Projects/worblehat-old
Projects
/
worblehat-old
Archived
12
0
Fork 0
This repository has been archived on 2024-07-04. You can view files and clone it, but cannot push or open issues or pull requests.
worblehat-old/python/gdata/__init__.py

836 lines
29 KiB
Python

#!/usr/bin/python
#
# Copyright (C) 2006 Google Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Contains classes representing Google Data elements.
Extends Atom classes to add Google Data specific elements.
"""
__author__ = 'j.s@google.com (Jeffrey Scudder)'
import os
import atom
try:
from xml.etree import cElementTree as ElementTree
except ImportError:
try:
import cElementTree as ElementTree
except ImportError:
try:
from xml.etree import ElementTree
except ImportError:
from elementtree import ElementTree
# XML namespaces which are often used in GData entities.
GDATA_NAMESPACE = 'http://schemas.google.com/g/2005'
GDATA_TEMPLATE = '{http://schemas.google.com/g/2005}%s'
OPENSEARCH_NAMESPACE = 'http://a9.com/-/spec/opensearchrss/1.0/'
OPENSEARCH_TEMPLATE = '{http://a9.com/-/spec/opensearchrss/1.0/}%s'
BATCH_NAMESPACE = 'http://schemas.google.com/gdata/batch'
GACL_NAMESPACE = 'http://schemas.google.com/acl/2007'
GACL_TEMPLATE = '{http://schemas.google.com/acl/2007}%s'
# Labels used in batch request entries to specify the desired CRUD operation.
BATCH_INSERT = 'insert'
BATCH_UPDATE = 'update'
BATCH_DELETE = 'delete'
BATCH_QUERY = 'query'
class Error(Exception):
pass
class MissingRequiredParameters(Error):
pass
class MediaSource(object):
"""GData Entries can refer to media sources, so this class provides a
place to store references to these objects along with some metadata.
"""
def __init__(self, file_handle=None, content_type=None, content_length=None,
file_path=None, file_name=None):
"""Creates an object of type MediaSource.
Args:
file_handle: A file handle pointing to the file to be encapsulated in the
MediaSource
content_type: string The MIME type of the file. Required if a file_handle
is given.
content_length: int The size of the file. Required if a file_handle is
given.
file_path: string (optional) A full path name to the file. Used in
place of a file_handle.
file_name: string The name of the file without any path information.
Required if a file_handle is given.
"""
self.file_handle = file_handle
self.content_type = content_type
self.content_length = content_length
self.file_name = file_name
if (file_handle is None and content_type is not None and
file_path is not None):
self.setFile(file_path, content_type)
def setFile(self, file_name, content_type):
"""A helper function which can create a file handle from a given filename
and set the content type and length all at once.
Args:
file_name: string The path and file name to the file containing the media
content_type: string A MIME type representing the type of the media
"""
self.file_handle = open(file_name, 'rb')
self.content_type = content_type
self.content_length = os.path.getsize(file_name)
self.file_name = os.path.basename(file_name)
class LinkFinder(atom.LinkFinder):
"""An "interface" providing methods to find link elements
GData Entry elements often contain multiple links which differ in the rel
attribute or content type. Often, developers are interested in a specific
type of link so this class provides methods to find specific classes of
links.
This class is used as a mixin in GData entries.
"""
def GetSelfLink(self):
"""Find the first link with rel set to 'self'
Returns:
An atom.Link or none if none of the links had rel equal to 'self'
"""
for a_link in self.link:
if a_link.rel == 'self':
return a_link
return None
def GetEditLink(self):
for a_link in self.link:
if a_link.rel == 'edit':
return a_link
return None
def GetEditMediaLink(self):
"""The Picasa API mistakenly returns media-edit rather than edit-media, but
this may change soon.
"""
for a_link in self.link:
if a_link.rel == 'edit-media':
return a_link
if a_link.rel == 'media-edit':
return a_link
return None
def GetHtmlLink(self):
"""Find the first link with rel of alternate and type of text/html
Returns:
An atom.Link or None if no links matched
"""
for a_link in self.link:
if a_link.rel == 'alternate' and a_link.type == 'text/html':
return a_link
return None
def GetPostLink(self):
"""Get a link containing the POST target URL.
The POST target URL is used to insert new entries.
Returns:
A link object with a rel matching the POST type.
