From 4283bd1cf9c3da2157c3642a7c4f105e9fac2636 Mon Sep 17 00:00:00 2001 From: Patrick Cloke Date: Mon, 17 Oct 2022 11:32:11 -0400 Subject: [PATCH] Support filtering the /messages API by relation type (MSC3874). (#14148) Gated behind an experimental configuration flag. --- changelog.d/14148.feature | 1 + synapse/api/filtering.py | 27 ++++- synapse/config/experimental.py | 3 + synapse/rest/client/versions.py | 2 + synapse/storage/databases/main/stream.py | 29 ++++- tests/api/test_filtering.py | 63 +++++++++- tests/rest/client/test_relations.py | 1 - tests/rest/client/test_rooms.py | 145 ++--------------------- tests/storage/test_stream.py | 118 ++++++++++++------ 9 files changed, 212 insertions(+), 177 deletions(-) create mode 100644 changelog.d/14148.feature diff --git a/changelog.d/14148.feature b/changelog.d/14148.feature new file mode 100644 index 0000000000..951d0cac80 --- /dev/null +++ b/changelog.d/14148.feature @@ -0,0 +1 @@ +Experimental support for [MSC3874](https://github.com/matrix-org/matrix-spec-proposals/pull/3874). diff --git a/synapse/api/filtering.py b/synapse/api/filtering.py index cc31cf8cc7..26be377d03 100644 --- a/synapse/api/filtering.py +++ b/synapse/api/filtering.py @@ -36,7 +36,7 @@ from jsonschema import FormatChecker from synapse.api.constants import EduTypes, EventContentFields from synapse.api.errors import SynapseError from synapse.api.presence import UserPresenceState -from synapse.events import EventBase +from synapse.events import EventBase, relation_from_event from synapse.types import JsonDict, RoomID, UserID if TYPE_CHECKING: @@ -53,6 +53,12 @@ FILTER_SCHEMA = { # check types are valid event types "types": {"type": "array", "items": {"type": "string"}}, "not_types": {"type": "array", "items": {"type": "string"}}, + # MSC3874, filtering /messages. + "org.matrix.msc3874.rel_types": {"type": "array", "items": {"type": "string"}}, + "org.matrix.msc3874.not_rel_types": { + "type": "array", + "items": {"type": "string"}, + }, }, } @@ -334,8 +340,15 @@ class Filter: self.labels = filter_json.get("org.matrix.labels", None) self.not_labels = filter_json.get("org.matrix.not_labels", []) - self.related_by_senders = self.filter_json.get("related_by_senders", None) - self.related_by_rel_types = self.filter_json.get("related_by_rel_types", None) + self.related_by_senders = filter_json.get("related_by_senders", None) + self.related_by_rel_types = filter_json.get("related_by_rel_types", None) + + # For compatibility with _check_fields. + self.rel_types = None + self.not_rel_types = [] + if hs.config.experimental.msc3874_enabled: + self.rel_types = filter_json.get("org.matrix.msc3874.rel_types", None) + self.not_rel_types = filter_json.get("org.matrix.msc3874.not_rel_types", []) def filters_all_types(self) -> bool: return "*" in self.not_types @@ -386,11 +399,19 @@ class Filter: # check if there is a string url field in the content for filtering purposes labels = content.get(EventContentFields.LABELS, []) + # Check if the event has a relation. + rel_type = None + if isinstance(event, EventBase): + relation = relation_from_event(event) + if relation: + rel_type = relation.rel_type + field_matchers = { "rooms": lambda v: room_id == v, "senders": lambda v: sender == v, "types": lambda v: _matches_wildcard(ev_type, v), "labels": lambda v: v in labels, + "rel_types": lambda v: rel_type == v, } result = self._check_fields(field_matchers) diff --git a/synapse/config/experimental.py b/synapse/config/experimental.py index f44655516e..f9a49451d8 100644 --- a/synapse/config/experimental.py +++ b/synapse/config/experimental.py @@ -117,3 +117,6 @@ class ExperimentalConfig(Config): self.msc3882_token_timeout = self.parse_duration( experimental.get("msc3882_token_timeout", "5m") ) + + # MSC3874: