migration: Rename 'general chat' topic to empty string topic.

Zulip now supports empty string as a valid topic name.

For clients predating this feature, such messages appear
in "general chat" topic. Messages sent to "general chat" are
stored in the database as having a "" topic.

This commit adds a migration to rename the existing
"general chat" topic in the database to "".

Fixes parts of #32996.

zerver/migrations/0679_zerver_message_edit_history_id.py

@@ -0,0 +1,23 @@
+# Generated by Django 5.1.6 on 2025-02-26 17:25
+
+from django.contrib.postgres.operations import AddIndexConcurrently
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+    atomic = False
+
+    dependencies = [
+        ("zerver", "0678_remove_realm_allow_edit_history"),
+    ]
+
+    operations = [
+        AddIndexConcurrently(
+            model_name="message",
+            index=models.Index(
+                condition=models.Q(("edit_history__isnull", False)),
+                fields=["id"],
+                name="zerver_message_edit_history_id",
+            ),
+        ),
+    ]

zerver/migrations/0680_rename_general_chat_to_empty_string_topic.py

@@ -0,0 +1,147 @@
+from typing import Any
+
+from django.db import connection, migrations, models, transaction
+from django.db.backends.base.schema import BaseDatabaseSchemaEditor
+from django.db.migrations.state import StateApps
+from psycopg2.sql import SQL, Identifier
+
+
+def update_edit_history(message_model: type[Any]) -> None:
+    BATCH_SIZE = 10000
+    lower_id_bound = 0
+
+    max_id = message_model.objects.aggregate(models.Max("id"))["id__max"]
+    if max_id is None:
+        return
+
+    while lower_id_bound < max_id:
+        upper_id_bound = min(lower_id_bound + BATCH_SIZE, max_id)
+        with connection.cursor() as cursor:
+            query = SQL(
+                """
+                UPDATE {table_name}
+                SET edit_history = (
+                    SELECT JSONB_AGG(
+                        elem
+                        ||
+                        (CASE
+                            WHEN elem ? 'prev_topic' AND elem->>'prev_topic' = 'general chat'
+                            THEN '{{"prev_topic": ""}}'::jsonb
+                            ELSE '{{}}'::jsonb
+                        END)
+                        ||
+                        (CASE
+                            WHEN elem ? 'topic' AND elem->>'topic' = 'general chat'
+                            THEN '{{"topic": ""}}'::jsonb
+                            ELSE '{{}}'::jsonb
+                        END)
+                    )::text
+                    FROM JSONB_ARRAY_ELEMENTS(edit_history::jsonb) AS elem
+                )
+                WHERE edit_history IS NOT NULL
+                AND id > %(lower_id_bound)s AND id <= %(upper_id_bound)s
+                AND (
+                    edit_history::jsonb @> '[{{"prev_topic": "general chat"}}]' OR
+                    edit_history::jsonb @> '[{{"topic": "general chat"}}]'
+                );
+            """
+            ).format(table_name=Identifier(message_model._meta.db_table))
+            cursor.execute(
+                query,
+                {
+                    "lower_id_bound": lower_id_bound,
+                    "upper_id_bound": upper_id_bound,
+                },
+            )
+
+        print(f"Processed {upper_id_bound} / {max_id}")
+        lower_id_bound += BATCH_SIZE
+
+
+def rename_general_chat_to_empty_string_topic(
+    apps: StateApps, schema_editor: BaseDatabaseSchemaEditor
+) -> None:
+    """Because legacy clients will be unable to distinguish the topic "general chat"
+    from the "" topic (displayed as italicized "general chat"), it's helpful
+    to not have both types of topics exist in an organization.
+
+    Further, the "general chat" topic is likely to in almost every
+    case be the result of an organization that followed our advice to
+    just make a "general chat" topic for topic-free chat; the new
+    "general chat" feature naturally replaces that learned
+    behavior.
+
+    So it makes sense to just consider those older "general chat"
+    topics to be the same as the modern general chat topic.
+
+    The technical way to do that is to rewrite those topics in the
+    database to be represented as `""` rather than "general chat",
+    since we've endeavored to make the distinction between those two
+    storage approaches invisible to legacy clients at the API layer.
+
+    Thus, we don't generate edit history entries for this, since we're
+    thinking of it as redefining how "general chat" is stored in the
+    database.
+    """
+    Realm = apps.get_model("zerver", "Realm")
+    Message = apps.get_model("zerver", "Message")
+    UserTopic = apps.get_model("zerver", "UserTopic")
+    ArchivedMessage = apps.get_model("zerver", "ArchivedMessage")
+    ScheduledMessage = apps.get_model("zerver", "ScheduledMessage")
+
+    for realm in Realm.objects.all():
+        with transaction.atomic(durable=True):
+            # Uses index "zerver_message_realm_upper_subject"
+            message_queryset = Message.objects.filter(realm=realm, subject__iexact="general chat")
+            channel_ids = list(
+                message_queryset.distinct("recipient__type_id").values_list(
+                    "recipient__type_id", flat=True
+                )
+            )
+
+            message_queryset.update(subject="")
+
+            # Limiting the UserTopic query to only those channels that
+            # contain an actual general chat topic does not guaranteed
+            # updating all UserTopic rows, since it's possible to
+            # follow/mute an empty topic. But it does guarantee that
+            # we update all rows that have any current effect.
+            #
+            # Uses index "zerver_mutedtopic_stream_topic"
+            UserTopic.objects.filter(
+                stream_id__in=channel_ids, topic_name__iexact="general chat"
+            ).update(topic_name="")
+
+            ArchivedMessage.objects.filter(realm=realm, subject__iexact="general chat").update(
+                subject=""
+            )
+            ScheduledMessage.objects.filter(realm=realm, subject__iexact="general chat").update(
+                subject=""
+            )
+
+    for message_model in [Message, ArchivedMessage]:
+        update_edit_history(message_model)
+
+
+class Migration(migrations.Migration):
+    """
+    Zulip now supports empty string as a valid topic name.
+    For clients predating this feature, such messages appear
+    in "general chat" topic. Messages sent to "general chat" are
+    stored in the database as having a "" topic. This migration
+    renames the existing "general chat" topic in the database to "".
+    """
+
+    atomic = False
+
+    dependencies = [
+        ("zerver", "0679_zerver_message_edit_history_id"),
+    ]
+
+    operations = [
+        migrations.RunPython(
+            rename_general_chat_to_empty_string_topic,
+            reverse_code=migrations.RunPython.noop,
+            elidable=True,
+        ),
+    ]

zerver/models/messages.py

@@ -163,7 +163,7 @@ class Message(AbstractMessage):
 
     # Name to be used for the empty topic with clients that have not
     # yet migrated to have the `empty_topic_name` client capability.
-    EMPTY_TOPIC_FALLBACK_NAME = "test general chat"
+    EMPTY_TOPIC_FALLBACK_NAME = "general chat"
 
     class Meta:
         indexes = [
@@ -238,6 +238,12 @@ class Message(AbstractMessage):
                 F("id").desc(nulls_last=True),
                 name="zerver_message_realm_id",
             ),
+            models.Index(
+                # Used by 0680_rename_general_chat_to_empty_string_topic
+                fields=["id"],
+                condition=Q(edit_history__isnull=False),
+                name="zerver_message_edit_history_id",
+            ),
         ]
 
     def topic_name(self) -> str: