is the same as Compat32).
+
+        """
+        name, value = sourcelines[0].split(':', 1)
+        value = value.lstrip(' \t') + ''.join(sourcelines[1:])
+        return (name, value.rstrip('\r\n'))
+
+    def header_store_parse(self, name, value):
+        """+
+        The name is returned unchanged.  If the input value has a 'name'
+        attribute and it matches the name ignoring case, the value is returned
+        unchanged.  Otherwise the name and value are passed to header_factory
+        method, and the resulting custom header object is returned as the
+        value.  In this case a ValueError is raised if the input value contains
+        CR or LF characters.
+
+        """
+        if hasattr(value, 'name') and value.name.lower() == name.lower():
+            return (name, value)
+        if len(value.splitlines())>1:
+            raise ValueError("Header values may not contain linefeed "
+                             "or carriage return characters")
+        return (name, self.header_factory(name, value))
+
+    def header_fetch_parse(self, name, value):
+        """+
+        If the value has a 'name' attribute, it is returned to unmodified.
+        Otherwise the name and the value with any linesep characters removed
+        are passed to the header_factory method, and the resulting custom
+        header object is returned.  Any surrogateescaped bytes get turned
+        into the unicode unknown-character glyph.
+
+        """
+        if hasattr(value, 'name'):
+            return value
+        return self.header_factory(name, ''.join(value.splitlines()))
+
+    def fold(self, name, value):
+        """+
+        Header folding is controlled by the refold_source policy setting.  A
+        value is considered to be a 'source value' if and only if it does not
+        have a 'name' attribute (having a 'name' attribute means it is a header
+        object of some sort).  If a source value needs to be refolded according
+        to the policy, it is converted into a custom header object by passing
+        the name and the value with any linesep characters removed to the
+        header_factory method.  Folding of a custom header object is done by
+        calling its fold method with the current policy.
+
+        Source values are split into lines using splitlines.  If the value is
+        not to be refolded, the lines are rejoined using the linesep from the
+        policy and returned.  The exception is lines containing non-ascii
+        binary data.  In that case the value is refolded regardless of the
+        refold_source setting, which causes the binary data to be CTE encoded
+        using the unknown-8bit charset.
+
+        """
+        return self._fold(name, value, refold_binary=True)
+
+    def fold_binary(self, name, value):
+        """+
+        The same as fold if cte_type is 7bit, except that the returned value is
+        bytes.
+
+        If cte_type is 8bit, non-ASCII binary data is converted back into
+        bytes.  Headers with binary data are not refolded, regardless of the
+        refold_header setting, since there is no way to know whether the binary
+        data consists of single byte characters or multibyte characters.
+
+        """
+        folded = self._fold(name, value, refold_binary=self.cte_type=='7bit')
+        return folded.encode('ascii', 'surrogateescape')
+
+    def _fold(self, name, value, refold_binary=False):
+        if hasattr(value, 'name'):
+            return value.fold(policy=self)
+        maxlen = self.max_line_length if self.max_line_length else float('inf')
+        lines = value.splitlines()
+        refold = (self.refold_source == 'all' or
+                  self.refold_source == 'long' and
+                    (len(lines[0])+len(name)+2 > maxlen or
+                     any(len(x) > maxlen for x in lines[1:])))
+        if refold or refold_binary and _has_surrogates(value):
+            return self.header_factory(name, ''.join(lines)).fold(policy=self)
+        return name + ': ' + self.linesep.join(lines) + self.linesep
+
+
+default = EmailPolicy()
+# Make the default policy use the class default header_factory
+del default.header_factory
 strict = default.clone(raise_on_defect=True)
 SMTP = default.clone(linesep='\r\n')
 HTTP = default.clone(linesep='\r\n', max_line_length=None)
diff --git a/Lib/email/utils.py b/Lib/email/utils.py
index b82d5c5..b7e1bb9 100644
--- a/Lib/email/utils.py
+++ b/Lib/email/utils.py
@@ -62,6 +62,13 @@ escapesre = re.compile(r'[\\"]')
 _has_surrogates = re.compile(
     '([^\ud800-\udbff]|\A)[\udc00-\udfff]([^\udc00-\udfff]|\Z)').search
 
+# How to deal with a string containing bytes before handing it to the
+# application through the 'normal' interface.
+def _sanitize(string):
+    # Turn any escaped bytes into unicode 'unknown' char.
+    original_bytes = string.encode('ascii', 'surrogateescape')
+    return original_bytes.decode('ascii', 'replace')
+
 
 # Helpers