Filtering /messages with rel_types / not_rel_types. + self.msc3874_enabled: bool = experimental.get("msc3874_enabled", False) diff --git a/synapse/rest/client/versions.py b/synapse/rest/client/versions.py index 4e1fd2bbe7..4b87ee978a 100644 --- a/synapse/rest/client/versions.py +++ b/synapse/rest/client/versions.py @@ -114,6 +114,8 @@ class VersionsRestServlet(RestServlet): "org.matrix.msc3882": self.config.experimental.msc3882_enabled, # Adds support for remotely enabling/disabling pushers, as per MSC3881 "org.matrix.msc3881": self.config.experimental.msc3881_enabled, + # Adds support for filtering /messages by event relation. + "org.matrix.msc3874": self.config.experimental.msc3874_enabled, }, }, ) diff --git a/synapse/storage/databases/main/stream.py b/synapse/storage/databases/main/stream.py index 5baffbfe55..09ce855aa8 100644 --- a/synapse/storage/databases/main/stream.py +++ b/synapse/storage/databases/main/stream.py @@ -357,6 +357,24 @@ def filter_to_clause(event_filter: Optional[Filter]) -> Tuple[str, List[str]]: ) args.extend(event_filter.related_by_rel_types) + if event_filter.rel_types: + clauses.append( + "(%s)" + % " OR ".join( + "event_relation.relation_type = ?" for _ in event_filter.rel_types + ) + ) + args.extend(event_filter.rel_types) + + if event_filter.not_rel_types: + clauses.append( + "((%s) OR event_relation.relation_type IS NULL)" + % " AND ".join( + "event_relation.relation_type != ?" for _ in event_filter.not_rel_types + ) + ) + args.extend(event_filter.not_rel_types) + return " AND ".join(clauses), args @@ -1278,8 +1296,8 @@ class StreamWorkerStore(EventsWorkerStore, SQLBaseStore): # Multiple labels could cause the same event to appear multiple times. needs_distinct = True - # If there is a filter on relation_senders and relation_types join to the - # relations table. + # If there is a relation_senders and relation_types filter join to the + # relations table to get events related to the current event. if event_filter and ( event_filter.related_by_senders or event_filter.related_by_rel_types ): @@ -1294,6 +1312,13 @@ class StreamWorkerStore(EventsWorkerStore, SQLBaseStore): LEFT JOIN events AS related_event ON (relation.event_id = related_event.event_id) """ + # If there is a not_rel_types filter join to the relations table to get + # the event's relation information. + if event_filter and (event_filter.rel_types or event_filter.not_rel_types): + join_clause += """ + LEFT JOIN event_relations AS event_relation USING (event_id) + """ + if needs_distinct: select_keywords += " DISTINCT" diff --git a/tests/api/test_filtering.py b/tests/api/test_filtering.py index a269c477fb..a82c4eed86 100644 --- a/tests/api/test_filtering.py +++ b/tests/api/test_filtering.py @@ -35,6 +35,8 @@ def MockEvent(**kwargs): kwargs["event_id"] = "fake_event_id" if "type" not in kwargs: kwargs["type"] = "fake_type" + if "content" not in kwargs: + kwargs["content"] = {} return make_event_from_dict(kwargs) @@ -357,6 +359,66 @@ class FilteringTestCase(unittest.HomeserverTestCase): self.assertTrue(Filter(self.hs, definition)._check(event)) + @unittest.override_config({"experimental_features": {"msc3874_enabled": True}}) + def test_filter_rel_type(self): + definition = {"org.matrix.msc3874.rel_types": ["m.thread"]} + event = MockEvent( + sender="@foo:bar", + type="m.room.message", + room_id="!secretbase:unknown", + content={}, + ) + + self.assertFalse(Filter(self.hs, definition)._check(event)) + + event = MockEvent( + sender="@foo:bar", + type="m.room.message", + room_id="!secretbase:unknown", + content={"m.relates_to": {"event_id": "$abc", "rel_type": "m.reference"}}, + ) + + self.assertFalse(Filter(self.hs, definition)._check(event)) + + event = MockEvent( + sender="@foo:bar", + type="m.room.message", + room_id="!secretbase:unknown", + content={"m.relates_to": {"event_id": "$abc", "rel_type": "m.thread"}}, + ) + + self.assertTrue(Filter(self.hs, definition)._check(event)) + + @unittest.override_config({"experimental_features": {"msc3874_enabled": True}}) + def test_filter_not_rel_type(self): + definition = {"org.matrix.msc3874.not_rel_types": ["m.thread"]} + event = MockEvent( + sender="@foo:bar", + type="m.room.message", + room_id="!secretbase:unknown", + content={"m.relates_to": {"event_id": "$abc", "rel_type": "m.thread"}}, + ) + + self.assertFalse(Filter(self.hs, definition)._check(event)) + + event = MockEvent( + sender="@foo:bar", + type="m.room.message", + room_id="!secretbase:unknown", + content={}, + ) + + self.assertTrue(Filter(self.hs, definition)._check(event)) + + event = MockEvent( + sender="@foo:bar", + type="m.room.message", + room_id="!secretbase:unknown", + content={"m.relates_to": {"event_id": "$abc", "rel_type": "m.reference"}}, + ) + + self.assertTrue(Filter(self.hs, definition)._check(event)) + def test_filter_presence_match(self): user_filter_json = {"presence": {"types": ["m.*"]}} filter_id = self.get_success( @@ -456,7 +518,6 @@ class FilteringTestCase(unittest.HomeserverTestCase): self.assertEqual(filtered_room_ids, ["!allowed:example.com"]) - @unittest.override_config({"experimental_features": {"msc3440_enabled": True}}) def test_filter_relations(self): events = [ # An event without a relation. diff --git a/tests/rest/client/test_relations.py b/tests/rest/client/test_relations.py index f5c1070b2c..ddf315b894 100644 --- a/tests/rest/client/test_relations.py +++ b/tests/rest/client/test_relations.py @@ -1677,7 +1677,6 @@ class RelationRedactionTestCase(BaseRelationsTestCase): {"chunk": [{"type": "m.reaction", "key": "👍", "count": 1}]}, ) - @unittest.override_config({"experimental_features": {"msc3440_enabled": True}}) def test_redact_parent_thread(self) -> None: """ Test that thread replies are still available when the root event is redacted. diff --git a/tests/rest/client/test_rooms.py b/tests/rest/client/test_rooms.py index 3612ebe7b9..71b1637be8 100644 --- a/tests/rest/client/test_rooms.py +++ b/tests/rest/client/test_rooms.py @@ -35,7 +35,6 @@ from synapse.api.constants import ( EventTypes, Membership, PublicRoomsFilterFields, - RelationTypes, RoomTypes, ) from synapse.api.errors import Codes, HttpResponseException @@ -50,6 +49,7 @@ from synapse.util.stringutils import random_string from tests import unittest from tests.http.server._base import make_request_with_cancellation_test +from tests.storage.test_stream import PaginationTestCase from tests.test_utils import make_awaitable PATH_PREFIX = b"/_matrix/client/api/v1" @@ -2915,149 +2915,20 @@ class LabelsTestCase(unittest.HomeserverTestCase): return event_id -class RelationsTestCase(unittest.HomeserverTestCase): - servlets = [ - synapse.rest.admin.register_servlets_for_client_rest_resource, - room.register_servlets, - login.register_servlets, - ] - - def default_config(self) -> Dict[str, Any]: - config = super().default_config() - config["experimental_features"] = {"msc3440_enabled": True} - return config - - def prepare(self, reactor: MemoryReactor, clock: Clock, hs: HomeServer) -> None: - self.user_id = self.register_user("test", "test") - self.tok = self.login("test", "test") - self.room_id = self.helper.create_room_as(self.user_id, tok=self.tok) - - self.second_user_id = self.register_user("second", "test") - self.second_tok = self.login("second", "test") - self.helper.join( - room=self.room_id, user=self.second_user_id, tok=self.second_tok - ) - - self.third_user_id = self.register_user("third", "test") - self.third_tok = self.login("third", "test") - self.helper.join(room=self.room_id, user=self.third_user_id, tok=self.third_tok) - - # An initial event with a relation from second user. - res = self.helper.send_event( - room_id=self.room_id, - type=EventTypes.Message, - content={"msgtype": "m.text", "body": "Message 1"}, - tok=self.tok, - ) - self.event_id_1 = res["event_id"] - self.helper.send_event( - room_id=self.room_id, - type="m.reaction", - content={ - "m.relates_to": { - "rel_type": RelationTypes.ANNOTATION, - "event_id": self.event_id_1, - "key": "👍", - } - }, - tok=self.second_tok, - ) - - # Another event with a relation from third user. - res = self.helper.send_event( - room_id=self.room_id, - type=EventTypes.Message, - content={"msgtype": "m.text", "body": "Message 2"}, - tok=self.tok, - ) - self.event_id_2 = res["event_id"] - self.helper.send_event( - room_id=self.room_id, - type="m.reaction", - content={ - "m.relates_to": { - "rel_type": RelationTypes.REFERENCE, - "event_id": self.event_id_2, - } - }, - tok=self.third_tok, - ) - - # An event with no relations. - self.helper.send_event( - room_id=self.room_id, - type=EventTypes.Message, - content={"msgtype": "m.text", "body": "No relations"}, - tok=self.tok, - ) - - def _filter_messages(self, filter: JsonDict) -> List[JsonDict]: +class RelationsTestCase(PaginationTestCase): + def _filter_messages(self, filter: JsonDict) -> List[str]: """Make a request to /messages with a filter, returns the chunk of events.""" + from_token = self.get_success( + self.from_token.to_string(self.hs.get_datastores().main) + ) channel = self.make_request( "GET", - "/rooms/%s/messages?filter=%s&dir=b" % (self.room_id, json.dumps(filter)), + f"/rooms/{self.room_id}/messages?filter={json.dumps(filter)}&dir=f&from={from_token}", access_token=self.tok, ) self.assertEqual(channel.code, HTTPStatus.OK, channel.result) - return channel.json_body["chunk"] - - def test_filter_relation_senders(self) -> None: - # Messages which second user reacted to. - filter = {"related_by_senders": [self.second_user_id]} - chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 1, chunk) - self.assertEqual(chunk[0]["event_id"], self.event_id_1) - - # Messages which third user reacted to. - filter = {"related_by_senders": [self.third_user_id]} - chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 1, chunk) - self.assertEqual(chunk[0]["event_id"], self.event_id_2) - - # Messages which either user reacted to. - filter = {"related_by_senders": [self.second_user_id, self.third_user_id]} - chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 2, chunk) - self.assertCountEqual( - [c["event_id"] for c in chunk], [self.event_id_1, self.event_id_2] - ) - - def test_filter_relation_type(self) -> None: - # Messages which have annotations. - filter = {"related_by_rel_types": [RelationTypes.ANNOTATION]} - chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 1, chunk) - self.assertEqual(chunk[0]["event_id"], self.event_id_1) - - # Messages which have references. - filter = {"related_by_rel_types": [RelationTypes.REFERENCE]} - chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 1, chunk) - self.assertEqual(chunk[0]["event_id"], self.event_id_2) - - # Messages which have either annotations or references. - filter = { - "related_by_rel_types": [ - RelationTypes.ANNOTATION, - RelationTypes.REFERENCE, - ] - } - chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 2, chunk) - self.assertCountEqual( - [c["event_id"] for c in chunk], [self.event_id_1, self.event_id_2] - ) - - def test_filter_relation_senders_and_type(self) -> None: - # Messages which second user reacted to. - filter = { - "related_by_senders": [self.second_user_id], - "related_by_rel_types": [RelationTypes.ANNOTATION], - } - chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 