Merge branch 'release-v0.17.1' of github.com:matrix-org/synapse v0.17.1

17 files changed, 506 insertions, 293 deletions

diff --git a/synapse/rest/__init__.py b/synapse/rest/__init__.py
index 14227f1cdb..326780405e 100644
--- a/synapse/rest/__init__.py
+++ b/synapse/rest/__init__.py
@@ -46,7 +46,9 @@ from synapse.rest.client.v2_alpha import (
     account_data,
     report_event,
     openid,
+    notifications,
     devices,
+    thirdparty,
 )
 
 from synapse.http.server import JsonResource
@@ -91,4 +93,6 @@ class ClientRestResource(JsonResource):
         account_data.register_servlets(hs, client_resource)
         report_event.register_servlets(hs, client_resource)
         openid.register_servlets(hs, client_resource)
+        notifications.register_servlets(hs, client_resource)
         devices.register_servlets(hs, client_resource)
+        thirdparty.register_servlets(hs, client_resource)
diff --git a/synapse/rest/client/v1/admin.py b/synapse/rest/client/v1/admin.py
index b0cb31a448..af21661d7c 100644
--- a/synapse/rest/client/v1/admin.py
+++ b/synapse/rest/client/v1/admin.py
@@ -28,6 +28,10 @@ logger = logging.getLogger(__name__)
 class WhoisRestServlet(ClientV1RestServlet):
     PATTERNS = client_path_patterns("/admin/whois/(?P<user_id>[^/]*)")
 
+    def __init__(self, hs):
+        super(WhoisRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     @defer.inlineCallbacks
     def on_GET(self, request, user_id):
         target_user = UserID.from_string(user_id)
@@ -82,6 +86,10 @@ class PurgeHistoryRestServlet(ClientV1RestServlet):
         "/admin/purge_history/(?P<room_id>[^/]*)/(?P<event_id>[^/]*)"
     )
 
+    def __init__(self, hs):
+        super(PurgeHistoryRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     @defer.inlineCallbacks
     def on_POST(self, request, room_id, event_id):
         requester = yield self.auth.get_user_by_req(request)
diff --git a/synapse/rest/client/v1/base.py b/synapse/rest/client/v1/base.py
index 96b49b01f2..c2a8447860 100644
--- a/synapse/rest/client/v1/base.py
+++ b/synapse/rest/client/v1/base.py
@@ -57,7 +57,6 @@ class ClientV1RestServlet(RestServlet):
             hs (synapse.server.HomeServer):
         """
         self.hs = hs
-        self.handlers = hs.get_handlers()
         self.builder_factory = hs.get_event_builder_factory()
         self.auth = hs.get_v1auth()
         self.txns = HttpTransactionStore()
diff --git a/synapse/rest/client/v1/directory.py b/synapse/rest/client/v1/directory.py
index 8ac09419dc..09d0831594 100644
--- a/synapse/rest/client/v1/directory.py
+++ b/synapse/rest/client/v1/directory.py
@@ -36,6 +36,10 @@ def register_servlets(hs, http_server):
 class ClientDirectoryServer(ClientV1RestServlet):
     PATTERNS = client_path_patterns("/directory/room/(?P<room_alias>[^/]*)$")
 
+    def __init__(self, hs):
+        super(ClientDirectoryServer, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     @defer.inlineCallbacks
     def on_GET(self, request, room_alias):
         room_alias = RoomAlias.from_string(room_alias)
@@ -146,6 +150,7 @@ class ClientDirectoryListServer(ClientV1RestServlet):
     def __init__(self, hs):
         super(ClientDirectoryListServer, self).__init__(hs)
         self.store = hs.get_datastore()
+        self.handlers = hs.get_handlers()
 
     @defer.inlineCallbacks
     def on_GET(self, request, room_id):
diff --git a/synapse/rest/client/v1/events.py b/synapse/rest/client/v1/events.py
index 498bb9e18a..701b6f549b 100644
--- a/synapse/rest/client/v1/events.py
+++ b/synapse/rest/client/v1/events.py
@@ -32,6 +32,10 @@ class EventStreamRestServlet(ClientV1RestServlet):
 
     DEFAULT_LONGPOLL_TIME_MS = 30000
 
+    def __init__(self, hs):
+        super(EventStreamRestServlet, self).__init__(hs)
+        self.event_stream_handler = hs.get_event_stream_handler()
+
     @defer.inlineCallbacks
     def on_GET(self, request):
         requester = yield self.auth.get_user_by_req(
@@ -46,7 +50,6 @@ class EventStreamRestServlet(ClientV1RestServlet):
         if "room_id" in request.args:
             room_id = request.args["room_id"][0]
 
-        handler = self.handlers.event_stream_handler
         pagin_config = PaginationConfig.from_request(request)
         timeout = EventStreamRestServlet.DEFAULT_LONGPOLL_TIME_MS
         if "timeout" in request.args:
@@ -57,7 +60,7 @@ class EventStreamRestServlet(ClientV1RestServlet):
 
         as_client_event = "raw" not in request.args
 
-        chunk = yield handler.get_stream(
+        chunk = yield self.event_stream_handler.get_stream(
             requester.user.to_string(),
             pagin_config,
             timeout=timeout,
@@ -80,12 +83,12 @@ class EventRestServlet(ClientV1RestServlet):
     def __init__(self, hs):
         super(EventRestServlet, self).__init__(hs)
         self.clock = hs.get_clock()
+        self.event_handler = hs.get_event_handler()
 
     @defer.inlineCallbacks
     def on_GET(self, request, event_id):
         requester = yield self.auth.get_user_by_req(request)
-        handler = self.handlers.event_handler
-        event = yield handler.get_event(requester.user, event_id)
+        event = yield self.event_handler.get_event(requester.user, event_id)
 
         time_now = self.clock.time_msec()
         if event:
diff --git a/synapse/rest/client/v1/initial_sync.py b/synapse/rest/client/v1/initial_sync.py
index 36c3520567..113a49e539 100644
--- a/synapse/rest/client/v1/initial_sync.py
+++ b/synapse/rest/client/v1/initial_sync.py
@@ -23,6 +23,10 @@ from .base import ClientV1RestServlet, client_path_patterns
 class InitialSyncRestServlet(ClientV1RestServlet):
     PATTERNS = client_path_patterns("/initialSync$")
 
