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This repository has been archived on 2024-07-04.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/atom/service.py

741 lines
28 KiB
Python

#!/usr/bin/python
#
# Copyright (C) 2006, 2007, 2008 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.
"""AtomService provides CRUD ops. in line with the Atom Publishing Protocol.
AtomService: Encapsulates the ability to perform insert, update and delete
operations with the Atom Publishing Protocol on which GData is
based. An instance can perform query, insertion, deletion, and
update.
HttpRequest: Function that performs a GET, POST, PUT, or DELETE HTTP request
to the specified end point. An AtomService object or a subclass can be
used to specify information about the request.
"""
__author__ = 'api.jscudder (Jeff Scudder)'
import atom.http_interface
import atom.url
import atom.http
import atom.token_store
import os
import httplib
import urllib
import re
import base64
import socket
import warnings
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
import atom
class AtomService(object):
"""Performs Atom Publishing Protocol CRUD operations.
The AtomService contains methods to perform HTTP CRUD operations.
"""
# Default values for members
port = 80
ssl = False
# Set the current_token to force the AtomService to use this token
# instead of searching for an appropriate token in the token_store.
current_token = None
auto_store_tokens = True
auto_set_current_token = True
def _get_override_token(self):
return self.current_token
def _set_override_token(self, token):
self.current_token = token
override_token = property(_get_override_token, _set_override_token)
#@atom.v1_deprecated('Please use atom.client.AtomPubClient instead.')
def __init__(self, server=None, additional_headers=None,
application_name='', http_client=None, token_store=None):
"""Creates a new AtomService client.
Args:
server: string (optional) The start of a URL for the server
to which all operations should be directed. Example:
'www.google.com'
additional_headers: dict (optional) Any additional HTTP headers which
should be included with CRUD operations.
http_client: An object responsible for making HTTP requests using a
request method. If none is provided, a new instance of
atom.http.ProxiedHttpClient will be used.
token_store: Keeps a collection of authorization tokens which can be
applied to requests for a specific URLs. Critical methods are
find_token based on a URL (atom.url.Url or a string), add_token,
and remove_token.
"""
self.http_client = http_client or atom.http.ProxiedHttpClient()
self.token_store = token_store or atom.token_store.TokenStore()
self.server = server
self.additional_headers = additional_headers or {}
self.additional_headers['User-Agent'] = atom.http_interface.USER_AGENT % (
application_name,)
# If debug is True, the HTTPConnection will display debug information
self._set_debug(False)
__init__ = atom.v1_deprecated(
'Please use atom.client.AtomPubClient instead.')(
__init__)
def _get_debug(self):
return self.http_client.debug
def _set_debug(self, value):
self.http_client.debug = value
debug = property(_get_debug, _set_debug,
doc='If True, HTTP debug information is printed.')
def use_basic_auth(self, username, password, scopes=None):
if username is not None and password is not None:
if scopes is None:
scopes = [atom.token_store.SCOPE_ALL]
base_64_string = base64.encodestring('%s:%s' % (username, password))
token = BasicAuthToken('Basic %s' % base_64_string.strip(),
scopes=[atom.token_store.SCOPE_ALL])
if self.auto_set_current_token:
self.current_token = token
if self.auto_store_tokens:
return self.token_store.add_token(token)
return True
return False
def UseBasicAuth(self, username, password, for_proxy=False):
"""Sets an Authenticaiton: Basic HTTP header containing plaintext.
Deprecated, use use_basic_auth instead.
The username and password are base64 encoded and added to an HTTP header
which will be included in each request. Note that your username and
password are sent in plaintext.
Args:
username: str
password: str
"""
self.use_basic_auth(username, password)
#@atom.v1_deprecated('Please use atom.client.AtomPubClient for requests.')
def request(self, operation, url, data=None, headers=None,
url_params=None):
if isinstance(url, (str, unicode)):
if url.startswith('http:') and self.ssl:
# Force all requests to be https if self.ssl is True.
url = atom.url.parse_url('https:' + url[5:])
elif not url.startswith('http') and self.ssl:
url = atom.url.parse_url('https://%s%s' % (self.server, url))
elif not url.startswith('http'):
url = atom.url.parse_url('http://%s%s' % (self.server, url))
else:
url = atom.url.parse_url(url)
if url_params:
for name, value in url_params.iteritems():
url.params[name] = value
all_headers = self.additional_headers.copy()
if headers:
all_headers.update(headers)
# If the list of headers does not include a Content-Length, attempt to
# calculate it based on the data object.
if data and 'Content-Length' not in all_headers:
content_length = CalculateDataLength(data)
if content_length:
all_headers['Content-Length'] = str(content_length)
# Find an Authorization token for this URL if one is available.
if self.override_token:
auth_token = self.override_token
else:
auth_token = self.token_store.find_token(url)
return auth_token.perform_request(self.http_client, operation, url,
data=data, headers=all_headers)
request = atom.v1_deprecated(
'Please use atom.client.AtomPubClient for requests.')(
request)
# CRUD operations
def Get(self, uri, extra_headers=None, url_params=None, escape_params=True):
"""Query the APP server with the given URI
The uri is the portion of the URI after the server value
(server example: 'www.google.com').
