# -*- coding: utf-8 -*-
"""
    werkzeug.urls
    ~~~~~~~~~~~~~

    ``werkzeug.urls`` used to provide several wrapper functions for Python 2
    urlparse, whose main purpose were to work around the behavior of the Py2
    stdlib and its lack of unicode support. While this was already a somewhat
    inconvenient situation, it got even more complicated because Python 3's
    ``urllib.parse`` actually does handle unicode properly. In other words,
    this module would wrap two libraries with completely different behavior. So
    now this module contains a 2-and-3-compatible backport of Python 3's
    ``urllib.parse``, which is mostly API-compatible.

    :copyright: (c) 2014 by the Werkzeug Team, see AUTHORS for more details.
    :license: BSD, see LICENSE for more details.
"""
import os
import re
from werkzeug._compat import text_type, PY2, to_unicode, \
    to_native, implements_to_string, try_coerce_native, \
    normalize_string_tuple, make_literal_wrapper, \
    fix_tuple_repr
from werkzeug._internal import _encode_idna, _decode_idna
from werkzeug.datastructures import MultiDict, iter_multi_items
from collections import namedtuple


# A regular expression for what a valid schema looks like
_scheme_re = re.compile(r'^[a-zA-Z0-9+-.]+$')

# Characters that are safe in any part of an URL.
_always_safe = (b'abcdefghijklmnopqrstuvwxyz'
                b'ABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789_.-+')

_hexdigits = '0123456789ABCDEFabcdef'
_hextobyte = dict(
    ((a + b).encode(), int(a + b, 16))
    for a in _hexdigits for b in _hexdigits
)
_bytetohex = [
    ('%%%02X' % char).encode('ascii') for char in range(256)
]


_URLTuple = fix_tuple_repr(namedtuple(
    '_URLTuple',
    ['scheme', 'netloc', 'path', 'query', 'fragment']
))


class BaseURL(_URLTuple):

    '''Superclass of :py:class:`URL` and :py:class:`BytesURL`.'''
    __slots__ = ()

    def replace(self, **kwargs):
        """Return an URL with the same values, except for those parameters
        given new values by whichever keyword arguments are specified."""
        return self._replace(**kwargs)

    @property
    def host(self):
        """The host part of the URL if available, otherwise `None`.  The
        host is either the hostname or the IP address mentioned in the
        URL.  It will not contain the port.
        """
        return self._split_host()[0]

    @property
    def ascii_host(self):
        """Works exactly like :attr:`host` but will return a result that
        is restricted to ASCII.  If it finds a netloc that is not ASCII
        it will attempt to idna decode it.  This is useful for socket
        operations when the URL might include internationalized characters.
        """
        rv = self.host
        if rv is not None and isinstance(rv, text_type):
            try:
                rv = _encode_idna(rv)
            except UnicodeError:
                rv = rv.encode('ascii', 'ignore')
        return to_native(rv, 'ascii', 'ignore')

    @property
    def port(self):
        """The port in the URL as an integer if it was present, `None`
        otherwise.  This does not fill in default ports.
        """
        try:
            rv = int(to_native(self._split_host()[1]))
            if 0 <= rv <= 65535:
                return rv
        except (ValueError, TypeError):
            pass

    @property
    def auth(self):
        """The authentication part in the URL if available, `None`
        otherwise.
        """
        return self._split_netloc()[0]

    @property
    def username(self):
        """The username if it was part of the URL, `None` otherwise.
        This undergoes URL decoding and will always be a unicode string.
        """
        rv = self._split_auth()[0]
        if rv is not None:
            return _url_unquote_legacy(rv)

    @property
    def raw_username(self):
        """The username if it was part of the URL, `None` otherwise.
        Unlike :attr:`username` this one is not being decoded.
        """
        return self._split_auth()[0]

    @property
    def password(self):
        """The password if it was part of the URL, `None` otherwise.
        This undergoes URL decoding and will always be a unicode string.
        """
        rv = self._split_auth()[1]
        if rv is not None:
            return _url_unquote_legacy(rv)

