Antalya 26.3 Backport of #100361 - Fix exception in updateFormatPrewhereInfo when only row-level filter is set

src/Storages/prepareReadingFromFormat.cpp

@@ -260,19 +260,23 @@ Names filterTupleColumnsToRead(NamesAndTypesList & requested_columns)
 
 ReadFromFormatInfo updateFormatPrewhereInfo(const ReadFromFormatInfo & info, const FilterDAGInfoPtr & row_level_filter, const PrewhereInfoPtr & prewhere_info)
 {
-    chassert(prewhere_info);
+    chassert(prewhere_info || row_level_filter);
 
-    if (info.prewhere_info || info.row_level_filter)
+    if (info.prewhere_info)
         throw Exception(ErrorCodes::LOGICAL_ERROR, "updateFormatPrewhereInfo called more than once");
 
     ReadFromFormatInfo new_info;
     new_info.prewhere_info = prewhere_info;
+    new_info.row_level_filter = row_level_filter;
 
     /// Removes columns that are only used as prewhere input.
     /// Adds prewhere outputs (the actual prewhere filter column is only added if
     /// !remove_prewhere_column; but there may also be subexpressions computed by prewhere
     /// expression and preserved for use further down the query pipeline).
-    new_info.format_header = SourceStepWithFilter::applyPrewhereActions(info.format_header, row_level_filter, prewhere_info);
+    /// If row_level_filter was already applied in a previous call, don't re-apply it;
+    /// only apply the new prewhere_info on top.
+    new_info.format_header = SourceStepWithFilter::applyPrewhereActions(
+        info.format_header, info.row_level_filter ? nullptr : row_level_filter, prewhere_info);
 
     /// We assume that any format that supports prewhere also supports subset of subcolumns, so we
     /// don't need to replace subcolumns with their nested columns etc.

tests/queries/0_stateless/04053_row_policy_object_storage.reference

@@ -0,0 +1,7 @@
+--- Row policy filters URL Parquet table ---
+1	a
+2	b
+--- Row policy with WHERE on URL Parquet table ---
+1	a
+--- Row policy count on URL Parquet table ---
+2

tests/queries/0_stateless/04053_row_policy_object_storage.sh

@@ -0,0 +1,46 @@
+#!/usr/bin/env bash
+# Tags: no-replicated-database, no-fasttest
+
+# Regression test: row policy on a URL table with Parquet format caused
+# "Logical error: 'prewhere_info'" because updateFormatPrewhereInfo asserted
+# prewhere_info was non-null, but only row_level_filter was set.
+
+CURDIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
+# shellcheck source=../shell_config.sh
+. "$CURDIR"/../shell_config.sh
+
+user="user04053_${CLICKHOUSE_DATABASE}_$RANDOM"
+db=${CLICKHOUSE_DATABASE}
+
+${CLICKHOUSE_CLIENT} <<EOF
+DROP TABLE IF EXISTS ${db}.source_data;
+CREATE TABLE ${db}.source_data (id UInt32, value String) ENGINE = MergeTree ORDER BY id;
+INSERT INTO ${db}.source_data VALUES (1, 'a'), (2, 'b'), (3, 'c');
+
+DROP TABLE IF EXISTS ${db}.url_parquet;
+CREATE TABLE ${db}.url_parquet (id UInt32, value String)
+ENGINE = URL('http://127.0.0.1:${CLICKHOUSE_PORT_HTTP}/?query=SELECT+*+FROM+${db}.source_data+FORMAT+Parquet', 'Parquet');
+
+DROP USER IF EXISTS ${user};
+CREATE USER ${user} IDENTIFIED WITH no_password;
+GRANT SELECT ON ${db}.url_parquet TO ${user};
+
+DROP ROW POLICY IF EXISTS rp_04053 ON ${db}.url_parquet;
+CREATE ROW POLICY rp_04053 ON ${db}.url_parquet FOR SELECT USING id <= 2 TO ${user};
+EOF
+
+echo "--- Row policy filters URL Parquet table ---"
+${CLICKHOUSE_CLIENT} --user ${user} --query "SELECT * FROM ${db}.url_parquet ORDER BY id"
+
+echo "--- Row policy with WHERE on URL Parquet table ---"
+${CLICKHOUSE_CLIENT} --user ${user} --query "SELECT * FROM ${db}.url_parquet WHERE value = 'a' ORDER BY id"
+
+echo "--- Row policy count on URL Parquet table ---"
+${CLICKHOUSE_CLIENT} --user ${user} --query "SELECT count() FROM ${db}.url_parquet"
+
+${CLICKHOUSE_CLIENT} <<EOF
+DROP ROW POLICY IF EXISTS rp_04053 ON ${db}.url_parquet;
+DROP USER IF EXISTS ${user};
+DROP TABLE IF EXISTS ${db}.url_parquet;
+DROP TABLE IF EXISTS ${db}.source_data;
+EOF