+    def __init__(self, hs):
+        super(InitialSyncRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     @defer.inlineCallbacks
     def on_GET(self, request):
         requester = yield self.auth.get_user_by_req(request)
diff --git a/synapse/rest/client/v1/login.py b/synapse/rest/client/v1/login.py
index 92fcae674a..6c0eec8fb3 100644
--- a/synapse/rest/client/v1/login.py
+++ b/synapse/rest/client/v1/login.py
@@ -54,12 +54,9 @@ class LoginRestServlet(ClientV1RestServlet):
         self.jwt_secret = hs.config.jwt_secret
         self.jwt_algorithm = hs.config.jwt_algorithm
         self.cas_enabled = hs.config.cas_enabled
-        self.cas_server_url = hs.config.cas_server_url
-        self.cas_required_attributes = hs.config.cas_required_attributes
-        self.servername = hs.config.server_name
-        self.http_client = hs.get_simple_http_client()
         self.auth_handler = self.hs.get_auth_handler()
         self.device_handler = self.hs.get_device_handler()
+        self.handlers = hs.get_handlers()
 
     def on_GET(self, request):
         flows = []
@@ -110,17 +107,6 @@ class LoginRestServlet(ClientV1RestServlet):
                                        LoginRestServlet.JWT_TYPE):
                 result = yield self.do_jwt_login(login_submission)
                 defer.returnValue(result)
-            # TODO Delete this after all CAS clients switch to token login instead
-            elif self.cas_enabled and (login_submission["type"] ==
-                                       LoginRestServlet.CAS_TYPE):
-                uri = "%s/proxyValidate" % (self.cas_server_url,)
-                args = {
-                    "ticket": login_submission["ticket"],
-                    "service": login_submission["service"]
-                }
-                body = yield self.http_client.get_raw(uri, args)
-                result = yield self.do_cas_login(body)
-                defer.returnValue(result)
             elif login_submission["type"] == LoginRestServlet.TOKEN_TYPE:
                 result = yield self.do_token_login(login_submission)
                 defer.returnValue(result)
@@ -191,51 +177,6 @@ class LoginRestServlet(ClientV1RestServlet):
 
         defer.returnValue((200, result))
 
-    # TODO Delete this after all CAS clients switch to token login instead
-    @defer.inlineCallbacks
-    def do_cas_login(self, cas_response_body):
-        user, attributes = self.parse_cas_response(cas_response_body)
-
-        for required_attribute, required_value in self.cas_required_attributes.items():
-            # If required attribute was not in CAS Response - Forbidden
-            if required_attribute not in attributes:
-                raise LoginError(401, "Unauthorized", errcode=Codes.UNAUTHORIZED)
-
-            # Also need to check value
-            if required_value is not None:
-                actual_value = attributes[required_attribute]
-                # If required attribute value does not match expected - Forbidden
-                if required_value != actual_value:
-                    raise LoginError(401, "Unauthorized", errcode=Codes.UNAUTHORIZED)
-
-        user_id = UserID.create(user, self.hs.hostname).to_string()
-        auth_handler = self.auth_handler
-        registered_user_id = yield auth_handler.check_user_exists(user_id)
-        if registered_user_id:
-            access_token, refresh_token = (
-                yield auth_handler.get_login_tuple_for_user_id(
-                    registered_user_id
-                )
-            )
-            result = {
-                "user_id": registered_user_id,  # may have changed
-                "access_token": access_token,
-                "refresh_token": refresh_token,
-                "home_server": self.hs.hostname,
-            }
-
-        else:
-            user_id, access_token = (
-                yield self.handlers.registration_handler.register(localpart=user)
-            )
-            result = {
-                "user_id": user_id,  # may have changed
-                "access_token": access_token,
-                "home_server": self.hs.hostname,
-            }
-
-        defer.returnValue((200, result))
-
     @defer.inlineCallbacks
     def do_jwt_login(self, login_submission):
         token = login_submission.get("token", None)
@@ -293,33 +234,6 @@ class LoginRestServlet(ClientV1RestServlet):
 
         defer.returnValue((200, result))
 
-    # TODO Delete this after all CAS clients switch to token login instead
-    def parse_cas_response(self, cas_response_body):
-        root = ET.fromstring(cas_response_body)
-        if not root.tag.endswith("serviceResponse"):
-            raise LoginError(401, "Invalid CAS response", errcode=Codes.UNAUTHORIZED)
-        if not root[0].tag.endswith("authenticationSuccess"):
-            raise LoginError(401, "Unsuccessful CAS response", errcode=Codes.UNAUTHORIZED)
-        for child in root[0]:
-            if child.tag.endswith("user"):
-                user = child.text
-            if child.tag.endswith("attributes"):
-                attributes = {}
-                for attribute in child:
-                    # ElementTree library expands the namespace in attribute tags
-                    # to the full URL of the namespace.
-                    # See (https://docs.python.org/2/library/xml.etree.elementtree.html)
-                    # We don't care about namespace here and it will always be encased in
-                    # curly braces, so we remove them.
-                    if "}" in attribute.tag:
-                        attributes[attribute.tag.split("}")[1]] = attribute.text
-                    else:
-                        attributes[attribute.tag] = attribute.text
-        if user is None or attributes is None:
-            raise LoginError(401, "Invalid CAS response", errcode=Codes.UNAUTHORIZED)
-
-        return (user, attributes)
-
     def _register_device(self, user_id, login_submission):
         """Register a device for a user.
 