    @property
    def raw_password(self):
        """The password if it was part of the URL, `None` otherwise.
        Unlike :attr:`password` this one is not being decoded.
        """
        return self._split_auth()[1]

    def decode_query(self, *args, **kwargs):
        """Decodes the query part of the URL.  Ths is a shortcut for
        calling :func:`url_decode` on the query argument.  The arguments and
        keyword arguments are forwarded to :func:`url_decode` unchanged.
        """
        return url_decode(self.query, *args, **kwargs)

    def join(self, *args, **kwargs):
        """Joins this URL with another one.  This is just a convenience
        function for calling into :meth:`url_join` and then parsing the
        return value again.
        """
        return url_parse(url_join(self, *args, **kwargs))

    def to_url(self):
        """Returns a URL string or bytes depending on the type of the
        information stored.  This is just a convenience function
        for calling :meth:`url_unparse` for this URL.
        """
        return url_unparse(self)

    def decode_netloc(self):
        """Decodes the netloc part into a string."""
        rv = _decode_idna(self.host or '')

        if ':' in rv:
            rv = '[%s]' % rv
        port = self.port
        if port is not None:
            rv = '%s:%d' % (rv, port)
        auth = ':'.join(filter(None, [
            _url_unquote_legacy(self.raw_username or '', '/:%@'),
            _url_unquote_legacy(self.raw_password or '', '/:%@'),
        ]))
        if auth:
            rv = '%s@%s' % (auth, rv)
        return rv

    def to_uri_tuple(self):
        """Returns a :class:`BytesURL` tuple that holds a URI.  This will
        encode all the information in the URL properly to ASCII using the
        rules a web browser would follow.

        It's usually more interesting to directly call :meth:`iri_to_uri` which
        will return a string.
        """
        return url_parse(iri_to_uri(self).encode('ascii'))

    def to_iri_tuple(self):
        """Returns a :class:`URL` tuple that holds a IRI.  This will try
        to decode as much information as possible in the URL without
        losing information similar to how a web browser does it for the
        URL bar.

        It's usually more interesting to directly call :meth:`uri_to_iri` which
        will return a string.
        """
        return url_parse(uri_to_iri(self))

    def get_file_location(self, pathformat=None):
        """Returns a tuple with the location of the file in the form
        ``(server, location)``.  If the netloc is empty in the URL or
        points to localhost, it's represented as ``None``.

        The `pathformat` by default is autodetection but needs to be set
        when working with URLs of a specific system.  The supported values
        are ``'windows'`` when working with Windows or DOS paths and
        ``'posix'`` when working with posix paths.

        If the URL does not point to to a local file, the server and location
        are both represented as ``None``.

        :param pathformat: The expected format of the path component.
                           Currently ``'windows'`` and ``'posix'`` are
                           supported.  Defaults to ``None`` which is
                           autodetect.
        """
        if self.scheme != 'file':
            return None, None

        path = url_unquote(self.path)
        host = self.netloc or None

        if pathformat is None:
            if os.name == 'nt':
                pathformat = 'windows'
            else:
                pathformat = 'posix'

        if pathformat == 'windows':
            if path[:1] == '/' and path[1:2].isalpha() and path[2:3] in '|:':
                path = path[1:2] + ':' + path[3:]
            windows_share = path[:3] in ('\\' * 3, '/' * 3)
            import ntpath
            path = ntpath.normpath(path)
            # Windows shared drives are represented as ``\\host\\directory``.
            # That results in a URL like ``file://///host/directory``, and a
            # path like ``///host/directory``. We need to special-case this
            # because the path contains the hostname.
            if windows_share and host is None:
                parts = path.lstrip('\\').split('\\', 1)
                if len(parts) == 2:
                    host, path = parts
                else:
                    host = parts[0]
                    path = ''
        elif pathformat == 'posix':
            import posixpath
            path = posixpath.normpath(path)
        else:
            raise TypeError('Invalid path format %s' % repr(pathformat))

        if host in ('127.0.0.1', '::1', 'localhost'):
            host = None

        return host, path

    def _split_netloc(self):
        if self._at in self.netloc:
            return self.netloc.split(self._at, 1)
        return None, self.netloc

    def _split_auth(self):
        auth = self._split_netloc()[0]
        if not auth:
            return None, None
        if self._colon not in auth:
            return auth, None
        return auth.split(self._colon, 1)

    def _split_host(self):
        rv = self._split_netloc()[1]
        if not rv:
            return None, None

        if not rv.startswith(self._lbracket):
            if self._colon in rv:
                return rv.split(self._colon, 1)
            return rv, None

        idx = rv.find(self._rbracket)
        if idx < 0:
            return rv, None

        host = rv[1:idx]
        rest = rv[idx + 1:]
        if rest.startswith(self._colon):
            return host, rest[1:]
        return host, None


