Fix Iceberg ARRAY columns with dot-separated names returning empty lists

src/DataTypes/NestedUtils.cpp

@@ -212,21 +212,37 @@ using NameToDataType = std::map<String, DataTypePtr>;
 
 NameToDataType getSubcolumnsOfNested(const NamesAndTypesList & names_and_types)
 {
-    std::unordered_map<String, NamesAndTypesList> nested;
+    /// Pass 1: count how many Array(T) columns share each dotted prefix.
+    /// A lone column like `a.b Array(T)` must not be collapsed into a synthetic
+    /// Nested parent — only genuine flat-Nested groups (n.x, n.y, ...) qualify.
+    std::unordered_map<String, size_t> prefix_count;
     for (const auto & name_type : names_and_types)
     {
         /// Skip subcolumns (e.g. `c0.c2.null` derived from `c0.c2 Array(Nullable(Tuple()))`).
-        /// They are not real flat-nested columns like `n.a Array(T)`, `n.b Array(T)`.
         if (name_type.isSubcolumn())
             continue;
 
         const auto * type_arr = typeid_cast<const DataTypeArray *>(name_type.type.get());
-
-        /// Ignore true Nested type, but try to unite flatten arrays to Nested type.
         if (!isNested(name_type.type) && type_arr)
         {
             auto split = splitName(name_type.name);
             if (!split.second.empty())
+                ++prefix_count[split.first];
+        }
+    }
+
+    /// Pass 2: build Nested only for prefixes shared by at least two columns.
+    std::unordered_map<String, NamesAndTypesList> nested;
+    for (const auto & name_type : names_and_types)
+    {
+        if (name_type.isSubcolumn())
+            continue;
+
+        const auto * type_arr = typeid_cast<const DataTypeArray *>(name_type.type.get());
+        if (!isNested(name_type.type) && type_arr)
+        {
+            auto split = splitName(name_type.name);
+            if (!split.second.empty() && prefix_count[split.first] >= 2)
                 nested[split.first].emplace_back(split.second, type_arr->getNestedType());
         }
     }

src/Storages/ColumnsDescription.cpp

@@ -976,10 +976,7 @@ std::vector<String> ColumnsDescription::getAllRegisteredNames() const
     std::vector<String> names;
     names.reserve(columns.size());
     for (const auto & column : columns)
-    {
-        if (!column.name.contains('.'))
-            names.push_back(column.name);
-    }
+        names.emplace_back(column.name);
     return names;
 }
 