@@ -347,6 +261,7 @@ class SAML2RestServlet(ClientV1RestServlet):
     def __init__(self, hs):
         super(SAML2RestServlet, self).__init__(hs)
         self.sp_config = hs.config.saml2_config_path
+        self.handlers = hs.get_handlers()
 
     @defer.inlineCallbacks
     def on_POST(self, request):
@@ -384,18 +299,6 @@ class SAML2RestServlet(ClientV1RestServlet):
         defer.returnValue((200, {"status": "not_authenticated"}))
 
 
-# TODO Delete this after all CAS clients switch to token login instead
-class CasRestServlet(ClientV1RestServlet):
-    PATTERNS = client_path_patterns("/login/cas", releases=())
-
-    def __init__(self, hs):
-        super(CasRestServlet, self).__init__(hs)
-        self.cas_server_url = hs.config.cas_server_url
-
-    def on_GET(self, request):
-        return (200, {"serverUrl": self.cas_server_url})
-
-
 class CasRedirectServlet(ClientV1RestServlet):
     PATTERNS = client_path_patterns("/login/cas/redirect", releases=())
 
@@ -427,6 +330,8 @@ class CasTicketServlet(ClientV1RestServlet):
         self.cas_server_url = hs.config.cas_server_url
         self.cas_service_url = hs.config.cas_service_url
         self.cas_required_attributes = hs.config.cas_required_attributes
+        self.auth_handler = hs.get_auth_handler()
+        self.handlers = hs.get_handlers()
 
     @defer.inlineCallbacks
     def on_GET(self, request):
@@ -479,30 +384,39 @@ class CasTicketServlet(ClientV1RestServlet):
         return urlparse.urlunparse(url_parts)
 
     def parse_cas_response(self, cas_response_body):
-        root = ET.fromstring(cas_response_body)
-        if not root.tag.endswith("serviceResponse"):
-            raise LoginError(401, "Invalid CAS response", errcode=Codes.UNAUTHORIZED)
-        if not root[0].tag.endswith("authenticationSuccess"):
-            raise LoginError(401, "Unsuccessful CAS response", errcode=Codes.UNAUTHORIZED)
-        for child in root[0]:
-            if child.tag.endswith("user"):
-                user = child.text
-            if child.tag.endswith("attributes"):
-                attributes = {}
-                for attribute in child:
-                    # ElementTree library expands the namespace in attribute tags
-                    # to the full URL of the namespace.
-                    # See (https://docs.python.org/2/library/xml.etree.elementtree.html)
-                    # We don't care about namespace here and it will always be encased in
-                    # curly braces, so we remove them.
-                    if "}" in attribute.tag:
-                        attributes[attribute.tag.split("}")[1]] = attribute.text
-                    else:
-                        attributes[attribute.tag] = attribute.text
-        if user is None or attributes is None:
-            raise LoginError(401, "Invalid CAS response", errcode=Codes.UNAUTHORIZED)
-
-        return (user, attributes)
+        user = None
+        attributes = None
+        try:
+            root = ET.fromstring(cas_response_body)
+            if not root.tag.endswith("serviceResponse"):
+                raise Exception("root of CAS response is not serviceResponse")
+            success = (root[0].tag.endswith("authenticationSuccess"))
+            for child in root[0]:
+                if child.tag.endswith("user"):
+                    user = child.text
+                if child.tag.endswith("attributes"):
+                    attributes = {}
+                    for attribute in child:
+                        # ElementTree library expands the namespace in
+                        # attribute tags to the full URL of the namespace.
+                        # We don't care about namespace here and it will always
+                        # be encased in curly braces, so we remove them.
+                        tag = attribute.tag
+                        if "}" in tag:
+                            tag = tag.split("}")[1]
+                        attributes[tag] = attribute.text
+            if user is None:
+                raise Exception("CAS response does not contain user")
+            if attributes is None:
+                raise Exception("CAS response does not contain attributes")
+        except Exception:
+            logger.error("Error parsing CAS response", exc_info=1)
+            raise LoginError(401, "Invalid CAS response",
+                             errcode=Codes.UNAUTHORIZED)
+        if not success:
+            raise LoginError(401, "Unsuccessful CAS response",
+                             errcode=Codes.UNAUTHORIZED)
+        return user, attributes
 
 
 def register_servlets(hs, http_server):
@@ -512,5 +426,3 @@ def register_servlets(hs, http_server):
     if hs.config.cas_enabled:
         CasRedirectServlet(hs).register(http_server)
         CasTicketServlet(hs).register(http_server)
-        CasRestServlet(hs).register(http_server)
-    # TODO PasswordResetRestServlet(hs).register(http_server)
diff --git a/synapse/rest/client/v1/profile.py b/synapse/rest/client/v1/profile.py
index 65c4e2ebef..355e82474b 100644
--- a/synapse/rest/client/v1/profile.py
+++ b/synapse/rest/client/v1/profile.py
@@ -24,6 +24,10 @@ from synapse.http.servlet import parse_json_object_from_request
 class ProfileDisplaynameRestServlet(ClientV1RestServlet):
     PATTERNS = client_path_patterns("/profile/(?P<user_id>[^/]*)/displayname")
 
+    def __init__(self, hs):
+        super(ProfileDisplaynameRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     @defer.inlineCallbacks
     def on_GET(self, request, user_id):
         user = UserID.from_string(user_id)
@@ -62,6 +66,10 @@ class ProfileDisplaynameRestServlet(ClientV1RestServlet):
 class ProfileAvatarURLRestServlet(ClientV1RestServlet):
     PATTERNS = client_path_patterns("/profile/(?P<user_id>[^/]*)/avatar_url")
 
+    def __init__(self, hs):
+        super(ProfileAvatarURLRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     @defer.inlineCallbacks
     def on_GET(self, request, user_id):
         user = UserID.from_string(user_id)
@@ -99,6 +107,10 @@ class ProfileAvatarURLRestServlet(ClientV1RestServlet):
 class ProfileRestServlet(ClientV1RestServlet):
     PATTERNS = client_path_patterns("/profile/(?P<user_id>[^/]*)")
 
+    def __init__(self, hs):
+        super(ProfileRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     @defer.inlineCallbacks
     def on_GET(self, request, user_id):
         user = UserID.from_string(user_id)
diff --git a/synapse/rest/client/v1/register.py b/synapse/rest/client/v1/register.py
index 2383b9df86..71d58c8e8d 100644
--- a/synapse/rest/client/v1/register.py
+++ b/synapse/rest/client/v1/register.py
@@ -65,6 +65,7 @@ class RegisterRestServlet(ClientV1RestServlet):
         self.sessions = {}
         self.enable_registration = hs.config.enable_registration
         self.auth_handler = hs.get_auth_handler()
+        self.handlers = hs.get_handlers()
 
     def on_GET(self, request):
         if self.hs.config.enable_registration_captcha:
@@ -383,6 +384,7 @@ class CreateUserRestServlet(ClientV1RestServlet):
         super(CreateUserRestServlet, self).__init__(hs)
         self.store = hs.get_datastore()
         self.direct_user_creation_max_duration = hs.config.user_creation_max_duration
+        self.handlers = hs.get_handlers()
 