@implements_to_string
class URL(BaseURL):

    """Represents a parsed URL.  This behaves like a regular tuple but
    also has some extra attributes that give further insight into the
    URL.
    """
    __slots__ = ()
    _at = '@'
    _colon = ':'
    _lbracket = '['
    _rbracket = ']'

    def __str__(self):
        return self.to_url()

    def encode_netloc(self):
        """Encodes the netloc part to an ASCII safe URL as bytes."""
        rv = self.ascii_host or ''
        if ':' in rv:
            rv = '[%s]' % rv
        port = self.port
        if port is not None:
            rv = '%s:%d' % (rv, port)
        auth = ':'.join(filter(None, [
            url_quote(self.raw_username or '', 'utf-8', 'strict', '/:%'),
            url_quote(self.raw_password or '', 'utf-8', 'strict', '/:%'),
        ]))
        if auth:
            rv = '%s@%s' % (auth, rv)
        return to_native(rv)

    def encode(self, charset='utf-8', errors='replace'):
        """Encodes the URL to a tuple made out of bytes.  The charset is
        only being used for the path, query and fragment.
        """
        return BytesURL(
            self.scheme.encode('ascii'),
            self.encode_netloc(),
            self.path.encode(charset, errors),
            self.query.encode(charset, errors),
            self.fragment.encode(charset, errors)
        )


class BytesURL(BaseURL):

    """Represents a parsed URL in bytes."""
    __slots__ = ()
    _at = b'@'
    _colon = b':'
    _lbracket = b'['
    _rbracket = b']'

    def __str__(self):
        return self.to_url().decode('utf-8', 'replace')

    def encode_netloc(self):
        """Returns the netloc unchanged as bytes."""
        return self.netloc

    def decode(self, charset='utf-8', errors='replace'):
        """Decodes the URL to a tuple made out of strings.  The charset is
        only being used for the path, query and fragment.
        """
        return URL(
            self.scheme.decode('ascii'),
            self.decode_netloc(),
            self.path.decode(charset, errors),
            self.query.decode(charset, errors),
            self.fragment.decode(charset, errors)
        )


def _unquote_to_bytes(string, unsafe=''):
    if isinstance(string, text_type):
        string = string.encode('utf-8')
    if isinstance(unsafe, text_type):
        unsafe = unsafe.encode('utf-8')
    unsafe = frozenset(bytearray(unsafe))
    bits = iter(string.split(b'%'))
    result = bytearray(next(bits, b''))
    for item in bits:
        try:
            char = _hextobyte[item[:2]]
            if char in unsafe:
                raise KeyError()
            result.append(char)
            result.extend(item[2:])
        except KeyError:
            result.extend(b'%')
            result.extend(item)
    return bytes(result)


def _url_encode_impl(obj, charset, encode_keys, sort, key):
    iterable = iter_multi_items(obj)
    if sort:
        iterable = sorted(iterable, key=key)
    for key, value in iterable:
        if value is None:
            continue
        if not isinstance(key, bytes):
            key = text_type(key).encode(charset)
        if not isinstance(value, bytes):
            value = text_type(value).encode(charset)
        yield url_quote_plus(key) + '=' + url_quote_plus(value)


def _url_unquote_legacy(value, unsafe=''):
    try:
        return url_unquote(value, charset='utf-8',
                           errors='strict', unsafe=unsafe)
    except UnicodeError:
        return url_unquote(value, charset='latin1', unsafe=unsafe)


def url_parse(url, scheme=None, allow_fragments=True):
    """Parses a URL from a string into a :class:`URL` tuple.  If the URL
    is lacking a scheme it can be provided as second argument. Otherwise,
    it is ignored.  Optionally fragments can be stripped from the URL
    by setting `allow_fragments` to `False`.

    The inverse of this function is :func:`url_unparse`.