"""
for a_link in self.link:
if a_link.rel == 'http://schemas.google.com/g/2005#post':
return a_link
return None
def GetAclLink(self):
for a_link in self.link:
if a_link.rel == 'http://schemas.google.com/acl/2007#accessControlList':
return a_link
return None
def GetFeedLink(self):
for a_link in self.link:
if a_link.rel == 'http://schemas.google.com/g/2005#feed':
return a_link
return None
def GetNextLink(self):
for a_link in self.link:
if a_link.rel == 'next':
return a_link
return None
def GetPrevLink(self):
for a_link in self.link:
if a_link.rel == 'previous':
return a_link
return None
class TotalResults(atom.AtomBase):
"""opensearch:TotalResults for a GData feed"""
_tag = 'totalResults'
_namespace = OPENSEARCH_NAMESPACE
_children = atom.AtomBase._children.copy()
_attributes = atom.AtomBase._attributes.copy()
def __init__(self, extension_elements=None,
extension_attributes=None, text=None):
self.text = text
self.extension_elements = extension_elements or []
self.extension_attributes = extension_attributes or {}
def TotalResultsFromString(xml_string):
return atom.CreateClassFromXMLString(TotalResults, xml_string)
class StartIndex(atom.AtomBase):
"""The opensearch:startIndex element in GData feed"""
_tag = 'startIndex'
_namespace = OPENSEARCH_NAMESPACE
_children = atom.AtomBase._children.copy()
_attributes = atom.AtomBase._attributes.copy()
def __init__(self, extension_elements=None,
extension_attributes=None, text=None):
self.text = text
self.extension_elements = extension_elements or []
self.extension_attributes = extension_attributes or {}
def StartIndexFromString(xml_string):
return atom.CreateClassFromXMLString(StartIndex, xml_string)
class ItemsPerPage(atom.AtomBase):
"""The opensearch:itemsPerPage element in GData feed"""
_tag = 'itemsPerPage'
_namespace = OPENSEARCH_NAMESPACE
_children = atom.AtomBase._children.copy()
_attributes = atom.AtomBase._attributes.copy()
def __init__(self, extension_elements=None,
extension_attributes=None, text=None):
self.text = text
self.extension_elements = extension_elements or []
self.extension_attributes = extension_attributes or {}
def ItemsPerPageFromString(xml_string):
return atom.CreateClassFromXMLString(ItemsPerPage, xml_string)
class ExtendedProperty(atom.AtomBase):
"""The Google Data extendedProperty element.
Used to store arbitrary key-value information specific to your
application. The value can either be a text string stored as an XML
attribute (.value), or an XML node (XmlBlob) as a child element.
This element is used in the Google Calendar data API and the Google
Contacts data API.
"""
_tag = 'extendedProperty'
_namespace = GDATA_NAMESPACE
_children = atom.AtomBase._children.copy()
_attributes = atom.AtomBase._attributes.copy()
_attributes['name'] = 'name'
_attributes['value'] = 'value'
def __init__(self, name=None, value=None, extension_elements=None,
extension_attributes=None, text=None):
self.name = name
self.value = value
self.text = text
self.extension_elements = extension_elements or []
self.extension_attributes = extension_attributes or {}
def GetXmlBlobExtensionElement(self):
"""Returns the XML blob as an atom.ExtensionElement.
Returns:
An atom.ExtensionElement representing the blob's XML, or None if no
blob was set.
"""
if len(self.extension_elements) < 1:
return None
else:
return self.extension_elements[0]
def GetXmlBlobString(self):
"""Returns the XML blob as a string.
Returns:
A string containing the blob's XML, or None if no blob was set.
"""
blob = self.GetXmlBlobExtensionElement()
if blob:
return blob.ToString()
return None
def SetXmlBlob(self, blob):
"""Sets the contents of the extendedProperty to XML as a child node.
Since the extendedProperty is only allowed one child element as an XML
blob, setting the XML blob will erase any preexisting extension elements
in this object.
Args:
blob: str, ElementTree Element or atom.ExtensionElement representing
the XML blob stored in the extendedProperty.