Example use:
To perform a query against Google Base, set the server to
'base.google.com' and set the uri to '/base/feeds/...', where ... is
your query. For example, to find snippets for all digital cameras uri
should be set to: '/base/feeds/snippets?bq=digital+camera'
Args:
uri: string The query in the form of a URI. Example:
'/base/feeds/snippets?bq=digital+camera'.
extra_headers: dicty (optional) Extra HTTP headers to be included
in the GET request. These headers are in addition to
those stored in the client's additional_headers property.
The client automatically sets the Content-Type and
Authorization headers.
url_params: dict (optional) Additional URL parameters to be included
in the query. These are translated into query arguments
in the form '&dict_key=value&...'.
Example: {'max-results': '250'} becomes &max-results=250
escape_params: boolean (optional) If false, the calling code has already
ensured that the query will form a valid URL (all
reserved characters have been escaped). If true, this
method will escape the query and any URL parameters
provided.
Returns:
httplib.HTTPResponse The server's response to the GET request.
"""
return self.request('GET', uri, data=None, headers=extra_headers,
url_params=url_params)
def Post(self, data, uri, extra_headers=None, url_params=None,
escape_params=True, content_type='application/atom+xml'):
"""Insert data into an APP server at the given URI.
Args:
data: string, ElementTree._Element, or something with a __str__ method
The XML to be sent to the uri.
uri: string The location (feed) to which the data should be inserted.
Example: '/base/feeds/items'.
extra_headers: dict (optional) HTTP headers which are to be included.
The client automatically sets the Content-Type,
Authorization, and Content-Length headers.
url_params: dict (optional) Additional URL parameters to be included
in the URI. These are translated into query arguments
in the form '&dict_key=value&...'.
Example: {'max-results': '250'} becomes &max-results=250
escape_params: boolean (optional) If false, the calling code has already
ensured that the query will form a valid URL (all
reserved characters have been escaped). If true, this
method will escape the query and any URL parameters
provided.
Returns:
httplib.HTTPResponse Server's response to the POST request.
"""
if extra_headers is None:
extra_headers = {}
if content_type:
extra_headers['Content-Type'] = content_type
return self.request('POST', uri, data=data, headers=extra_headers,
url_params=url_params)
def Put(self, data, uri, extra_headers=None, url_params=None,
escape_params=True, content_type='application/atom+xml'):
"""Updates an entry at the given URI.
Args:
data: string, ElementTree._Element, or xml_wrapper.ElementWrapper The
XML containing the updated data.
uri: string A URI indicating entry to which the update will be applied.
Example: '/base/feeds/items/ITEM-ID'
extra_headers: dict (optional) HTTP headers which are to be included.
The client automatically sets the Content-Type,
Authorization, and Content-Length headers.
url_params: dict (optional) Additional URL parameters to be included
in the URI. These are translated into query arguments
in the form '&dict_key=value&...'.
Example: {'max-results': '250'} becomes &max-results=250
escape_params: boolean (optional) If false, the calling code has already
ensured that the query will form a valid URL (all
reserved characters have been escaped). If true, this
method will escape the query and any URL parameters
provided.
Returns:
httplib.HTTPResponse Server's response to the PUT request.
"""
if extra_headers is None:
extra_headers = {}
if content_type:
extra_headers['Content-Type'] = content_type
return self.request('PUT', uri, data=data, headers=extra_headers,
url_params=url_params)
def Delete(self, uri, extra_headers=None, url_params=None,
escape_params=True):
"""Deletes the entry at the given URI.
Args:
uri: string The URI of the entry to be deleted. Example:
'/base/feeds/items/ITEM-ID'
extra_headers: dict (optional) HTTP headers which are to be included.
The client automatically sets the Content-Type and
Authorization headers.
url_params: dict (optional) Additional URL parameters to be included
in the URI. These are translated into query arguments
in the form '&dict_key=value&...'.
Example: {'max-results': '250'} becomes &max-results=250
escape_params: boolean (optional) If false, the calling code has already
ensured that the query will form a valid URL (all
reserved characters have been escaped). If true, this
method will escape the query and any URL parameters
provided.
