Backport #63962 to 24.3: Analyzer: Fix COLUMNS resolve

src/Analyzer/Passes/QueryAnalysisPass.cpp

@@ -4613,6 +4613,36 @@ QueryAnalyzer::QueryTreeNodesWithNames QueryAnalyzer::resolveUnqualifiedMatcher(
 
     std::unordered_set<std::string> table_expression_column_names_to_skip;
 
+    QueryTreeNodesWithNames result;
+
+    if (matcher_node_typed.getMatcherType() == MatcherNodeType::COLUMNS_LIST)
+    {
+        auto identifiers = matcher_node_typed.getColumnsIdentifiers();
+        result.reserve(identifiers.size());
+
+        for (const auto & identifier : identifiers)
+        {
+            auto resolve_result = tryResolveIdentifier(IdentifierLookup{identifier, IdentifierLookupContext::EXPRESSION}, scope);
+            if (!resolve_result.isResolved())
+                throw Exception(ErrorCodes::UNKNOWN_IDENTIFIER,
+                        "Unknown identifier '{}' inside COLUMNS matcher. In scope {}",
+                        identifier.getFullName(), scope.dump());
+
+            // TODO: Introduce IdentifierLookupContext::COLUMN and get rid of this check
+            auto * resolved_column = resolve_result.resolved_identifier->as<ColumnNode>();
+            if (!resolved_column)
+                throw Exception(ErrorCodes::UNKNOWN_IDENTIFIER,
+                        "Identifier '{}' inside COLUMNS matcher must resolve into a column, but got {}. In scope {}",
+                        identifier.getFullName(),
+                        resolve_result.resolved_identifier->getNodeTypeName(),
+                        scope.scope_node->formatASTForErrorMessage());
+            result.emplace_back(resolve_result.resolved_identifier, resolved_column->getColumnName());
+        }
+        return result;
+    }
+
+    result.resize(matcher_node_typed.getColumnsIdentifiers().size());
+
     for (auto & table_expression : table_expressions_stack)
     {
         bool table_expression_in_resolve_process = scope.table_expressions_in_resolve_process.contains(table_expression.get());
@@ -4780,8 +4810,6 @@ QueryAnalyzer::QueryTreeNodesWithNames QueryAnalyzer::resolveUnqualifiedMatcher(
         table_expressions_column_nodes_with_names_stack.push_back(std::move(matched_column_nodes_with_names));
     }
 
-    QueryTreeNodesWithNames result;
-
     for (auto & table_expression_column_nodes_with_names : table_expressions_column_nodes_with_names_stack)
     {
         for (auto && table_expression_column_node_with_name : table_expression_column_nodes_with_names)

tests/queries/0_stateless/02271_fix_column_matcher_and_column_transformer.sql

@@ -61,6 +61,11 @@ CREATE TABLE github_events
 )
 ENGINE = MergeTree ORDER BY (event_type, repo_name, created_at);
 
-with top_repos as ( select repo_name from github_events where event_type = 'WatchEvent' and toDate(created_at) = today() - 1 group by repo_name order by count() desc limit 100 union distinct select repo_name from github_events where event_type = 'WatchEvent' and toMonday(created_at) = toMonday(today() - interval 1 week) group by repo_name order by count() desc limit 100 union distinct select repo_name from github_events where event_type = 'WatchEvent' and toStartOfMonth(created_at) = toStartOfMonth(today()) - interval 1 month group by repo_name order by count() desc limit 100 union distinct select repo_name from github_events where event_type = 'WatchEvent' and toYear(created_at) = toYear(today()) - 1 group by repo_name order by count() desc limit 100 ), last_day as ( select repo_name, count() as count_last_day, rowNumberInAllBlocks() + 1 as position_last_day from github_events where repo_name in (select repo_name from top_repos) and toDate(created_at) = today() - 1 group by repo_name order by count_last_day desc ), last_week as ( select repo_name, count() as count_last_week, rowNumberInAllBlocks() + 1 as position_last_week from github_events where repo_name in (select repo_name from top_repos) and toMonday(created_at) = toMonday(today()) - interval 1 week group by repo_name order by count_last_week desc ), last_month as ( select repo_name, count() as count_last_month, rowNumberInAllBlocks() + 1 as position_last_month from github_events where repo_name in (select repo_name from top_repos) and toStartOfMonth(created_at) = toStartOfMonth(today()) - interval 1 month group by repo_name order by count_last_month desc ) select d.repo_name, columns(count) from last_day d join last_week w on d.repo_name = w.repo_name join last_month m on d.repo_name = m.repo_name;
+with
+    top_repos as ( select repo_name from github_events where event_type = 'WatchEvent' and toDate(created_at) = today() - 1 group by repo_name order by count() desc limit 100 union distinct select repo_name from github_events where event_type = 'WatchEvent' and toMonday(created_at) = toMonday(today() - interval 1 week) group by repo_name order by count() desc limit 100 union distinct select repo_name from github_events where event_type = 'WatchEvent' and toStartOfMonth(created_at) = toStartOfMonth(today()) - interval 1 month group by repo_name order by count() desc limit 100 union distinct select repo_name from github_events where event_type = 'WatchEvent' and toYear(created_at) = toYear(today()) - 1 group by repo_name order by count() desc limit 100 ),
+    last_day as ( select repo_name, count() as count_last_day, rowNumberInAllBlocks() + 1 as position_last_day from github_events where repo_name in (select repo_name from top_repos) and toDate(created_at) = today() - 1 group by repo_name order by count_last_day desc ),
+    last_week as ( select repo_name, count() as count_last_week, rowNumberInAllBlocks() + 1 as position_last_week from github_events where repo_name in (select repo_name from top_repos) and toMonday(created_at) = toMonday(today()) - interval 1 week group by repo_name order by count_last_week desc ),
+    last_month as ( select repo_name, count() as count_last_month, rowNumberInAllBlocks() + 1 as position_last_month from github_events where repo_name in (select repo_name from top_repos) and toStartOfMonth(created_at) = toStartOfMonth(today()) - interval 1 month group by repo_name order by count_last_month desc )
+select d.repo_name, columns('count') from last_day d join last_week w on d.repo_name = w.repo_name join last_month m on d.repo_name = m.repo_name;
 
 DROP TABLE github_events;

tests/queries/0_stateless/03152_analyzer_columns_list.reference

@@ -0,0 +1 @@
+4	3

tests/queries/0_stateless/03152_analyzer_columns_list.sql

@@ -0,0 +1,13 @@
+CREATE TABLE test
+(
+    foo String,
+    bar String,
+)
+ENGINE = MergeTree()
+ORDER BY (foo, bar);
+
+INSERT INTO test VALUES ('foo', 'bar1');
+
+SELECT COLUMNS(bar, foo) APPLY (length) FROM test;
+
+SELECT COLUMNS(bar, foo, xyz) APPLY (length) FROM test; -- { serverError UNKNOWN_IDENTIFIER }