     @defer.inlineCallbacks
     def on_POST(self, request):
diff --git a/synapse/rest/client/v1/room.py b/synapse/rest/client/v1/room.py
index 866a1e9120..0d81757010 100644
--- a/synapse/rest/client/v1/room.py
+++ b/synapse/rest/client/v1/room.py
@@ -35,6 +35,10 @@ logger = logging.getLogger(__name__)
 class RoomCreateRestServlet(ClientV1RestServlet):
     # No PATTERN; we have custom dispatch rules here
 
+    def __init__(self, hs):
+        super(RoomCreateRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     def register(self, http_server):
         PATTERNS = "/createRoom"
         register_txn_path(self, PATTERNS, http_server)
@@ -82,6 +86,10 @@ class RoomCreateRestServlet(ClientV1RestServlet):
 
 # TODO: Needs unit testing for generic events
 class RoomStateEventRestServlet(ClientV1RestServlet):
+    def __init__(self, hs):
+        super(RoomStateEventRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     def register(self, http_server):
         # /room/$roomid/state/$eventtype
         no_state_key = "/rooms/(?P<room_id>[^/]*)/state/(?P<event_type>[^/]*)$"
@@ -166,6 +174,10 @@ class RoomStateEventRestServlet(ClientV1RestServlet):
 # TODO: Needs unit testing for generic events + feedback
 class RoomSendEventRestServlet(ClientV1RestServlet):
 
+    def __init__(self, hs):
+        super(RoomSendEventRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     def register(self, http_server):
         # /rooms/$roomid/send/$event_type[/$txn_id]
         PATTERNS = ("/rooms/(?P<room_id>[^/]*)/send/(?P<event_type>[^/]*)")
@@ -210,6 +222,9 @@ class RoomSendEventRestServlet(ClientV1RestServlet):
 
 # TODO: Needs unit testing for room ID + alias joins
 class JoinRoomAliasServlet(ClientV1RestServlet):
+    def __init__(self, hs):
+        super(JoinRoomAliasServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
 
     def register(self, http_server):
         # /join/$room_identifier[/$txn_id]
@@ -253,6 +268,7 @@ class JoinRoomAliasServlet(ClientV1RestServlet):
             action="join",
             txn_id=txn_id,
             remote_room_hosts=remote_room_hosts,
+            content=content,
             third_party_signed=content.get("third_party_signed", None),
         )
 
@@ -296,6 +312,10 @@ class PublicRoomListRestServlet(ClientV1RestServlet):
 class RoomMemberListRestServlet(ClientV1RestServlet):
     PATTERNS = client_path_patterns("/rooms/(?P<room_id>[^/]*)/members$")
 
+    def __init__(self, hs):
+        super(RoomMemberListRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     @defer.inlineCallbacks
     def on_GET(self, request, room_id):
         # TODO support Pagination stream API (limit/tokens)
@@ -322,6 +342,10 @@ class RoomMemberListRestServlet(ClientV1RestServlet):
 class RoomMessageListRestServlet(ClientV1RestServlet):
     PATTERNS = client_path_patterns("/rooms/(?P<room_id>[^/]*)/messages$")
 
+    def __init__(self, hs):
+        super(RoomMessageListRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     @defer.inlineCallbacks
     def on_GET(self, request, room_id):
         requester = yield self.auth.get_user_by_req(request, allow_guest=True)
@@ -351,6 +375,10 @@ class RoomMessageListRestServlet(ClientV1RestServlet):
 class RoomStateRestServlet(ClientV1RestServlet):
     PATTERNS = client_path_patterns("/rooms/(?P<room_id>[^/]*)/state$")
 
+    def __init__(self, hs):
+        super(RoomStateRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     @defer.inlineCallbacks
     def on_GET(self, request, room_id):
         requester = yield self.auth.get_user_by_req(request, allow_guest=True)
@@ -368,6 +396,10 @@ class RoomStateRestServlet(ClientV1RestServlet):
 class RoomInitialSyncRestServlet(ClientV1RestServlet):
     PATTERNS = client_path_patterns("/rooms/(?P<room_id>[^/]*)/initialSync$")
 
+    def __init__(self, hs):
+        super(RoomInitialSyncRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     @defer.inlineCallbacks
     def on_GET(self, request, room_id):
         requester = yield self.auth.get_user_by_req(request, allow_guest=True)
@@ -388,6 +420,7 @@ class RoomEventContext(ClientV1RestServlet):
     def __init__(self, hs):
         super(RoomEventContext, self).__init__(hs)
         self.clock = hs.get_clock()
+        self.handlers = hs.get_handlers()
 
     @defer.inlineCallbacks
     def on_GET(self, request, room_id, event_id):
@@ -424,6 +457,10 @@ class RoomEventContext(ClientV1RestServlet):
 
 
 class RoomForgetRestServlet(ClientV1RestServlet):
+    def __init__(self, hs):
+        super(RoomForgetRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     def register(self, http_server):
         PATTERNS = ("/rooms/(?P<room_id>[^/]*)/forget")
         register_txn_path(self, PATTERNS, http_server)
@@ -462,6 +499,10 @@ class RoomForgetRestServlet(ClientV1RestServlet):
 # TODO: Needs unit testing
 class RoomMembershipRestServlet(ClientV1RestServlet):
 
+    def __init__(self, hs):
+        super(RoomMembershipRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     def register(self, http_server):
         # /rooms/$roomid/[invite|join|leave]
         PATTERNS = ("/rooms/(?P<room_id>[^/]*)/"
@@ -542,6 +583,10 @@ class RoomMembershipRestServlet(ClientV1RestServlet):
 
 
 class RoomRedactEventRestServlet(ClientV1RestServlet):
+    def __init__(self, hs):
+        super(RoomRedactEventRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     def register(self, http_server):
         PATTERNS = ("/rooms/(?P<room_id>[^/]*)/redact/(?P<event_id>[^/]*)")
         register_txn_path(self, PATTERNS, http_server)
@@ -624,6 +669,10 @@ class SearchRestServlet(ClientV1RestServlet):
         "/search$"
     )
 