    :param url: the URL to parse.
    :param scheme: the default schema to use if the URL is schemaless.
    :param allow_fragments: if set to `False` a fragment will be removed
                            from the URL.
    """
    s = make_literal_wrapper(url)
    is_text_based = isinstance(url, text_type)

    if scheme is None:
        scheme = s('')
    netloc = query = fragment = s('')
    i = url.find(s(':'))
    if i > 0 and _scheme_re.match(to_native(url[:i], errors='replace')):
        # make sure "iri" is not actually a port number (in which case
        # "scheme" is really part of the path)
        rest = url[i + 1:]
        if not rest or any(c not in s('0123456789') for c in rest):
            # not a port number
            scheme, url = url[:i].lower(), rest

    if url[:2] == s('//'):
        delim = len(url)
        for c in s('/?#'):
            wdelim = url.find(c, 2)
            if wdelim >= 0:
                delim = min(delim, wdelim)
        netloc, url = url[2:delim], url[delim:]
        if (s('[') in netloc and s(']') not in netloc) or \
           (s(']') in netloc and s('[') not in netloc):
            raise ValueError('Invalid IPv6 URL')

    if allow_fragments and s('#') in url:
        url, fragment = url.split(s('#'), 1)
    if s('?') in url:
        url, query = url.split(s('?'), 1)

    result_type = is_text_based and URL or BytesURL
    return result_type(scheme, netloc, url, query, fragment)


def url_quote(string, charset='utf-8', errors='strict', safe='/:', unsafe=''):
    """URL encode a single string with a given encoding.

    :param s: the string to quote.
    :param charset: the charset to be used.
    :param safe: an optional sequence of safe characters.
    :param unsafe: an optional sequence of unsafe characters.

    .. versionadded:: 0.9.2
       The `unsafe` parameter was added.
    """
    if not isinstance(string, (text_type, bytes, bytearray)):
        string = text_type(string)
    if isinstance(string, text_type):
        string = string.encode(charset, errors)
    if isinstance(safe, text_type):
        safe = safe.encode(charset, errors)
    if isinstance(unsafe, text_type):
        unsafe = unsafe.encode(charset, errors)
    safe = frozenset(bytearray(safe) + _always_safe) - frozenset(bytearray(unsafe))
    rv = bytearray()
    for char in bytearray(string):
        if char in safe:
            rv.append(char)
        else:
            rv.extend(_bytetohex[char])
    return to_native(bytes(rv))


def url_quote_plus(string, charset='utf-8', errors='strict', safe=''):
    """URL encode a single string with the given encoding and convert
    whitespace to "+".

    :param s: The string to quote.
    :param charset: The charset to be used.
    :param safe: An optional sequence of safe characters.
    """
    return url_quote(string, charset, errors, safe + ' ', '+').replace(' ', '+')


def url_unparse(components):
    """The reverse operation to :meth:`url_parse`.  This accepts arbitrary
    as well as :class:`URL` tuples and returns a URL as a string.

    :param components: the parsed URL as tuple which should be converted
                       into a URL string.
    """
    scheme, netloc, path, query, fragment = \
        normalize_string_tuple(components)
    s = make_literal_wrapper(scheme)
    url = s('')

    # We generally treat file:///x and file:/x the same which is also
    # what browsers seem to do.  This also allows us to ignore a schema
    # register for netloc utilization or having to differenciate between
    # empty and missing netloc.
    if netloc or (scheme and path.startswith(s('/'))):
        if path and path[:1] != s('/'):
            path = s('/') + path
        url = s('//') + (netloc or s('')) + path
    elif path:
        url += path
    if scheme:
        url = scheme + s(':') + url
    if query:
        url = url + s('?') + query
    if fragment:
        url = url + s('#') + fragment
    return url


def url_unquote(string, charset='utf-8', errors='replace', unsafe=''):
    """URL decode a single string with a given encoding.  If the charset
    is set to `None` no unicode decoding is performed and raw bytes
    are returned.

    :param s: the string to unquote.
    :param charset: the charset of the query string.  If set to `None`
                    no unicode decoding will take place.
    :param errors: the error handling for the charset decoding.
    """
    rv = _unquote_to_bytes(string, unsafe)
    if charset is not None:
        rv = rv.decode(charset, errors)
    return rv


def url_unquote_plus(s, charset='utf-8', errors='replace'):
    """URL decode a single string with the given `charset` and decode "+" to
    whitespace.