Returns:
httplib.HTTPResponse Server's response to the DELETE request.
"""
return self.request('DELETE', uri, data=None, headers=extra_headers,
url_params=url_params)
class BasicAuthToken(atom.http_interface.GenericToken):
def __init__(self, auth_header, scopes=None):
"""Creates a token used to add Basic Auth headers to HTTP requests.
Args:
auth_header: str The value for the Authorization header.
scopes: list of str or atom.url.Url specifying the beginnings of URLs
for which this token can be used. For example, if scopes contains
'http://example.com/foo', then this token can be used for a request to
'http://example.com/foo/bar' but it cannot be used for a request to
'http://example.com/baz'
"""
self.auth_header = auth_header
self.scopes = scopes or []
def perform_request(self, http_client, operation, url, data=None,
headers=None):
"""Sets the Authorization header to the basic auth string."""
if headers is None:
headers = {'Authorization':self.auth_header}
else:
headers['Authorization'] = self.auth_header
return http_client.request(operation, url, data=data, headers=headers)
def __str__(self):
return self.auth_header
def valid_for_scope(self, url):
"""Tells the caller if the token authorizes access to the desired URL.
"""
if isinstance(url, (str, unicode)):
url = atom.url.parse_url(url)
for scope in self.scopes:
if scope == atom.token_store.SCOPE_ALL:
return True
if isinstance(scope, (str, unicode)):
scope = atom.url.parse_url(scope)
if scope == url:
return True
# Check the host and the path, but ignore the port and protocol.
elif scope.host == url.host and not scope.path:
return True
elif scope.host == url.host and scope.path and not url.path:
continue
elif scope.host == url.host and url.path.startswith(scope.path):
return True
return False
def PrepareConnection(service, full_uri):
"""Opens a connection to the server based on the full URI.
This method is deprecated, instead use atom.http.HttpClient.request.
Examines the target URI and the proxy settings, which are set as
environment variables, to open a connection with the server. This
connection is used to make an HTTP request.
Args:
service: atom.AtomService or a subclass. It must have a server string which
represents the server host to which the request should be made. It may also
have a dictionary of additional_headers to send in the HTTP request.
full_uri: str Which is the target relative (lacks protocol and host) or
absolute URL to be opened. Example:
'https://www.google.com/accounts/ClientLogin' or
'base/feeds/snippets' where the server is set to www.google.com.
Returns:
A tuple containing the httplib.HTTPConnection and the full_uri for the
request.
"""
deprecation('calling deprecated function PrepareConnection')
(server, port, ssl, partial_uri) = ProcessUrl(service, full_uri)
if ssl:
# destination is https
proxy = os.environ.get('https_proxy')
if proxy:
(p_server, p_port, p_ssl, p_uri) = ProcessUrl(service, proxy, True)
proxy_username = os.environ.get('proxy-username')
if not proxy_username:
proxy_username = os.environ.get('proxy_username')
proxy_password = os.environ.get('proxy-password')
if not proxy_password:
proxy_password = os.environ.get('proxy_password')
if proxy_username:
user_auth = base64.encodestring('%s:%s' % (proxy_username,
proxy_password))
proxy_authorization = ('Proxy-authorization: Basic %s\r\n' % (
user_auth.strip()))
else:
proxy_authorization = ''
proxy_connect = 'CONNECT %s:%s HTTP/1.0\r\n' % (server, port)
user_agent = 'User-Agent: %s\r\n' % (
service.additional_headers['User-Agent'])
proxy_pieces = (proxy_connect + proxy_authorization + user_agent
+ '\r\n')
#now connect, very simple recv and error checking
p_sock = socket.socket(socket.AF_INET,socket.SOCK_STREAM)
p_sock.connect((p_server,p_port))
p_sock.sendall(proxy_pieces)
response = ''
# Wait for the full response.
while response.find("\r\n\r\n") == -1:
response += p_sock.recv(8192)
p_status=response.split()[1]
if p_status!=str(200):
raise 'Error status=',str(p_status)
# Trivial setup for ssl socket.
ssl = socket.ssl(p_sock, None, None)
fake_sock = httplib.FakeSocket(p_sock, ssl)