+    def __init__(self, hs):
+        super(SearchRestServlet, self).__init__(hs)
+        self.handlers = hs.get_handlers()
+
     @defer.inlineCallbacks
     def on_POST(self, request):
         requester = yield self.auth.get_user_by_req(request)
diff --git a/synapse/rest/client/v2_alpha/notifications.py b/synapse/rest/client/v2_alpha/notifications.py
new file mode 100644
index 0000000000..f1a48acf07
--- /dev/null
+++ b/synapse/rest/client/v2_alpha/notifications.py
@@ -0,0 +1,99 @@
+# -*- coding: utf-8 -*-
+# Copyright 2016 OpenMarket Ltd
+#
+# 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.
+
+from twisted.internet import defer
+
+from synapse.http.servlet import (
+    RestServlet, parse_string, parse_integer
+)
+from synapse.events.utils import (
+    serialize_event, format_event_for_client_v2_without_room_id,
+)
+
+from ._base import client_v2_patterns
+
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+class NotificationsServlet(RestServlet):
+    PATTERNS = client_v2_patterns("/notifications$", releases=())
+
+    def __init__(self, hs):
+        super(NotificationsServlet, self).__init__()
+        self.store = hs.get_datastore()
+        self.auth = hs.get_auth()
+        self.clock = hs.get_clock()
+
+    @defer.inlineCallbacks
+    def on_GET(self, request):
+        requester = yield self.auth.get_user_by_req(request)
+        user_id = requester.user.to_string()
+
+        from_token = parse_string(request, "from", required=False)
+        limit = parse_integer(request, "limit", default=50)
+
+        limit = min(limit, 500)
+
+        push_actions = yield self.store.get_push_actions_for_user(
+            user_id, from_token, limit
+        )
+
+        receipts_by_room = yield self.store.get_receipts_for_user_with_orderings(
+            user_id, 'm.read'
+        )
+
+        notif_event_ids = [pa["event_id"] for pa in push_actions]
+        notif_events = yield self.store.get_events(notif_event_ids)
+
+        returned_push_actions = []
+
+        next_token = None
+
+        for pa in push_actions:
+            returned_pa = {
+                "room_id": pa["room_id"],
+                "profile_tag": pa["profile_tag"],
+                "actions": pa["actions"],
+                "ts": pa["received_ts"],
+                "event": serialize_event(
+                    notif_events[pa["event_id"]],
+                    self.clock.time_msec(),
+                    event_format=format_event_for_client_v2_without_room_id,
+                ),
+            }
+
+            if pa["room_id"] not in receipts_by_room:
+                returned_pa["read"] = False
+            else:
+                receipt = receipts_by_room[pa["room_id"]]
+
+                returned_pa["read"] = (
+                    receipt["topological_ordering"], receipt["stream_ordering"]
+                ) >= (
+                    pa["topological_ordering"], pa["stream_ordering"]
+                )
+            returned_push_actions.append(returned_pa)
+            next_token = pa["stream_ordering"]
+
+        defer.returnValue((200, {
+            "notifications": returned_push_actions,
+            "next_token": next_token,
+        }))
+
+
+def register_servlets(hs, http_server):
+    NotificationsServlet(hs).register(http_server)
diff --git a/synapse/rest/client/v2_alpha/register.py b/synapse/rest/client/v2_alpha/register.py
index 943f5676a3..2121bd75ea 100644
--- a/synapse/rest/client/v2_alpha/register.py
+++ b/synapse/rest/client/v2_alpha/register.py
@@ -403,10 +403,9 @@ class RegisterRestServlet(RestServlet):
         # register the user's device
         device_id = params.get("device_id")
         initial_display_name = params.get("initial_device_display_name")
-        device_id = self.device_handler.check_device_registered(
+        return self.device_handler.check_device_registered(
             user_id, device_id, initial_display_name
         )
-        return device_id
 
     @defer.inlineCallbacks
     def _do_guest_registration(self):
diff --git a/synapse/rest/client/v2_alpha/sync.py b/synapse/rest/client/v2_alpha/sync.py
index 43d8e0bf39..b11acdbea7 100644
--- a/synapse/rest/client/v2_alpha/sync.py
+++ b/synapse/rest/client/v2_alpha/sync.py
@@ -146,7 +146,7 @@ class SyncRestServlet(RestServlet):
         affect_presence = set_presence != PresenceState.OFFLINE
 
         if affect_presence:
-            yield self.presence_handler.set_state(user, {"presence": set_presence})
+            yield self.presence_handler.set_state(user, {"presence": set_presence}, True)
 
         context = yield self.presence_handler.user_syncing(
             user.to_string(), affect_presence=affect_presence,
diff --git a/synapse/rest/client/v2_alpha/thirdparty.py b/synapse/rest/client/v2_alpha/thirdparty.py
new file mode 100644
index 0000000000..9abca3a8ad
--- /dev/null
+++ b/synapse/rest/client/v2_alpha/thirdparty.py
@@ -0,0 +1,78 @@
+# -*- coding: utf-8 -*-
+# Copyright 2015, 2016 OpenMarket Ltd
+#
+# 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.
+
+
+import logging
+
+from twisted.internet import defer
+
+from synapse.http.servlet import RestServlet
+from synapse.types import ThirdPartyEntityKind
+from ._base import client_v2_patterns
+
+logger = logging.getLogger(__name__)
+
+
+class ThirdPartyUserServlet(RestServlet):
+    PATTERNS = client_v2_patterns("/3pu(/(?P<protocol>[^/]+))?$",
+                                  releases=())
+
+    def __init__(self, hs):
+        super(ThirdPartyUserServlet, self).__init__()
+
+        self.auth = hs.get_auth()
+        self.appservice_handler = hs.get_application_service_handler()
+
+    @defer.inlineCallbacks
+    def on_GET(self, request, protocol):
+        yield self.auth.get_user_by_req(request)
+
+        fields = request.args
+        del fields["access_token"]
+
+        results = yield self.appservice_handler.query_3pe(
+            ThirdPartyEntityKind.USER, protocol, fields
+        )
+
+        defer.returnValue((200, results))
+
+
+class ThirdPartyLocationServlet(RestServlet):
+    PATTERNS = client_v2_patterns("/3pl(/(?P<protocol>[^/]+))?$",
+                                  releases=())
+
+    def __init__(self, hs):
+        super(ThirdPartyLocationServlet, self).__init__()
+
+        self.auth = hs.get_auth()
+        self.appservice_handler = hs.get_application_service_handler()
+
+    @defer.inlineCallbacks
+    def on_GET(self, request, protocol):
+        yield self.auth.get_user_by_req(request)
+
+        fields = request.args
+        del fields["access_token"]
+
+        results = yield self.appservice_handler.query_3pe(
+            ThirdPartyEntityKind.LOCATION, protocol, fields
+        )
+
+        defer.returnValue((200, results))
+
+
+def register_servlets(hs, http_server):
+    ThirdPartyUserServlet(hs).register(http_server)
+    ThirdPartyLocationServlet(hs).register(http_server)
diff --git a/synapse/rest/key/v2/remote_key_resource.py b/synapse/rest/key/v2/remote_key_resource.py
index 7209d5a37d..9fe2013657 100644
--- a/synapse/rest/key/v2/remote_key_resource.py
+++ b/synapse/rest/key/v2/remote_key_resource.py
@@ -15,6 +15,7 @@
 from synapse.http.server import request_handler, respond_with_json_bytes
 from synapse.http.servlet import parse_integer, parse_json_object_from_request
 from synapse.api.errors import SynapseError, Codes
+from synapse.crypto.keyring import KeyLookupError
 
