Re-implement unread counts (again) (#8059)

synapse/push/bulk_push_rule_evaluator.py

@@ -19,8 +19,10 @@ from collections import namedtuple
 
 from prometheus_client import Counter
 
-from synapse.api.constants import EventTypes, Membership
+from synapse.api.constants import EventTypes, Membership, RelationTypes
 from synapse.event_auth import get_user_power_level
+from synapse.events import EventBase
+from synapse.events.snapshot import EventContext
 from synapse.state import POWER_KEY
 from synapse.util.async_helpers import Linearizer
 from synapse.util.caches import register_cache
@@ -51,6 +53,48 @@ push_rules_delta_state_cache_metric = register_cache(
 )
 
 
+STATE_EVENT_TYPES_TO_MARK_UNREAD = {
+    EventTypes.Topic,
+    EventTypes.Name,
+    EventTypes.RoomAvatar,
+    EventTypes.Tombstone,
+}
+
+
+def _should_count_as_unread(event: EventBase, context: EventContext) -> bool:
+    # Exclude rejected and soft-failed events.
+    if context.rejected or event.internal_metadata.is_soft_failed():
+        return False
+
+    # Exclude notices.
+    if (
+        not event.is_state()
+        and event.type == EventTypes.Message
+        and event.content.get("msgtype") == "m.notice"
+    ):
+        return False
+
+    # Exclude edits.
+    relates_to = event.content.get("m.relates_to", {})
+    if relates_to.get("rel_type") == RelationTypes.REPLACE:
+        return False
+
+    # Mark events that have a non-empty string body as unread.
+    body = event.content.get("body")
+    if isinstance(body, str) and body:
+        return True
+
+    # Mark some state events as unread.
+    if event.is_state() and event.type in STATE_EVENT_TYPES_TO_MARK_UNREAD:
+        return True
+
+    # Mark encrypted events as unread.
+    if not event.is_state() and event.type == EventTypes.Encrypted:
+        return True
+
+    return False
+
+
 class BulkPushRuleEvaluator(object):
     """Calculates the outcome of push rules for an event for all users in the
     room at once.
@@ -133,9 +177,12 @@ class BulkPushRuleEvaluator(object):
         return pl_event.content if pl_event else {}, sender_level
 
     async def action_for_event_by_user(self, event, context) -> None:
-        """Given an event and context, evaluate the push rules and insert the
-        results into the event_push_actions_staging table.
+        """Given an event and context, evaluate the push rules, check if the message
+        should increment the unread count, and insert the results into the
+        event_push_actions_staging table.
         """
+        count_as_unread = _should_count_as_unread(event, context)
+
         rules_by_user = await self._get_rules_for_event(event, context)
         actions_by_user = {}
 
@@ -172,6 +219,8 @@ class BulkPushRuleEvaluator(object):
                 if event.type == EventTypes.Member and event.state_key == uid:
                     display_name = event.content.get("displayname", None)
 
+            actions_by_user[uid] = []
+
             for rule in rules:
                 if "enabled" in rule and not rule["enabled"]:
                     continue
@@ -189,7 +238,9 @@ class BulkPushRuleEvaluator(object):
         # Mark in the DB staging area the push actions for users who should be
         # notified for this event. (This will then get handled when we persist
         # the event)
-        await self.store.add_push_actions_to_staging(event.event_id, actions_by_user)
+        await self.store.add_push_actions_to_staging(
+            event.event_id, actions_by_user, count_as_unread,
+        )
 
 
 def _condition_checker(evaluator, conditions, uid, display_name, cache):
@@ -369,8 +420,8 @@ class RulesForRoom(object):
         Args:
             ret_rules_by_user (dict): Partiallly filled dict of push rules. Gets
                 updated with any new rules.
-            member_event_ids (list): List of event ids for membership events that
-                have happened since the last time we filled rules_by_user
+            member_event_ids (dict): Dict of user id to event id for membership events
+                that have happened since the last time we filled rules_by_user
             state_group: The state group we are currently computing push rules
                 for. Used when updating the cache.
         """
@@ -390,34 +441,19 @@ class RulesForRoom(object):
         if logger.isEnabledFor(logging.DEBUG):
             logger.debug("Found members %r: %r", self.room_id, members.values())
 
-        interested_in_user_ids = {
+        user_ids = {
             user_id
             for user_id, membership in members.values()
             if membership == Membership.JOIN
         }
 
-        logger.debug("Joined: %r", interested_in_user_ids)
+        logger.debug("Joined: %r", user_ids)
 
-        if_users_with_pushers = await self.store.get_if_users_have_pushers(
-            interested_in_user_ids, on_invalidate=self.invalidate_all_cb
-        )
-
-        user_ids = {
-            uid for uid, have_pusher in if_users_with_pushers.items() if have_pusher
-        }
-
-        logger.debug("With pushers: %r", user_ids)
-
-        users_with_receipts = await self.store.get_users_with_read_receipts_in_room(
-            self.room_id, on_invalidate=self.invalidate_all_cb
-        )
-
-        logger.debug("With receipts: %r", users_with_receipts)
-
-        # any users with pushers must be ours: they have pushers
-        for uid in users_with_receipts:
-            if uid in interested_in_user_ids:
-                user_ids.add(uid)
+        # Previously we only considered users with pushers or read receipts in that
+        # room. We can't do this anymore because we use push actions to calculate unread
+        # counts, which don't rely on the user having pushers or sent a read receipt into
+        # the room. Therefore we just need to filter for local users here.
+        user_ids = list(filter(self.is_mine_id, user_ids))
 
         rules_by_user = await self.store.bulk_get_push_rules(
             user_ids, on_invalidate=self.invalidate_all_cb