"""
# Erase any existing extension_elements, clears the child nodes from the
# extendedProperty.
self.extension_elements = []
if isinstance(blob, atom.ExtensionElement):
self.extension_elements.append(blob)
elif ElementTree.iselement(blob):
self.extension_elements.append(atom._ExtensionElementFromElementTree(
blob))
else:
self.extension_elements.append(atom.ExtensionElementFromString(blob))
def ExtendedPropertyFromString(xml_string):
return atom.CreateClassFromXMLString(ExtendedProperty, xml_string)
class GDataEntry(atom.Entry, LinkFinder):
"""Extends Atom Entry to provide data processing"""
_tag = atom.Entry._tag
_namespace = atom.Entry._namespace
_children = atom.Entry._children.copy()
_attributes = atom.Entry._attributes.copy()
def __GetId(self):
return self.__id
# This method was created to strip the unwanted whitespace from the id's
# text node.
def __SetId(self, id):
self.__id = id
if id is not None and id.text is not None:
self.__id.text = id.text.strip()
id = property(__GetId, __SetId)
def IsMedia(self):
"""Determines whether or not an entry is a GData Media entry.
"""
if (self.GetEditMediaLink()):
return True
else:
return False
def GetMediaURL(self):
"""Returns the URL to the media content, if the entry is a media entry.
Otherwise returns None.
"""
if not self.IsMedia():
return None
else:
return self.content.src
def GDataEntryFromString(xml_string):
"""Creates a new GDataEntry instance given a string of XML."""
return atom.CreateClassFromXMLString(GDataEntry, xml_string)
class GDataFeed(atom.Feed, LinkFinder):
"""A Feed from a GData service"""
_tag = 'feed'
_namespace = atom.ATOM_NAMESPACE
_children = atom.Feed._children.copy()
_attributes = atom.Feed._attributes.copy()
_children['{%s}totalResults' % OPENSEARCH_NAMESPACE] = ('total_results',
TotalResults)
_children['{%s}startIndex' % OPENSEARCH_NAMESPACE] = ('start_index',
StartIndex)
_children['{%s}itemsPerPage' % OPENSEARCH_NAMESPACE] = ('items_per_page',
ItemsPerPage)
# Add a conversion rule for atom:entry to make it into a GData
# Entry.
_children['{%s}entry' % atom.ATOM_NAMESPACE] = ('entry', [GDataEntry])
def __GetId(self):
return self.__id
def __SetId(self, id):
self.__id = id
if id is not None and id.text is not None:
self.__id.text = id.text.strip()
id = property(__GetId, __SetId)
def __GetGenerator(self):
return self.__generator
def __SetGenerator(self, generator):
self.__generator = generator
if generator is not None:
self.__generator.text = generator.text.strip()
generator = property(__GetGenerator, __SetGenerator)
def __init__(self, author=None, category=None, contributor=None,
generator=None, icon=None, atom_id=None, link=None, logo=None,
rights=None, subtitle=None, title=None, updated=None, entry=None,
total_results=None, start_index=None, items_per_page=None,
extension_elements=None, extension_attributes=None, text=None):
"""Constructor for Source
Args:
author: list (optional) A list of Author instances which belong to this
class.
category: list (optional) A list of Category instances
contributor: list (optional) A list on Contributor instances
generator: Generator (optional)
icon: Icon (optional)
id: Id (optional) The entry's Id element
link: list (optional) A list of Link instances
logo: Logo (optional)
rights: Rights (optional) The entry's Rights element
subtitle: Subtitle (optional) The entry's subtitle element
title: Title (optional) the entry's title element
updated: Updated (optional) the entry's updated element
entry: list (optional) A list of the Entry instances contained in the
feed.
text: String (optional) The text contents of the element. This is the
contents of the Entry's XML text node.
(Example: <foo>This is the text</foo>)
extension_elements: list (optional) A list of ExtensionElement instances
which are children of this element.
extension_attributes: dict (optional) A dictionary of strings which are
the values for additional XML attributes of this element.