# Initalize httplib and replace with the proxy socket.
connection = httplib.HTTPConnection(server)
connection.sock=fake_sock
full_uri = partial_uri
else:
connection = httplib.HTTPSConnection(server, port)
full_uri = partial_uri
else:
# destination is http
proxy = os.environ.get('http_proxy')
if proxy:
(p_server, p_port, p_ssl, p_uri) = ProcessUrl(service.server, proxy, True)
proxy_username = os.environ.get('proxy-username')
if not proxy_username:
proxy_username = os.environ.get('proxy_username')
proxy_password = os.environ.get('proxy-password')
if not proxy_password:
proxy_password = os.environ.get('proxy_password')
if proxy_username:
UseBasicAuth(service, proxy_username, proxy_password, True)
connection = httplib.HTTPConnection(p_server, p_port)
if not full_uri.startswith("http://"):
if full_uri.startswith("/"):
full_uri = "http://%s%s" % (service.server, full_uri)
else:
full_uri = "http://%s/%s" % (service.server, full_uri)
else:
connection = httplib.HTTPConnection(server, port)
full_uri = partial_uri
return (connection, full_uri)
def UseBasicAuth(service, username, password, for_proxy=False):
"""Sets an Authenticaiton: Basic HTTP header containing plaintext.
Deprecated, use AtomService.use_basic_auth insread.
The username and password are base64 encoded and added to an HTTP header
which will be included in each request. Note that your username and
password are sent in plaintext. The auth header is added to the
additional_headers dictionary in the service object.
Args:
service: atom.AtomService or a subclass which has an
additional_headers dict as a member.
username: str
password: str
"""
deprecation('calling deprecated function UseBasicAuth')
base_64_string = base64.encodestring('%s:%s' % (username, password))
base_64_string = base_64_string.strip()
if for_proxy:
header_name = 'Proxy-Authorization'
else:
header_name = 'Authorization'
service.additional_headers[header_name] = 'Basic %s' % (base_64_string,)
def ProcessUrl(service, url, for_proxy=False):
"""Processes a passed URL. If the URL does not begin with https?, then
the default value for server is used
This method is deprecated, use atom.url.parse_url instead.
"""
if not isinstance(url, atom.url.Url):
url = atom.url.parse_url(url)
server = url.host
ssl = False
port = 80
if not server:
if hasattr(service, 'server'):
server = service.server
else:
server = service
if not url.protocol and hasattr(service, 'ssl'):
ssl = service.ssl
if hasattr(service, 'port'):
port = service.port
else:
if url.protocol == 'https':
ssl = True
elif url.protocol == 'http':
ssl = False
if url.port:
port = int(url.port)
elif port == 80 and ssl:
port = 443
return (server, port, ssl, url.get_request_uri())
def DictionaryToParamList(url_parameters, escape_params=True):
"""Convert a dictionary of URL arguments into a URL parameter string.
This function is deprcated, use atom.url.Url instead.
Args:
url_parameters: The dictionaty of key-value pairs which will be converted
into URL parameters. For example,
{'dry-run': 'true', 'foo': 'bar'}
will become ['dry-run=true', 'foo=bar'].
Returns:
A list which contains a string for each key-value pair. The strings are
ready to be incorporated into a URL by using '&'.join([] + parameter_list)
"""
# Choose which function to use when modifying the query and parameters.
# Use quote_plus when escape_params is true.
transform_op = [str, urllib.quote_plus][bool(escape_params)]
# Create a list of tuples containing the escaped version of the
# parameter-value pairs.
parameter_tuples = [(transform_op(param), transform_op(value))
for param, value in (url_parameters or {}).items()]
# Turn parameter-value tuples into a list of strings in the form
# 'PARAMETER=VALUE'.
return ['='.join(x) for x in parameter_tuples]
def BuildUri(uri, url_params=None, escape_params=True):
"""Converts a uri string and a collection of parameters into a URI.
This function is deprcated, use atom.url.Url instead.
Args:
uri: string
url_params: dict (optional)
escape_params: boolean (optional)
uri: string The start of the desired URI. This string can alrady contain
URL parameters. Examples: '/base/feeds/snippets',
'/base/feeds/snippets?bq=digital+camera'
url_parameters: dict (optional) Additional URL parameters to be included
in the query. These are translated into query arguments
in the form '&dict_key=value&...'.
Example: {'max-results': '250'} becomes &max-results=250
escape_params: boolean (optional) If false, the calling code has already
ensured that the query will form a valid URL (all
reserved characters have been escaped). If true, this
method will escape the query and any URL parameters
provided.
Returns:
string The URI consisting of the escaped URL parameters appended to the
initial uri string.