 from twisted.web.resource import Resource
 from twisted.web.server import NOT_DONE_YET
@@ -210,9 +211,10 @@ class RemoteKey(Resource):
                     yield self.keyring.get_server_verify_key_v2_direct(
                         server_name, key_ids
                     )
+                except KeyLookupError as e:
+                    logger.info("Failed to fetch key: %s", e)
                 except:
                     logger.exception("Failed to get key for %r", server_name)
-                    pass
             yield self.query_keys(
                 request, query, query_remote_on_cache_miss=False
             )
diff --git a/synapse/rest/media/v1/download_resource.py b/synapse/rest/media/v1/download_resource.py
index 9f69620772..9f0625a822 100644
--- a/synapse/rest/media/v1/download_resource.py
+++ b/synapse/rest/media/v1/download_resource.py
@@ -45,6 +45,7 @@ class DownloadResource(Resource):
     @request_handler()
     @defer.inlineCallbacks
     def _async_render_GET(self, request):
+        request.setHeader("Content-Security-Policy", "sandbox")
         server_name, media_id, name = parse_media_id(request)
         if server_name == self.server_name:
             yield self._respond_local_file(request, media_id, name)
diff --git a/synapse/rest/media/v1/preview_url_resource.py b/synapse/rest/media/v1/preview_url_resource.py
index bdd0e60c5b..33f35fb44e 100644
--- a/synapse/rest/media/v1/preview_url_resource.py
+++ b/synapse/rest/media/v1/preview_url_resource.py
@@ -29,14 +29,13 @@ from synapse.http.server import (
 from synapse.util.async import ObservableDeferred
 from synapse.util.stringutils import is_ascii
 
-from copy import deepcopy
-
 import os
 import re
 import fnmatch
 import cgi
 import ujson as json
 import urlparse
+import itertools
 
 import logging
 logger = logging.getLogger(__name__)
@@ -163,7 +162,7 @@ class PreviewUrlResource(Resource):
 
         logger.debug("got media_info of '%s'" % media_info)
 
-        if self._is_media(media_info['media_type']):
+        if _is_media(media_info['media_type']):
             dims = yield self.media_repo._generate_local_thumbnails(
                 media_info['filesystem_id'], media_info
             )
@@ -184,11 +183,9 @@ class PreviewUrlResource(Resource):
                 logger.warn("Couldn't get dims for %s" % url)
 
             # define our OG response for this media
-        elif self._is_html(media_info['media_type']):
+        elif _is_html(media_info['media_type']):
             # TODO: somehow stop a big HTML tree from exploding synapse's RAM
 
-            from lxml import etree
-
             file = open(media_info['filename'])
             body = file.read()
             file.close()
@@ -199,17 +196,35 @@ class PreviewUrlResource(Resource):
             match = re.match(r'.*; *charset=(.*?)(;|$)', media_info['media_type'], re.I)
             encoding = match.group(1) if match else "utf-8"
 
-            try:
-                parser = etree.HTMLParser(recover=True, encoding=encoding)
-                tree = etree.fromstring(body, parser)
-                og = yield self._calc_og(tree, media_info, requester)
-            except UnicodeDecodeError:
-                # blindly try decoding the body as utf-8, which seems to fix
-                # the charset mismatches on https://google.com
-                parser = etree.HTMLParser(recover=True, encoding=encoding)
-                tree = etree.fromstring(body.decode('utf-8', 'ignore'), parser)
-                og = yield self._calc_og(tree, media_info, requester)
+            og = decode_and_calc_og(body, media_info['uri'], encoding)
+
+            # pre-cache the image for posterity
+            # FIXME: it might be cleaner to use the same flow as the main /preview_url
+            # request itself and benefit from the same caching etc.  But for now we
+            # just rely on the caching on the master request to speed things up.
+            if 'og:image' in og and og['og:image']:
+                image_info = yield self._download_url(
+                    _rebase_url(og['og:image'], media_info['uri']), requester.user
+                )
 
+                if _is_media(image_info['media_type']):
+                    # TODO: make sure we don't choke on white-on-transparent images
+                    dims = yield self.media_repo._generate_local_thumbnails(
+                        image_info['filesystem_id'], image_info
+                    )
+                    if dims:
+                        og["og:image:width"] = dims['width']
+                        og["og:image:height"] = dims['height']
+                    else:
+                        logger.warn("Couldn't get dims for %s" % og["og:image"])
+
+                    og["og:image"] = "mxc://%s/%s" % (
+                        self.server_name, image_info['filesystem_id']
+                    )
+                    og["og:image:type"] = image_info['media_type']
+                    og["matrix:image:size"] = image_info['media_length']
+                else:
+                    del og["og:image"]
         else:
             logger.warn("Failed to find any OG data in %s", url)
             og = {}
@@ -233,139 +248,6 @@ class PreviewUrlResource(Resource):
         respond_with_json_bytes(request, 200, json.dumps(og), send_cors=True)
 