"""
self.author = author or []
self.category = category or []
self.contributor = contributor or []
self.generator = generator
self.icon = icon
self.id = atom_id
self.link = link or []
self.logo = logo
self.rights = rights
self.subtitle = subtitle
self.title = title
self.updated = updated
self.entry = entry or []
self.total_results = total_results
self.start_index = start_index
self.items_per_page = items_per_page
self.text = text
self.extension_elements = extension_elements or []
self.extension_attributes = extension_attributes or {}
def GDataFeedFromString(xml_string):
return atom.CreateClassFromXMLString(GDataFeed, xml_string)
class BatchId(atom.AtomBase):
_tag = 'id'
_namespace = BATCH_NAMESPACE
_children = atom.AtomBase._children.copy()
_attributes = atom.AtomBase._attributes.copy()
def BatchIdFromString(xml_string):
return atom.CreateClassFromXMLString(BatchId, xml_string)
class BatchOperation(atom.AtomBase):
_tag = 'operation'
_namespace = BATCH_NAMESPACE
_children = atom.AtomBase._children.copy()
_attributes = atom.AtomBase._attributes.copy()
_attributes['type'] = 'type'
def __init__(self, op_type=None, extension_elements=None,
extension_attributes=None,
text=None):
self.type = op_type
atom.AtomBase.__init__(self,
extension_elements=extension_elements,
extension_attributes=extension_attributes,
text=text)
def BatchOperationFromString(xml_string):
return atom.CreateClassFromXMLString(BatchOperation, xml_string)
class BatchStatus(atom.AtomBase):
"""The batch:status element present in a batch response entry.
A status element contains the code (HTTP response code) and
reason as elements. In a single request these fields would
be part of the HTTP response, but in a batch request each
Entry operation has a corresponding Entry in the response
feed which includes status information.
See http://code.google.com/apis/gdata/batch.html#Handling_Errors
"""
_tag = 'status'
_namespace = BATCH_NAMESPACE
_children = atom.AtomBase._children.copy()
_attributes = atom.AtomBase._attributes.copy()
_attributes['code'] = 'code'
_attributes['reason'] = 'reason'
_attributes['content-type'] = 'content_type'
def __init__(self, code=None, reason=None, content_type=None,
extension_elements=None, extension_attributes=None, text=None):
self.code = code
self.reason = reason
self.content_type = content_type
atom.AtomBase.__init__(self, extension_elements=extension_elements,
extension_attributes=extension_attributes,
text=text)
def BatchStatusFromString(xml_string):
return atom.CreateClassFromXMLString(BatchStatus, xml_string)
class BatchEntry(GDataEntry):
"""An atom:entry for use in batch requests.
The BatchEntry contains additional members to specify the operation to be
performed on this entry and a batch ID so that the server can reference
individual operations in the response feed. For more information, see:
http://code.google.com/apis/gdata/batch.html
"""
_tag = GDataEntry._tag
_namespace = GDataEntry._namespace
_children = GDataEntry._children.copy()
_children['{%s}operation' % BATCH_NAMESPACE] = ('batch_operation', BatchOperation)
_children['{%s}id' % BATCH_NAMESPACE] = ('batch_id', BatchId)
_children['{%s}status' % BATCH_NAMESPACE] = ('batch_status', BatchStatus)
_attributes = GDataEntry._attributes.copy()
def __init__(self, author=None, category=None, content=None,
contributor=None, atom_id=None, link=None, published=None, rights=None,
source=None, summary=None, control=None, title=None, updated=None,
batch_operation=None, batch_id=None, batch_status=None,
extension_elements=None, extension_attributes=None, text=None):
self.batch_operation = batch_operation
self.batch_id = batch_id
self.batch_status = batch_status
GDataEntry.__init__(self, author=author, category=category,
content=content, contributor=contributor, atom_id=atom_id, link=link,
published=published, rights=rights, source=source, summary=summary,
control=control, title=title, updated=updated,
extension_elements=extension_elements,
extension_attributes=extension_attributes, text=text)
def BatchEntryFromString(xml_string):
return atom.CreateClassFromXMLString(BatchEntry, xml_string)
class BatchInterrupted(atom.AtomBase):
"""The batch:interrupted element sent if batch request was interrupted.
Only appears in a feed if some of the batch entries could not be processed.
See: http://code.google.com/apis/gdata/batch.html#Handling_Errors
"""
_tag = 'interrupted'
_namespace = BATCH_NAMESPACE
_children = atom.AtomBase._children.copy()
_attributes = atom.AtomBase._attributes.copy()
_attributes['reason'] = 'reason'
_attributes['success'] = 'success'
_attributes['failures'] = 'failures'
_attributes['parsed'] = 'parsed'
def __init__(self, reason=None, success=None, failures=None, parsed=None,
extension_elements=None, extension_attributes=None, text=None):
self.reason = reason
self.success = success
self.failures = failures
self.parsed = parsed
atom.AtomBase.__init__(self, extension_elements=extension_elements,
extension_attributes=extension_attributes,
text=text)
def BatchInterruptedFromString(xml_string):
return atom.CreateClassFromXMLString(BatchInterrupted, xml_string)
class BatchFeed(GDataFeed):
"""A feed containing a list of batch request entries."""