1, chunk) - self.assertEqual(chunk[0]["event_id"], self.event_id_1) + return [ev["event_id"] for ev in channel.json_body["chunk"]] class ContextTestCase(unittest.HomeserverTestCase): diff --git a/tests/storage/test_stream.py b/tests/storage/test_stream.py index 78663a53fe..34fa810cf6 100644 --- a/tests/storage/test_stream.py +++ b/tests/storage/test_stream.py @@ -16,7 +16,6 @@ from typing import List from synapse.api.constants import EventTypes, RelationTypes from synapse.api.filtering import Filter -from synapse.events import EventBase from synapse.rest import admin from synapse.rest.client import login, room from synapse.types import JsonDict @@ -40,7 +39,7 @@ class PaginationTestCase(HomeserverTestCase): def default_config(self): config = super().default_config() - config["experimental_features"] = {"msc3440_enabled": True} + config["experimental_features"] = {"msc3874_enabled": True} return config def prepare(self, reactor, clock, homeserver): @@ -58,6 +57,11 @@ class PaginationTestCase(HomeserverTestCase): self.third_tok = self.login("third", "test") self.helper.join(room=self.room_id, user=self.third_user_id, tok=self.third_tok) + # Store a token which is after all the room creation events. + self.from_token = self.get_success( + self.hs.get_event_sources().get_current_token_for_pagination(self.room_id) + ) + # An initial event with a relation from second user. res = self.helper.send_event( room_id=self.room_id, @@ -66,7 +70,7 @@ class PaginationTestCase(HomeserverTestCase): tok=self.tok, ) self.event_id_1 = res["event_id"] - self.helper.send_event( + res = self.helper.send_event( room_id=self.room_id, type="m.reaction", content={ @@ -78,6 +82,7 @@ class PaginationTestCase(HomeserverTestCase): }, tok=self.second_tok, ) + self.event_id_annotation = res["event_id"] # Another event with a relation from third user. res = self.helper.send_event( @@ -87,7 +92,7 @@ class PaginationTestCase(HomeserverTestCase): tok=self.tok, ) self.event_id_2 = res["event_id"] - self.helper.send_event( + res = self.helper.send_event( room_id=self.room_id, type="m.reaction", content={ @@ -98,68 +103,59 @@ class PaginationTestCase(HomeserverTestCase): }, tok=self.third_tok, ) + self.event_id_reference = res["event_id"] # An event with no relations. - self.helper.send_event( + res = self.helper.send_event( room_id=self.room_id, type=EventTypes.Message, content={"msgtype": "m.text", "body": "No relations"}, tok=self.tok, ) + self.event_id_none = res["event_id"] - def _filter_messages(self, filter: JsonDict) -> List[EventBase]: + def _filter_messages(self, filter: JsonDict) -> List[str]: """Make a request to /messages with a filter, returns the chunk of events.""" - from_token = self.get_success( - self.hs.get_event_sources().get_current_token_for_pagination(self.room_id) - ) - events, next_key = self.get_success( self.hs.get_datastores().main.paginate_room_events( room_id=self.room_id, - from_key=from_token.room_key, + from_key=self.from_token.room_key, to_key=None, - direction="b", + direction="f", limit=10, event_filter=Filter(self.hs, filter), ) ) - return events + return [ev.event_id for ev in events] def test_filter_relation_senders(self): # Messages which second user reacted to. filter = {"related_by_senders": [self.second_user_id]} chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 1, chunk) - self.assertEqual(chunk[0].event_id, self.event_id_1) + self.assertEqual(chunk, [self.event_id_1]) # Messages which third user reacted to. filter = {"related_by_senders": [self.third_user_id]} chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 1, chunk) - self.assertEqual(chunk[0].event_id, self.event_id_2) + self.assertEqual(chunk, [self.event_id_2]) # Messages