     @defer.inlineCallbacks
-    def _calc_og(self, tree, media_info, requester):
-        # suck our tree into lxml and define our OG response.
-
-        # if we see any image URLs in the OG response, then spider them
-        # (although the client could choose to do this by asking for previews of those
-        # URLs to avoid DoSing the server)
-
-        # "og:type"         : "video",
-        # "og:url"          : "https://www.youtube.com/watch?v=LXDBoHyjmtw",
-        # "og:site_name"    : "YouTube",
-        # "og:video:type"   : "application/x-shockwave-flash",
-        # "og:description"  : "Fun stuff happening here",
-        # "og:title"        : "RemoteJam - Matrix team hack for Disrupt Europe Hackathon",
-        # "og:image"        : "https://i.ytimg.com/vi/LXDBoHyjmtw/maxresdefault.jpg",
-        # "og:video:url"    : "http://www.youtube.com/v/LXDBoHyjmtw?version=3&autohide=1",
-        # "og:video:width"  : "1280"
-        # "og:video:height" : "720",
-        # "og:video:secure_url": "https://www.youtube.com/v/LXDBoHyjmtw?version=3",
-
-        og = {}
-        for tag in tree.xpath("//*/meta[starts-with(@property, 'og:')]"):
-            if 'content' in tag.attrib:
-                og[tag.attrib['property']] = tag.attrib['content']
-
-        # TODO: grab article: meta tags too, e.g.:
-
-        # "article:publisher" : "https://www.facebook.com/thethudonline" />
-        # "article:author" content="https://www.facebook.com/thethudonline" />
-        # "article:tag" content="baby" />
-        # "article:section" content="Breaking News" />
-        # "article:published_time" content="2016-03-31T19:58:24+00:00" />
-        # "article:modified_time" content="2016-04-01T18:31:53+00:00" />
-
-        if 'og:title' not in og:
-            # do some basic spidering of the HTML
-            title = tree.xpath("(//title)[1] | (//h1)[1] | (//h2)[1] | (//h3)[1]")
-            og['og:title'] = title[0].text.strip() if title else None
-
-        if 'og:image' not in og:
-            # TODO: extract a favicon failing all else
-            meta_image = tree.xpath(
-                "//*/meta[translate(@itemprop, 'IMAGE', 'image')='image']/@content"
-            )
-            if meta_image:
-                og['og:image'] = self._rebase_url(meta_image[0], media_info['uri'])
-            else:
-                # TODO: consider inlined CSS styles as well as width & height attribs
-                images = tree.xpath("//img[@src][number(@width)>10][number(@height)>10]")
-                images = sorted(images, key=lambda i: (
-                    -1 * float(i.attrib['width']) * float(i.attrib['height'])
-                ))
-                if not images:
-                    images = tree.xpath("//img[@src]")
-                if images:
-                    og['og:image'] = images[0].attrib['src']
-
-        # pre-cache the image for posterity
-        # FIXME: it might be cleaner to use the same flow as the main /preview_url
-        # request itself and benefit from the same caching etc.  But for now we
-        # just rely on the caching on the master request to speed things up.
-        if 'og:image' in og and og['og:image']:
-            image_info = yield self._download_url(
-                self._rebase_url(og['og:image'], media_info['uri']), requester.user
-            )
-
-            if self._is_media(image_info['media_type']):
-                # TODO: make sure we don't choke on white-on-transparent images
-                dims = yield self.media_repo._generate_local_thumbnails(
-                    image_info['filesystem_id'], image_info
-                )
-                if dims:
-                    og["og:image:width"] = dims['width']
-                    og["og:image:height"] = dims['height']
-                else:
-                    logger.warn("Couldn't get dims for %s" % og["og:image"])
-
-                og["og:image"] = "mxc://%s/%s" % (
-                    self.server_name, image_info['filesystem_id']
-                )
-                og["og:image:type"] = image_info['media_type']
-                og["matrix:image:size"] = image_info['media_length']
-            else:
-                del og["og:image"]
-
-        if 'og:description' not in og:
-            meta_description = tree.xpath(
-                "//*/meta"
-                "[translate(@name, 'DESCRIPTION', 'description')='description']"
-                "/@content")
-            if meta_description:
-                og['og:description'] = meta_description[0]
-            else:
-                # grab any text nodes which are inside the <body/> tag...
-                # unless they are within an HTML5 semantic markup tag...
-                # <header/>, <nav/>, <aside/>, <footer/>
-                # ...or if they are within a <script/> or <style/> tag.
-                # This is a very very very coarse approximation to a plain text
-                # render of the page.
-
-                # We don't just use XPATH here as that is slow on some machines.
-
-                # We clone `tree` as we modify it.
-                cloned_tree = deepcopy(tree.find("body"))
-
-                TAGS_TO_REMOVE = ("header", "nav", "aside", "footer", "script", "style",)
-                for el in cloned_tree.iter(TAGS_TO_REMOVE):
-                    el.getparent().remove(el)
-
-                # Split all the text nodes into paragraphs (by splitting on new
-                # lines)
-                text_nodes = (
-                    re.sub(r'\s+', '\n', el.text).strip()
-                    for el in cloned_tree.iter()
-                    if el.text and isinstance(el.tag, basestring)  # Removes comments
-                )
-                og['og:description'] = summarize_paragraphs(text_nodes)
-
-        # TODO: delete the url downloads to stop diskfilling,
-        # as we only ever cared about its OG
-        defer.returnValue(og)
-
-    def _rebase_url(self, url, base):
-        base = list(urlparse.urlparse(base))
-        url = list(urlparse.urlparse(url))
-        if not url[0]:  # fix up schema
-            url[0] = base[0] or "http"
-        if not url[1]:  # fix up hostname
-            url[1] = base[1]
-            if not url[2].startswith('/'):
-                url[2] = re.sub(r'/[^/]+$', '/', base[2]) + url[2]
-        return urlparse.urlunparse(url)
-
-    @defer.inlineCallbacks
     def _download_url(self, url, user):
         # TODO: we should probably honour robots.txt... except in practice
         # we're most likely being explicitly triggered by a human rather than a
@@ -445,17 +327,171 @@ class PreviewUrlResource(Resource):
             "etag": headers["ETag"][0] if "ETag" in headers else None,
         })
 
-    def _is_media(self, content_type):
-        if content_type.lower().startswith("image/"):
-            return True
 
