Fix alias on default columns

src/Storages/MergeTree/MergeTreeBlockReadUtils.cpp

@@ -12,59 +12,76 @@ namespace ErrorCodes
     extern const int LOGICAL_ERROR;
 }
 
+namespace
+{
+
+/// Columns absent in part may depend on other absent columns so we are
+/// searching all required columns recursively. Return true if found at least
+/// one existing column in part.
+bool injectRequiredColumnsRecursively(
+    const String & column_name,
+    const ColumnsDescription & storage_columns,
+    const MergeTreeData::AlterConversions & alter_conversions,
+    const MergeTreeData::DataPartPtr & part,
+    Names & columns,
+    NameSet & required_columns,
+    NameSet & injected_columns)
+{
+    String column_name_in_part = column_name;
+    if (alter_conversions.isColumnRenamed(column_name_in_part))
+        column_name_in_part = alter_conversions.getColumnOldName(column_name_in_part);
+
+    /// column has files and hence does not require evaluation
+    if (storage_columns.hasPhysical(column_name) && part->hasColumnFiles(column_name_in_part, *storage_columns.getPhysical(column_name).type))
+    {
+        /// ensure each column is added only once
+        if (required_columns.count(column_name) == 0)
+        {
+            columns.emplace_back(column_name);
+            required_columns.emplace(column_name);
+            injected_columns.emplace(column_name);
+        }
+        return true;
+    }
+
+    /// Column doesn't have default value and don't exist in part
+    /// don't need to add to required set.
+    const auto column_default = storage_columns.getDefault(column_name);
+    if (!column_default)
+        return false;
+
+    /// collect identifiers required for evaluation
+    IdentifierNameSet identifiers;
+    column_default->expression->collectIdentifierNames(identifiers);
+
+    bool result = false;
+    for (const auto & identifier : identifiers)
+        result |= injectRequiredColumnsRecursively(identifier, storage_columns, alter_conversions, part, columns, required_columns, injected_columns);
+
+    return result;
+}
+
+}
 
 NameSet injectRequiredColumns(const MergeTreeData & storage, const StorageMetadataPtr & metadata_snapshot, const MergeTreeData::DataPartPtr & part, Names & columns)
 {
     NameSet required_columns{std::begin(columns), std::end(columns)};
     NameSet injected_columns;
 
-    auto all_column_files_missing = true;
+    bool have_at_least_one_physical_column = false;
 
     const auto & storage_columns = metadata_snapshot->getColumns();
     auto alter_conversions = storage.getAlterConversionsForPart(part);
     for (size_t i = 0; i < columns.size(); ++i)
-    {
-        /// possibly renamed
-        auto column_name_in_part = columns[i];
-
-        if (alter_conversions.isColumnRenamed(column_name_in_part))
-            column_name_in_part = alter_conversions.getColumnOldName(column_name_in_part);
-
-        /// column has files and hence does not require evaluation
-        if (part->hasColumnFiles(column_name_in_part, *storage_columns.getPhysical(columns[i]).type))
-        {
-            all_column_files_missing = false;
-            continue;
-        }
-
-        const auto column_default = storage_columns.getDefault(columns[i]);
-        if (!column_default)
-            continue;
-
-        /// collect identifiers required for evaluation
-        IdentifierNameSet identifiers;
-        column_default->expression->collectIdentifierNames(identifiers);
-
-        for (const auto & identifier : identifiers)
-        {
-            if (storage_columns.hasPhysical(identifier))
-            {
-                /// ensure each column is added only once
-                if (required_columns.count(identifier) == 0)
-                {
-                    columns.emplace_back(identifier);
-                    required_columns.emplace(identifier);
-                    injected_columns.emplace(identifier);
-                }
-            }
-        }
-    }
+        have_at_least_one_physical_column |= injectRequiredColumnsRecursively(
+            columns[i], storage_columns, alter_conversions,
+            part, columns, required_columns, injected_columns);
 
     /** Add a column of the minimum size.
         * Used in case when no column is needed or files are missing, but at least you need to know number of rows.
         * Adds to the columns.
         */
-    if (all_column_files_missing)
+    if (!have_at_least_one_physical_column)
     {
         const auto minimum_size_column_name = part->getColumnNameWithMinumumCompressedSize(metadata_snapshot);
         columns.push_back(minimum_size_column_name);

tests/queries/0_stateless/01084_defaults_on_aliases.reference

@@ -1,5 +1,6 @@
 1	1
 1	1	1
+1
 2	2	4
 2	2	2	4
 3	3	9

tests/queries/0_stateless/01084_defaults_on_aliases.sql

@@ -2,12 +2,16 @@ DROP TABLE IF EXISTS table_with_defaults_on_aliases;
 
 CREATE TABLE table_with_defaults_on_aliases (col1 UInt32, col2 ALIAS col1, col3 DEFAULT col2) Engine = MergeTree() ORDER BY tuple();
 
+SYSTEM STOP MERGES table_with_defaults_on_aliases;
+
 INSERT INTO table_with_defaults_on_aliases (col1) VALUES (1);
 
 SELECT * FROM table_with_defaults_on_aliases WHERE col1 = 1;
 
 SELECT col1, col2, col3 FROM table_with_defaults_on_aliases WHERE col1 = 1;
 
+SELECT col3 FROM table_with_defaults_on_aliases; -- important to check without WHERE
+
 ALTER TABLE table_with_defaults_on_aliases ADD COLUMN col4 UInt64 DEFAULT col2 * col3;
 
 INSERT INTO table_with_defaults_on_aliases (col1) VALUES (2);
@@ -24,7 +28,6 @@ SELECT * FROM table_with_defaults_on_aliases WHERE col1 = 3;
 
 SELECT col1, col2, col3, col4, col5 FROM table_with_defaults_on_aliases WHERE col1 = 3;
 
-
 ALTER TABLE table_with_defaults_on_aliases ADD COLUMN col6 UInt64 MATERIALIZED col2 * col4;
 
 DROP TABLE IF EXISTS table_with_defaults_on_aliases;

tests/queries/0_stateless/01497_alias_on_default_array.reference

@@ -0,0 +1,6 @@
+a1	b1
+a2	b2
+a3	b3
+c1
+c2
+c3

tests/queries/0_stateless/01497_alias_on_default_array.sql

@@ -0,0 +1,21 @@
+DROP TABLE IF EXISTS test_new_col;
+
+CREATE TABLE test_new_col
+(
+  `_csv` String,
+  `csv_as_array` Array(String) ALIAS splitByChar(';',_csv),
+  `csv_col1` String DEFAULT csv_as_array[1],
+  `csv_col2` String DEFAULT csv_as_array[2]
+)
+ENGINE = MergeTree
+ORDER BY tuple();
+
+INSERT INTO test_new_col (_csv) VALUES ('a1;b1;c1;d1'), ('a2;b2;c2;d2'), ('a3;b3;c3;d3');
+
+SELECT csv_col1, csv_col2 FROM test_new_col ORDER BY csv_col1;
+
+ALTER TABLE test_new_col ADD COLUMN `csv_col3` String DEFAULT csv_as_array[3];
+
+SELECT csv_col3 FROM test_new_col ORDER BY csv_col3;
+
+DROP TABLE IF EXISTS test_new_col;