which either user reacted to. filter = {"related_by_senders": [self.second_user_id, self.third_user_id]} chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 2, chunk) - self.assertCountEqual( - [c.event_id for c in chunk], [self.event_id_1, self.event_id_2] - ) + self.assertCountEqual(chunk, [self.event_id_1, self.event_id_2]) def test_filter_relation_type(self): # Messages which have annotations. filter = {"related_by_rel_types": [RelationTypes.ANNOTATION]} chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 1, chunk) - self.assertEqual(chunk[0].event_id, self.event_id_1) + self.assertEqual(chunk, [self.event_id_1]) # Messages which have references. filter = {"related_by_rel_types": [RelationTypes.REFERENCE]} chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 1, chunk) - self.assertEqual(chunk[0].event_id, self.event_id_2) + self.assertEqual(chunk, [self.event_id_2]) # Messages which have either annotations or references. filter = { @@ -169,10 +165,7 @@ class PaginationTestCase(HomeserverTestCase): ] } chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 2, chunk) - self.assertCountEqual( - [c.event_id for c in chunk], [self.event_id_1, self.event_id_2] - ) + self.assertCountEqual(chunk, [self.event_id_1, self.event_id_2]) def test_filter_relation_senders_and_type(self): # Messages which second user reacted to. @@ -181,8 +174,7 @@ class PaginationTestCase(HomeserverTestCase): "related_by_rel_types": [RelationTypes.ANNOTATION], } chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 1, chunk) - self.assertEqual(chunk[0].event_id, self.event_id_1) + self.assertEqual(chunk, [self.event_id_1]) def test_duplicate_relation(self): """An event should only be returned once if there are multiple relations to it.""" @@ -201,5 +193,65 @@ class PaginationTestCase(HomeserverTestCase): filter = {"related_by_senders": [self.second_user_id]} chunk = self._filter_messages(filter) - self.assertEqual(len(chunk), 1, chunk) - self.assertEqual(chunk[0].event_id, self.event_id_1) + self.assertEqual(chunk, [self.event_id_1]) + + def test_filter_rel_types(self) -> None: + # Messages which are annotations. + filter = {"org.matrix.msc3874.rel_types": [RelationTypes.ANNOTATION]} + chunk = self._filter_messages(filter) + self.assertEqual(chunk, [self.event_id_annotation]) + + # Messages which are references. + filter = {"org.matrix.msc3874.rel_types": [RelationTypes.REFERENCE]} + chunk = self._filter_messages(filter) + self.assertEqual(chunk, [self.event_id_reference]) + + # Messages which are either annotations or references. + filter = { + "org.matrix.msc3874.rel_types": [ + RelationTypes.ANNOTATION, + RelationTypes.REFERENCE, + ] + } + chunk = self._filter_messages(filter) + self.assertCountEqual( + chunk, + [self.event_id_annotation, self.event_id_reference], + ) + + def test_filter_not_rel_types(self) -> None: + # Messages which are not annotations. + filter = {"org.matrix.msc3874.not_rel_types": [RelationTypes.ANNOTATION]} + chunk = self._filter_messages(filter) + self.assertEqual( + chunk, + [ + self.event_id_1, + self.event_id_2, + self.event_id_reference, + self.event_id_none, + ], + ) + + # Messages which are not references. + filter = {"org.matrix.msc3874.not_rel_types": [RelationTypes.REFERENCE]} + chunk = self._filter_messages(filter) + self.assertEqual( + chunk, + [ + self.event_id_1, + self.event_id_annotation, + self.event_id_2, + self.event_id_none, + ], + ) + + # Messages which are neither annotations or references. + filter = { + "org.matrix.msc3874.not_rel_types": [ + RelationTypes.ANNOTATION, + RelationTypes.REFERENCE, + ] + } + chunk = self._filter_messages(filter) + self.assertEqual(chunk, [self.event_id_1, self.event_id_2, self.event_id_none])