    Per default encoding errors are ignored.  If you want a different behavior
    you can set `errors` to ``'replace'`` or ``'strict'``.  In strict mode a
    :exc:`HTTPUnicodeError` is raised.

    :param s: The string to unquote.
    :param charset: the charset of the query string.  If set to `None`
                    no unicode decoding will take place.
    :param errors: The error handling for the `charset` decoding.
    """
    if isinstance(s, text_type):
        s = s.replace(u'+', u' ')
    else:
        s = s.replace(b'+', b' ')
    return url_unquote(s, charset, errors)


def url_fix(s, charset='utf-8'):
    r"""Sometimes you get an URL by a user that just isn't a real URL because
    it contains unsafe characters like ' ' and so on. This function can fix
    some of the problems in a similar way browsers handle data entered by the
    user:

    >>> url_fix(u'http://de.wikipedia.org/wiki/Elf (Begriffskl\xe4rung)')
    'http://de.wikipedia.org/wiki/Elf%20(Begriffskl%C3%A4rung)'

    :param s: the string with the URL to fix.
    :param charset: The target charset for the URL if the url was given as
                    unicode string.
    """
    # First step is to switch to unicode processing and to convert
    # backslashes (which are invalid in URLs anyways) to slashes.  This is
    # consistent with what Chrome does.
    s = to_unicode(s, charset, 'replace').replace('\\', '/')

    # For the specific case that we look like a malformed windows URL
    # we want to fix this up manually:
    if s.startswith('file://') and s[7:8].isalpha() and s[8:10] in (':/', '|/'):
        s = 'file:///' + s[7:]

    url = url_parse(s)
    path = url_quote(url.path, charset, safe='/%+$!*\'(),')
    qs = url_quote_plus(url.query, charset, safe=':&%=+$!*\'(),')
    anchor = url_quote_plus(url.fragment, charset, safe=':&%=+$!*\'(),')
    return to_native(url_unparse((url.scheme, url.encode_netloc(),
                                  path, qs, anchor)))


def uri_to_iri(uri, charset='utf-8', errors='replace'):
    r"""
    Converts a URI in a given charset to a IRI.

    Examples for URI versus IRI:

    >>> uri_to_iri(b'http://xn--n3h.net/')
    u'http://\u2603.net/'
    >>> uri_to_iri(b'http://%C3%BCser:p%C3%A4ssword@xn--n3h.net/p%C3%A5th')
    u'http://\xfcser:p\xe4ssword@\u2603.net/p\xe5th'

    Query strings are left unchanged:

    >>> uri_to_iri('/?foo=24&x=%26%2f')
    u'/?foo=24&x=%26%2f'

    .. versionadded:: 0.6

    :param uri: The URI to convert.
    :param charset: The charset of the URI.
    :param errors: The error handling on decode.
    """
    if isinstance(uri, tuple):
        uri = url_unparse(uri)
    uri = url_parse(to_unicode(uri, charset))
    path = url_unquote(uri.path, charset, errors, '%/;?')
    query = url_unquote(uri.query, charset, errors, '%;/?:@&=+,$#')
    fragment = url_unquote(uri.fragment, charset, errors, '%;/?:@&=+,$#')
    return url_unparse((uri.scheme, uri.decode_netloc(),
                        path, query, fragment))


def iri_to_uri(iri, charset='utf-8', errors='strict', safe_conversion=False):
    r"""
    Converts any unicode based IRI to an acceptable ASCII URI. Werkzeug always
    uses utf-8 URLs internally because this is what browsers and HTTP do as
    well. In some places where it accepts an URL it also accepts a unicode IRI
    and converts it into a URI.

    Examples for IRI versus URI:

    >>> iri_to_uri(u'http://☃.net/')
    'http://xn--n3h.net/'
    >>> iri_to_uri(u'http://üser:pässword@☃.net/påth')
    'http://%C3%BCser:p%C3%A4ssword@xn--n3h.net/p%C3%A5th'

    There is a general problem with IRI and URI conversion with some
    protocols that appear in the wild that are in violation of the URI
    specification.  In places where Werkzeug goes through a forced IRI to
    URI conversion it will set the `safe_conversion` flag which will
    not perform a conversion if the end result is already ASCII.  This
    can mean that the return value is not an entirely correct URI but
    it will not destroy such invalid URLs in the process.