tests/integration/test_storage_iceberg_with_spark/test_column_names_with_dots.py

@@ -216,3 +216,101 @@ def test_deeply_nested_struct_with_dotted_names(started_cluster_iceberg_with_spa
     ).strip()
     expected = "deep_value1\ndeep_value2\ndeep_value3"
     assert result == expected, f"Expected:\n{expected}\nGot:\n{result}"
+
+
+@pytest.mark.parametrize("storage_type", ["s3", "azure", "local"])
+def test_dotted_array_column(started_cluster_iceberg_with_spark, storage_type):
+    """
+    Regression test for issue #90731.
+    A top-level ARRAY column whose name literally contains a dot (e.g. `a.b`)
+    must be returned with its actual values, not as an empty array.
+    """
+    instance = started_cluster_iceberg_with_spark.instances["node1"]
+    spark = started_cluster_iceberg_with_spark.spark_session
+    TABLE_NAME = "test_dotted_array_column_" + storage_type + "_" + get_uuid_str()
+
+    from pyspark.sql.types import ArrayType
+
+    data = [(["a", "b", "c"],)]
+    schema = StructType([
+        StructField("a.b", ArrayType(StringType())),
+    ])
+    df = spark.createDataFrame(data=data, schema=schema)
+
+    write_iceberg_from_df(spark, df, TABLE_NAME, mode="overwrite", format_version="2")
+
+    default_upload_directory(
+        started_cluster_iceberg_with_spark,
+        storage_type,
+        f"/iceberg_data/default/{TABLE_NAME}/",
+        f"/iceberg_data/default/{TABLE_NAME}/",
+    )
+
+    # Test via table function
+    table_function_expr = get_creation_expression(
+        storage_type, TABLE_NAME, started_cluster_iceberg_with_spark, table_function=True
+    )
+
+    result = instance.query(
+        f"SELECT `a.b` FROM {table_function_expr}"
+    ).strip()
+    assert result == "['a','b','c']", f"Expected ['a','b','c'], got: {result}"
+
+    # Test via table engine
+    create_iceberg_table(storage_type, instance, TABLE_NAME, started_cluster_iceberg_with_spark)
+
+    result = instance.query(
+        f"SELECT `a.b` FROM {TABLE_NAME}"
+    ).strip()
+    assert result == "['a','b','c']", f"Expected ['a','b','c'], got: {result}"
+
+
+@pytest.mark.parametrize("storage_type", ["s3", "azure", "local"])
+def test_dotted_array_alongside_real_nested(started_cluster_iceberg_with_spark, storage_type):
+    """
+    Regression guard: a lone dotted Array column (`a.b`) must not interfere with
+    a genuine flat-Nested group (`c.x`, `c.y`) that shares a different prefix.
+    All three columns must round-trip correctly.
+    """
+    instance = started_cluster_iceberg_with_spark.instances["node1"]
+    spark = started_cluster_iceberg_with_spark.spark_session
+    TABLE_NAME = "test_dotted_array_alongside_real_nested_" + storage_type + "_" + get_uuid_str()
+
+    from pyspark.sql.types import ArrayType, IntegerType as SparkIntegerType
+
+    data = [(["a", "b", "c"], [1, 2], ["p", "q"])]
+    schema = StructType([
+        StructField("a.b", ArrayType(StringType())),
+        StructField("c.x", ArrayType(SparkIntegerType())),
+        StructField("c.y", ArrayType(StringType())),
+    ])
+    df = spark.createDataFrame(data=data, schema=schema)
+
+    write_iceberg_from_df(spark, df, TABLE_NAME, mode="overwrite", format_version="2")
+
+    default_upload_directory(
+        started_cluster_iceberg_with_spark,
+        storage_type,
+        f"/iceberg_data/default/{TABLE_NAME}/",
+        f"/iceberg_data/default/{TABLE_NAME}/",
+    )
+
+    # Test via table function
+    table_function_expr = get_creation_expression(
+        storage_type, TABLE_NAME, started_cluster_iceberg_with_spark, table_function=True
+    )
+
+    result = instance.query(
+        f"SELECT `a.b`, `c.x`, `c.y` FROM {table_function_expr}"
+    ).strip()
+    assert result == "['a','b','c']\t[1,2]\t['p','q']", \
+        f"Unexpected result via table function: {result}"
+
+    # Test via table engine
+    create_iceberg_table(storage_type, instance, TABLE_NAME, started_cluster_iceberg_with_spark)
+
+    result = instance.query(
+        f"SELECT `a.b`, `c.x`, `c.y` FROM {TABLE_NAME}"
+    ).strip()
+    assert result == "['a','b','c']\t[1,2]\t['p','q']", \
+        f"Unexpected result via table engine: {result}"

tests/queries/0_stateless/04259_dotted_array_not_nested.reference

@@ -0,0 +1,3 @@
+['a','b','c']
+['a','b','c']
+[1,2]	['p','q']

tests/queries/0_stateless/04259_dotted_array_not_nested.sql

@@ -0,0 +1,18 @@
+-- Regression test for #90731.
+-- A lone Array(T) column with a dot in its name must not be collapsed into
+-- a synthetic Nested structure and must be readable as a plain array.
+
+CREATE TABLE t1 (`a.b` Array(String)) ENGINE = Memory;
+INSERT INTO t1 VALUES (['a','b','c']);
+SELECT `a.b` FROM t1;
+
+-- In a mixed table, the lone dotted column must not interfere with the
+-- genuine flat-Nested group (c.x / c.y share prefix 'c').
+CREATE TABLE t2 (`a.b` Array(String), `c.x` Array(Int32), `c.y` Array(String))
+    ENGINE = Memory;
+INSERT INTO t2 VALUES (['a','b','c'], [1,2], ['p','q']);
+SELECT `a.b` FROM t2;
+SELECT `c.x`, `c.y` FROM t2;
+
+DROP TABLE t1;
+DROP TABLE t2;

tests/queries/0_stateless/04260_dotted_column_in_hints.reference

@@ -0,0 +1 @@
+ok

tests/queries/0_stateless/04260_dotted_column_in_hints.sh

@@ -0,0 +1,19 @@
+#!/usr/bin/env bash
+# Regression test for #90731.
+# ColumnsDescription::getAllRegisteredNames must include columns whose names
+# contain a dot, so they appear in IHints suggestions after a typo.
+
+CURDIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
+# shellcheck source=../shell_config.sh
+. "$CURDIR"/../shell_config.sh
+
+$CLICKHOUSE_CLIENT -q "
+    CREATE TABLE t_dotted_hint (\`a.b\` Array(String))
+    ENGINE = MergeTree ORDER BY tuple();
+"
+
+# Misspell the column name; the error message must suggest the real name 'a.b'.
+$CLICKHOUSE_CLIENT -q "ALTER TABLE t_dotted_hint MODIFY COLUMN a_b Array(String);" 2>&1 \
+    | grep -qF "a.b" && echo "ok" || echo "FAIL"
+
+$CLICKHOUSE_CLIENT -q "DROP TABLE t_dotted_hint;"