_tag = GDataFeed._tag
_namespace = GDataFeed._namespace
_children = GDataFeed._children.copy()
_attributes = GDataFeed._attributes.copy()
_children['{%s}entry' % atom.ATOM_NAMESPACE] = ('entry', [BatchEntry])
_children['{%s}interrupted' % BATCH_NAMESPACE] = ('interrupted', BatchInterrupted)
def __init__(self, author=None, category=None, contributor=None,
generator=None, icon=None, atom_id=None, link=None, logo=None,
rights=None, subtitle=None, title=None, updated=None, entry=None,
total_results=None, start_index=None, items_per_page=None,
interrupted=None,
extension_elements=None, extension_attributes=None, text=None):
self.interrupted = interrupted
GDataFeed.__init__(self, author=author, category=category,
contributor=contributor, generator=generator,
icon=icon, atom_id=atom_id, link=link,
logo=logo, rights=rights, subtitle=subtitle,
title=title, updated=updated, entry=entry,
total_results=total_results, start_index=start_index,
items_per_page=items_per_page,
extension_elements=extension_elements,
extension_attributes=extension_attributes,
text=text)
def AddBatchEntry(self, entry=None, id_url_string=None,
batch_id_string=None, operation_string=None):
"""Logic for populating members of a BatchEntry and adding to the feed.
If the entry is not a BatchEntry, it is converted to a BatchEntry so
that the batch specific members will be present.
The id_url_string can be used in place of an entry if the batch operation
applies to a URL. For example query and delete operations require just
the URL of an entry, no body is sent in the HTTP request. If an
id_url_string is sent instead of an entry, a BatchEntry is created and
added to the feed.
This method also assigns the desired batch id to the entry so that it
can be referenced in the server's response. If the batch_id_string is
None, this method will assign a batch_id to be the index at which this
entry will be in the feed's entry list.
Args:
entry: BatchEntry, atom.Entry, or another Entry flavor (optional) The
entry which will be sent to the server as part of the batch request.
The item must have a valid atom id so that the server knows which
entry this request references.
id_url_string: str (optional) The URL of the entry to be acted on. You
can find this URL in the text member of the atom id for an entry.
If an entry is not sent, this id will be used to construct a new
BatchEntry which will be added to the request feed.
batch_id_string: str (optional) The batch ID to be used to reference
this batch operation in the results feed. If this parameter is None,
the current length of the feed's entry array will be used as a
count. Note that batch_ids should either always be specified or
never, mixing could potentially result in duplicate batch ids.
operation_string: str (optional) The desired batch operation which will
set the batch_operation.type member of the entry. Options are
'insert', 'update', 'delete', and 'query'
Raises:
MissingRequiredParameters: Raised if neither an id_ url_string nor an
entry are provided in the request.
Returns:
The added entry.
"""
if entry is None and id_url_string is None:
raise MissingRequiredParameters('supply either an entry or URL string')
if entry is None and id_url_string is not None:
entry = BatchEntry(atom_id=atom.Id(text=id_url_string))
# TODO: handle cases in which the entry lacks batch_... members.
#if not isinstance(entry, BatchEntry):
# Convert the entry to a batch entry.
if batch_id_string is not None:
entry.batch_id = BatchId(text=batch_id_string)
elif entry.batch_id is None or entry.batch_id.text is None:
entry.batch_id = BatchId(text=str(len(self.entry)))
if operation_string is not None:
entry.batch_operation = BatchOperation(op_type=operation_string)
self.entry.append(entry)
return entry
def AddInsert(self, entry, batch_id_string=None):
"""Add an insert request to the operations in this batch request feed.
If the entry doesn't yet have an operation or a batch id, these will
be set to the insert operation and a batch_id specified as a parameter.
Args:
entry: BatchEntry The entry which will be sent in the batch feed as an
insert request.
batch_id_string: str (optional) The batch ID to be used to reference
this batch operation in the results feed. If this parameter is None,
the current length of the feed's entry array will be used as a
count. Note that batch_ids should either always be specified or
never, mixing could potentially result in duplicate batch ids.