    As an example consider the following two IRIs::

      magnet:?xt=uri:whatever
      itms-services://?action=download-manifest

    The internal representation after parsing of those URLs is the same
    and there is no way to reconstruct the original one.  If safe
    conversion is enabled however this function becomes a noop for both of
    those strings as they both can be considered URIs.

    .. versionadded:: 0.6

    .. versionchanged:: 0.9.6
       The `safe_conversion` parameter was added.

    :param iri: The IRI to convert.
    :param charset: The charset for the URI.
    :param safe_conversion: indicates if a safe conversion should take place.
                            For more information see the explanation above.
    """
    if isinstance(iri, tuple):
        iri = url_unparse(iri)

    if safe_conversion:
        try:
            native_iri = to_native(iri)
            ascii_iri = to_native(iri).encode('ascii')
            if ascii_iri.split() == [ascii_iri]:
                return native_iri
        except UnicodeError:
            pass

    iri = url_parse(to_unicode(iri, charset, errors))

    netloc = iri.encode_netloc()
    path = url_quote(iri.path, charset, errors, '/:~+%')
    query = url_quote(iri.query, charset, errors, '%&[]:;$*()+,!?*/=')
    fragment = url_quote(iri.fragment, charset, errors, '=%&[]:;$()+,!?*/')

    return to_native(url_unparse((iri.scheme, netloc,
                                  path, query, fragment)))


def url_decode(s, charset='utf-8', decode_keys=False, include_empty=True,
               errors='replace', separator='&', cls=None):
    """
    Parse a querystring and return it as :class:`MultiDict`.  There is a
    difference in key decoding on different Python versions.  On Python 3
    keys will always be fully decoded whereas on Python 2, keys will
    remain bytestrings if they fit into ASCII.  On 2.x keys can be forced
    to be unicode by setting `decode_keys` to `True`.

    If the charset is set to `None` no unicode decoding will happen and
    raw bytes will be returned.

    Per default a missing value for a key will default to an empty key.  If
    you don't want that behavior you can set `include_empty` to `False`.

    Per default encoding errors are ignored.  If you want a different behavior
    you can set `errors` to ``'replace'`` or ``'strict'``.  In strict mode a
    `HTTPUnicodeError` is raised.

    .. versionchanged:: 0.5
       In previous versions ";" and "&" could be used for url decoding.
       This changed in 0.5 where only "&" is supported.  If you want to
       use ";" instead a different `separator` can be provided.

       The `cls` parameter was added.

    :param s: a string with the query string to decode.
    :param charset: the charset of the query string.  If set to `None`
                    no unicode decoding will take place.
    :param decode_keys: Used on Python 2.x to control whether keys should
                        be forced to be unicode objects.  If set to `True`
                        then keys will be unicode in all cases. Otherwise,
                        they remain `str` if they fit into ASCII.
    :param include_empty: Set to `False` if you don't want empty values to
                          appear in the dict.
    :param errors: the decoding error behavior.
    :param separator: the pair separator to be used, defaults to ``&``
    :param cls: an optional dict class to use.  If this is not specified
                       or `None` the default :class:`MultiDict` is used.
    """
    if cls is None:
        cls = MultiDict
    if isinstance(s, text_type) and not isinstance(separator, text_type):
        separator = separator.decode(charset or 'ascii')
    elif isinstance(s, bytes) and not isinstance(separator, bytes):
        separator = separator.encode(charset or 'ascii')
    return cls(_url_decode_impl(s.split(separator), charset, decode_keys,
                                include_empty, errors))


def url_decode_stream(stream, charset='utf-8', decode_keys=False,
                      include_empty=True, errors='replace', separator='&',
                      cls=None, limit=None, return_iterator=False):
    """Works like :func:`url_decode` but decodes a stream.  The behavior
    of stream and limit follows functions like
    :func:`~werkzeug.wsgi.make_line_iter`.  The generator of pairs is
    directly fed to the `cls` so you can consume the data while it's
    parsed.