-    def _is_html(self, content_type):
-        content_type = content_type.lower()
-        if (
-            content_type.startswith("text/html") or
-            content_type.startswith("application/xhtml")
-        ):
-            return True
+def decode_and_calc_og(body, media_uri, request_encoding=None):
+    from lxml import etree
+
+    try:
+        parser = etree.HTMLParser(recover=True, encoding=request_encoding)
+        tree = etree.fromstring(body, parser)
+        og = _calc_og(tree, media_uri)
+    except UnicodeDecodeError:
+        # blindly try decoding the body as utf-8, which seems to fix
+        # the charset mismatches on https://google.com
+        parser = etree.HTMLParser(recover=True, encoding=request_encoding)
+        tree = etree.fromstring(body.decode('utf-8', 'ignore'), parser)
+        og = _calc_og(tree, media_uri)
+
+    return og
+
+
+def _calc_og(tree, media_uri):
+    # suck our tree into lxml and define our OG response.
+
+    # if we see any image URLs in the OG response, then spider them
+    # (although the client could choose to do this by asking for previews of those
+    # URLs to avoid DoSing the server)
+
+    # "og:type"         : "video",
+    # "og:url"          : "https://www.youtube.com/watch?v=LXDBoHyjmtw",
+    # "og:site_name"    : "YouTube",
+    # "og:video:type"   : "application/x-shockwave-flash",
+    # "og:description"  : "Fun stuff happening here",
+    # "og:title"        : "RemoteJam - Matrix team hack for Disrupt Europe Hackathon",
+    # "og:image"        : "https://i.ytimg.com/vi/LXDBoHyjmtw/maxresdefault.jpg",
+    # "og:video:url"    : "http://www.youtube.com/v/LXDBoHyjmtw?version=3&autohide=1",
+    # "og:video:width"  : "1280"
+    # "og:video:height" : "720",
+    # "og:video:secure_url": "https://www.youtube.com/v/LXDBoHyjmtw?version=3",
+
+    og = {}
+    for tag in tree.xpath("//*/meta[starts-with(@property, 'og:')]"):
+        if 'content' in tag.attrib:
+            og[tag.attrib['property']] = tag.attrib['content']
+
+    # TODO: grab article: meta tags too, e.g.:
+
+    # "article:publisher" : "https://www.facebook.com/thethudonline" />
+    # "article:author" content="https://www.facebook.com/thethudonline" />
+    # "article:tag" content="baby" />
+    # "article:section" content="Breaking News" />
+    # "article:published_time" content="2016-03-31T19:58:24+00:00" />
+    # "article:modified_time" content="2016-04-01T18:31:53+00:00" />
+
+    if 'og:title' not in og:
+        # do some basic spidering of the HTML
+        title = tree.xpath("(//title)[1] | (//h1)[1] | (//h2)[1] | (//h3)[1]")
+        og['og:title'] = title[0].text.strip() if title else None
+
+    if 'og:image' not in og:
+        # TODO: extract a favicon failing all else
+        meta_image = tree.xpath(
+            "//*/meta[translate(@itemprop, 'IMAGE', 'image')='image']/@content"
+        )
+        if meta_image:
+            og['og:image'] = _rebase_url(meta_image[0], media_uri)
+        else:
+            # TODO: consider inlined CSS styles as well as width & height attribs
+            images = tree.xpath("//img[@src][number(@width)>10][number(@height)>10]")
+            images = sorted(images, key=lambda i: (
+                -1 * float(i.attrib['width']) * float(i.attrib['height'])
+            ))
+            if not images:
+                images = tree.xpath("//img[@src]")
+            if images:
+                og['og:image'] = images[0].attrib['src']
+
+    if 'og:description' not in og:
+        meta_description = tree.xpath(
+            "//*/meta"
+            "[translate(@name, 'DESCRIPTION', 'description')='description']"
+            "/@content")
+        if meta_description:
+            og['og:description'] = meta_description[0]
+        else:
+            # grab any text nodes which are inside the <body/> tag...
+            # unless they are within an HTML5 semantic markup tag...
+            # <header/>, <nav/>, <aside/>, <footer/>
+            # ...or if they are within a <script/> or <style/> tag.
+            # This is a very very very coarse approximation to a plain text
+            # render of the page.
+
+            # We don't just use XPATH here as that is slow on some machines.
+
+            from lxml import etree
+
+            TAGS_TO_REMOVE = (
+                "header", "nav", "aside", "footer", "script", "style", etree.Comment
+            )
+
+            # Split all the text nodes into paragraphs (by splitting on new
+            # lines)
+            text_nodes = (
+                re.sub(r'\s+', '\n', el).strip()
+                for el in _iterate_over_text(tree.find("body"), *TAGS_TO_REMOVE)
+            )
+            og['og:description'] = summarize_paragraphs(text_nodes)
+
+    # TODO: delete the url downloads to stop diskfilling,
+    # as we only ever cared about its OG
+    return og
+
+
+def _iterate_over_text(tree, *tags_to_ignore):
+    """Iterate over the tree returning text nodes in a depth first fashion,
+    skipping text nodes inside certain tags.
+    """
+    # This is basically a stack that we extend using itertools.chain.
+    # This will either consist of an element to iterate over *or* a string
+    # to be returned.
+    elements = iter([tree])
+    while True:
+        el = elements.next()
+        if isinstance(el, basestring):
+            yield el
+        elif el is not None and el.tag not in tags_to_ignore:
+            # el.text is the text before the first child, so we can immediately
+            # return it if the text exists.
+            if el.text:
+                yield el.text
+
+            # We add to the stack all the elements children, interspersed with
+            # each child's tail text (if it exists). The tail text of a node
+            # is text that comes *after* the node, so we always include it even
+            # if we ignore the child node.
+            elements = itertools.chain(
+                itertools.chain.from_iterable(  # Basically a flatmap
+                    [child, child.tail] if child.tail else [child]
+                    for child in el.iterchildren()
+                ),
+                elements
+            )
+
+
+def _rebase_url(url, base):
+    base = list(urlparse.urlparse(base))
+    url = list(urlparse.urlparse(url))
+    if not url[0]:  # fix up schema
+        url[0] = base[0] or "http"
+    if not url[1]:  # fix up hostname
+        url[1] = base[1]
+        if not url[2].startswith('/'):
+            url[2] = re.sub(r'/[^/]+$', '/', base[2]) + url[2]
+    return urlparse.urlunparse(url)
+
+
+def _is_media(content_type):
+    if content_type.lower().startswith("image/"):
+        return True
+
+
+def _is_html(content_type):
+    content_type = content_type.lower()
+    if (
+        content_type.startswith("text/html") or
+        content_type.startswith("application/xhtml")
+    ):
+        return True
 
 
 def summarize_paragraphs(text_nodes, min_size=200, max_size=500):