"""
entry = self.AddBatchEntry(entry=entry, batch_id_string=batch_id_string,
operation_string=BATCH_INSERT)
def AddUpdate(self, entry, batch_id_string=None):
"""Add an update request to the list of batch operations in this feed.
Sets the operation type of the entry to insert if it is not already set
and assigns the desired batch id to the entry so that it can be
referenced in the server's response.
Args:
entry: BatchEntry The entry which will be sent to the server as an
update (HTTP PUT) request. The item must have a valid atom id
so that the server knows which entry to replace.
batch_id_string: str (optional) The batch ID to be used to reference
this batch operation in the results feed. If this parameter is None,
the current length of the feed's entry array will be used as a
count. See also comments for AddInsert.
"""
entry = self.AddBatchEntry(entry=entry, batch_id_string=batch_id_string,
operation_string=BATCH_UPDATE)
def AddDelete(self, url_string=None, entry=None, batch_id_string=None):
"""Adds a delete request to the batch request feed.
This method takes either the url_string which is the atom id of the item
to be deleted, or the entry itself. The atom id of the entry must be
present so that the server knows which entry should be deleted.
Args:
url_string: str (optional) The URL of the entry to be deleted. You can
find this URL in the text member of the atom id for an entry.
entry: BatchEntry (optional) The entry to be deleted.
batch_id_string: str (optional)
Raises:
MissingRequiredParameters: Raised if neither a url_string nor an entry
are provided in the request.
"""
entry = self.AddBatchEntry(entry=entry, id_url_string=url_string,
batch_id_string=batch_id_string,
operation_string=BATCH_DELETE)
def AddQuery(self, url_string=None, entry=None, batch_id_string=None):
"""Adds a query request to the batch request feed.
This method takes either the url_string which is the query URL
whose results will be added to the result feed. The query URL will
be encapsulated in a BatchEntry, and you may pass in the BatchEntry
with a query URL instead of sending a url_string.
Args:
url_string: str (optional)
entry: BatchEntry (optional)
batch_id_string: str (optional)
Raises:
MissingRequiredParameters
"""
entry = self.AddBatchEntry(entry=entry, id_url_string=url_string,
batch_id_string=batch_id_string,
operation_string=BATCH_QUERY)
def GetBatchLink(self):
for link in self.link:
if link.rel == 'http://schemas.google.com/g/2005#batch':
return link
return None
def BatchFeedFromString(xml_string):
return atom.CreateClassFromXMLString(BatchFeed, xml_string)
class EntryLink(atom.AtomBase):
"""The gd:entryLink element"""
_tag = 'entryLink'
_namespace = GDATA_NAMESPACE
_children = atom.AtomBase._children.copy()
_attributes = atom.AtomBase._attributes.copy()
# The entry used to be an atom.Entry, now it is a GDataEntry.
_children['{%s}entry' % atom.ATOM_NAMESPACE] = ('entry', GDataEntry)
_attributes['rel'] = 'rel'
_attributes['readOnly'] = 'read_only'
_attributes['href'] = 'href'
def __init__(self, href=None, read_only=None, rel=None,
entry=None, extension_elements=None,
extension_attributes=None, text=None):
self.href = href
self.read_only = read_only
self.rel = rel
self.entry = entry
self.text = text
self.extension_elements = extension_elements or []
self.extension_attributes = extension_attributes or {}
def EntryLinkFromString(xml_string):
return atom.CreateClassFromXMLString(EntryLink, xml_string)
class FeedLink(atom.AtomBase):
"""The gd:feedLink element"""
_tag = 'feedLink'
_namespace = GDATA_NAMESPACE
_children = atom.AtomBase._children.copy()
_attributes = atom.AtomBase._attributes.copy()
_children['{%s}feed' % atom.ATOM_NAMESPACE] = ('feed', GDataFeed)
_attributes['rel'] = 'rel'
_attributes['readOnly'] = 'read_only'
_attributes['countHint'] = 'count_hint'
_attributes['href'] = 'href'
def __init__(self, count_hint=None, href=None, read_only=None, rel=None,
feed=None, extension_elements=None, extension_attributes=None,
text=None):
self.count_hint = count_hint
self.href = href
self.read_only = read_only
self.rel = rel
self.feed = feed
self.text = text
self.extension_elements = extension_elements or []
self.extension_attributes = extension_attributes or {}
def FeedLinkFromString(xml_string):
return atom.CreateClassFromXMLString(FeedLink, xml_string)