    .. versionadded:: 0.8

    :param stream: a stream with the encoded querystring
    :param charset: the charset of the query string.  If set to `None`
                    no unicode decoding will take place.
    :param decode_keys: Used on Python 2.x to control whether keys should
                        be forced to be unicode objects.  If set to `True`,
                        keys will be unicode in all cases. Otherwise, they
                        remain `str` if they fit into ASCII.
    :param include_empty: Set to `False` if you don't want empty values to
                          appear in the dict.
    :param errors: the decoding error behavior.
    :param separator: the pair separator to be used, defaults to ``&``
    :param cls: an optional dict class to use.  If this is not specified
                       or `None` the default :class:`MultiDict` is used.
    :param limit: the content length of the URL data.  Not necessary if
                  a limited stream is provided.
    :param return_iterator: if set to `True` the `cls` argument is ignored
                            and an iterator over all decoded pairs is
                            returned
    """
    from werkzeug.wsgi import make_chunk_iter
    if return_iterator:
        cls = lambda x: x
    elif cls is None:
        cls = MultiDict
    pair_iter = make_chunk_iter(stream, separator, limit)
    return cls(_url_decode_impl(pair_iter, charset, decode_keys,
                                include_empty, errors))


def _url_decode_impl(pair_iter, charset, decode_keys, include_empty, errors):
    for pair in pair_iter:
        if not pair:
            continue
        s = make_literal_wrapper(pair)
        equal = s('=')
        if equal in pair:
            key, value = pair.split(equal, 1)
        else:
            if not include_empty:
                continue
            key = pair
            value = s('')
        key = url_unquote_plus(key, charset, errors)
        if charset is not None and PY2 and not decode_keys:
            key = try_coerce_native(key)
        yield key, url_unquote_plus(value, charset, errors)


def url_encode(obj, charset='utf-8', encode_keys=False, sort=False, key=None,
               separator=b'&'):
    """URL encode a dict/`MultiDict`.  If a value is `None` it will not appear
    in the result string.  Per default only values are encoded into the target
    charset strings.  If `encode_keys` is set to ``True`` unicode keys are
    supported too.

    If `sort` is set to `True` the items are sorted by `key` or the default
    sorting algorithm.

    .. versionadded:: 0.5
        `sort`, `key`, and `separator` were added.

    :param obj: the object to encode into a query string.
    :param charset: the charset of the query string.
    :param encode_keys: set to `True` if you have unicode keys. (Ignored on
                        Python 3.x)
    :param sort: set to `True` if you want parameters to be sorted by `key`.
    :param separator: the separator to be used for the pairs.
    :param key: an optional function to be used for sorting.  For more details
                check out the :func:`sorted` documentation.
    """
    separator = to_native(separator, 'ascii')
    return separator.join(_url_encode_impl(obj, charset, encode_keys, sort, key))


def url_encode_stream(obj, stream=None, charset='utf-8', encode_keys=False,
                      sort=False, key=None, separator=b'&'):
    """Like :meth:`url_encode` but writes the results to a stream
    object.  If the stream is `None` a generator over all encoded
    pairs is returned.

    .. versionadded:: 0.8

    :param obj: the object to encode into a query string.
    :param stream: a stream to write the encoded object into or `None` if
                   an iterator over the encoded pairs should be returned.  In
                   that case the separator argument is ignored.
    :param charset: the charset of the query string.
    :param encode_keys: set to `True` if you have unicode keys. (Ignored on
                        Python 3.x)
    :param sort: set to `True` if you want parameters to be sorted by `key`.
    :param separator: the separator to be used for the pairs.
    :param key: an optional function to be used for sorting.  For more details
                check out the :func:`sorted` documentation.
    """
    separator = to_native(separator, 'ascii')
    gen = _url_encode_impl(obj, charset, encode_keys, sort, key)
    if stream is None:
        return gen
    for idx, chunk in enumerate(gen):
        if idx:
            stream.write(separator)
        stream.write(chunk)


def url_join(base, url, allow_fragments=True):
    """Join a base URL and a possibly relative URL to form an absolute
    interpretation of the latter.