"""
# Prepare URL parameters for inclusion into the GET request.
parameter_list = DictionaryToParamList(url_params, escape_params)
# Append the URL parameters to the URL.
if parameter_list:
if uri.find('?') != -1:
# If there are already URL parameters in the uri string, add the
# parameters after a new & character.
full_uri = '&'.join([uri] + parameter_list)
else:
# The uri string did not have any URL parameters (no ? character)
# so put a ? between the uri and URL parameters.
full_uri = '%s%s' % (uri, '?%s' % ('&'.join([] + parameter_list)))
else:
full_uri = uri
return full_uri
def HttpRequest(service, operation, data, uri, extra_headers=None,
url_params=None, escape_params=True, content_type='application/atom+xml'):
"""Performs an HTTP call to the server, supports GET, POST, PUT, and DELETE.
This method is deprecated, use atom.http.HttpClient.request instead.
Usage example, perform and HTTP GET on http://www.google.com/:
import atom.service
client = atom.service.AtomService()
http_response = client.Get('http://www.google.com/')
or you could set the client.server to 'www.google.com' and use the
following:
client.server = 'www.google.com'
http_response = client.Get('/')
Args:
service: atom.AtomService object which contains some of the parameters
needed to make the request. The following members are used to
construct the HTTP call: server (str), additional_headers (dict),
port (int), and ssl (bool).
operation: str The HTTP operation to be performed. This is usually one of
'GET', 'POST', 'PUT', or 'DELETE'
data: ElementTree, filestream, list of parts, or other object which can be
converted to a string.
Should be set to None when performing a GET or PUT.
If data is a file-like object which can be read, this method will read
a chunk of 100K bytes at a time and send them.
If the data is a list of parts to be sent, each part will be evaluated
and sent.
uri: The beginning of the URL to which the request should be sent.
Examples: '/', '/base/feeds/snippets',
'/m8/feeds/contacts/default/base'
extra_headers: dict of strings. HTTP headers which should be sent
in the request. These headers are in addition to those stored in
service.additional_headers.
url_params: dict of strings. Key value pairs to be added to the URL as
URL parameters. For example {'foo':'bar', 'test':'param'} will
become ?foo=bar&test=param.
escape_params: bool default True. If true, the keys and values in
url_params will be URL escaped when the form is constructed
(Special characters converted to %XX form.)
content_type: str The MIME type for the data being sent. Defaults to
'application/atom+xml', this is only used if data is set.
"""
deprecation('call to deprecated function HttpRequest')
full_uri = BuildUri(uri, url_params, escape_params)
(connection, full_uri) = PrepareConnection(service, full_uri)
if extra_headers is None:
extra_headers = {}
# Turn on debug mode if the debug member is set.
if service.debug:
connection.debuglevel = 1
connection.putrequest(operation, full_uri)
# If the list of headers does not include a Content-Length, attempt to
# calculate it based on the data object.
if (data and not service.additional_headers.has_key('Content-Length') and
not extra_headers.has_key('Content-Length')):
content_length = CalculateDataLength(data)
if content_length:
extra_headers['Content-Length'] = str(content_length)
if content_type:
extra_headers['Content-Type'] = content_type
# Send the HTTP headers.
if isinstance(service.additional_headers, dict):
for header in service.additional_headers:
connection.putheader(header, service.additional_headers[header])
if isinstance(extra_headers, dict):
for header in extra_headers:
connection.putheader(header, extra_headers[header])
connection.endheaders()
# If there is data, send it in the request.
if data:
if isinstance(data, list):
for data_part in data:
__SendDataPart(data_part, connection)
else:
__SendDataPart(data, connection)
# Return the HTTP Response from the server.
return connection.getresponse()
def __SendDataPart(data, connection):
"""This method is deprecated, use atom.http._send_data_part"""
deprecated('call to deprecated function __SendDataPart')
if isinstance(data, str):
#TODO add handling for unicode.
connection.send(data)
return
elif ElementTree.iselement(data):
connection.send(ElementTree.tostring(data))
return
# Check to see if data is a file-like object that has a read method.
elif hasattr(data, 'read'):
# Read the file and send it a chunk at a time.
while 1:
binarydata = data.read(100000)
if binarydata == '': break
connection.send(binarydata)
return
else:
# The data object was not a file.
# Try to convert to a string and send the data.
connection.send(str(data))
return
def CalculateDataLength(data):
"""Attempts to determine the length of the data to send.
This method will respond with a length only if the data is a string or
and ElementTree element.
Args:
data: object If this is not a string or ElementTree element this funtion
will return None.
"""
if isinstance(data, str):
return len(data)
elif isinstance(data, list):
return None
elif ElementTree.iselement(data):
return len(ElementTree.tostring(data))
elif hasattr(data, 'read'):
# If this is a file-like object, don't try to guess the length.
return None
else:
return len(str(data))
def deprecation(message):
warnings.warn(message, DeprecationWarning, stacklevel=2)