    :param base: the base URL for the join operation.
    :param url: the URL to join.
    :param allow_fragments: indicates whether fragments should be allowed.
    """
    if isinstance(base, tuple):
        base = url_unparse(base)
    if isinstance(url, tuple):
        url = url_unparse(url)

    base, url = normalize_string_tuple((base, url))
    s = make_literal_wrapper(base)

    if not base:
        return url
    if not url:
        return base

    bscheme, bnetloc, bpath, bquery, bfragment = \
        url_parse(base, allow_fragments=allow_fragments)
    scheme, netloc, path, query, fragment = \
        url_parse(url, bscheme, allow_fragments)
    if scheme != bscheme:
        return url
    if netloc:
        return url_unparse((scheme, netloc, path, query, fragment))
    netloc = bnetloc

    if path[:1] == s('/'):
        segments = path.split(s('/'))
    elif not path:
        segments = bpath.split(s('/'))
        if not query:
            query = bquery
    else:
        segments = bpath.split(s('/'))[:-1] + path.split(s('/'))

    # If the rightmost part is "./" we want to keep the slash but
    # remove the dot.
    if segments[-1] == s('.'):
        segments[-1] = s('')

    # Resolve ".." and "."
    segments = [segment for segment in segments if segment != s('.')]
    while 1:
        i = 1
        n = len(segments) - 1
        while i < n:
            if segments[i] == s('..') and \
               segments[i - 1] not in (s(''), s('..')):
                del segments[i - 1:i + 1]
                break
            i += 1
        else:
            break

    # Remove trailing ".." if the URL is absolute
    unwanted_marker = [s(''), s('..')]
    while segments[:2] == unwanted_marker:
        del segments[1]

    path = s('/').join(segments)
    return url_unparse((scheme, netloc, path, query, fragment))


class Href(object):

    """Implements a callable that constructs URLs with the given base. The
    function can be called with any number of positional and keyword
    arguments which than are used to assemble the URL.  Works with URLs
    and posix paths.

    Positional arguments are appended as individual segments to
    the path of the URL:

    >>> href = Href('/foo')
    >>> href('bar', 23)
    '/foo/bar/23'
    >>> href('foo', bar=23)
    '/foo/foo?bar=23'

    If any of the arguments (positional or keyword) evaluates to `None` it
    will be skipped.  If no keyword arguments are given the last argument
    can be a :class:`dict` or :class:`MultiDict` (or any other dict subclass),
    otherwise the keyword arguments are used for the query parameters, cutting
    off the first trailing underscore of the parameter name:

    >>> href(is_=42)
    '/foo?is=42'
    >>> href({'foo': 'bar'})
    '/foo?foo=bar'

    Combining of both methods is not allowed:

    >>> href({'foo': 'bar'}, bar=42)
    Traceback (most recent call last):
      ...
    TypeError: keyword arguments and query-dicts can't be combined

    Accessing attributes on the href object creates a new href object with
    the attribute name as prefix:

    >>> bar_href = href.bar
    >>> bar_href("blub")
    '/foo/bar/blub'

    If `sort` is set to `True` the items are sorted by `key` or the default
    sorting algorithm:

    >>> href = Href("/", sort=True)
    >>> href(a=1, b=2, c=3)
    '/?a=1&b=2&c=3'

    .. versionadded:: 0.5
        `sort` and `key` were added.
    """

    def __init__(self, base='./', charset='utf-8', sort=False, key=None):
        if not base:
            base = './'
        self.base = base
        self.charset = charset
        self.sort = sort
        self.key = key

    def __getattr__(self, name):
        if name[:2] == '__':
            raise AttributeError(name)
        base = self.base
        if base[-1:] != '/':
            base += '/'
        return Href(url_join(base, name), self.charset, self.sort, self.key)

    def __call__(self, *path, **query):
        if path and isinstance(path[-1], dict):
            if query:
                raise TypeError('keyword arguments and query-dicts '
                                'can\'t be combined')
            query, path = path[-1], path[:-1]
        elif query:
            query = dict([(k.endswith('_') and k[:-1] or k, v)
                          for k, v in query.items()])
        path = '/'.join([to_unicode(url_quote(x, self.charset), 'ascii')
                         for x in path if x is not None]).lstrip('/')
        rv = self.base
        if path:
            if not rv.endswith('/'):
                rv += '/'
            rv = url_join(rv, './' + path)
        if query:
            rv += '?' + to_unicode(url_encode(query, self.charset, sort=self.sort,
                                              key=self.key), 'ascii')
        return to_native(rv)