/* Copyright (c) 2020, 2024, Oracle and/or its affiliates. This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License, version 2.0, as published by the Free Software Foundation. This program is designed to work with certain software (including but not limited to OpenSSL) that is licensed under separate terms, as designated in a particular file or component or in included license documentation. The authors of MySQL hereby grant you an additional permission to link the program and your derivative works with the separately licensed software that they have either included with the program or referenced in the documentation. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License, version 2.0, for more details. You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA */ #ifndef SQL_JOIN_OPTIMIZER_WALK_ACCESS_PATHS_H #define SQL_JOIN_OPTIMIZER_WALK_ACCESS_PATHS_H #include #include "sql/join_optimizer/access_path.h" #include "sql/range_optimizer/range_optimizer.h" enum class WalkAccessPathPolicy { // Stop on _any_ MATERIALIZE or STREAM path, even if they do not cross query // blocks. // Also stops on APPEND paths, which always cross query blocks. STOP_AT_MATERIALIZATION, // Stop on MATERIALIZE, STREAM or APPEND paths that cross query blocks. ENTIRE_QUERY_BLOCK, // Do not stop at any kind of access path, unless func() returns true. ENTIRE_TREE }; /** Traverse every access path below `path` (possibly limited to the current query block with the `cross_query_blocks` parameter), calling func() for each one with pre- or post-order traversal. If func() returns true, the traversal does not descend into the children of the current path. For post-order traversal, the children have already been traversed when func() is called, so it is too late to skip them, and the return value of func() is effectively ignored. The `join` parameter signifies what query block `path` is part of, since that is not implicit from the path itself. The function will track this as it changes throughout the tree (in MATERIALIZE or STREAM access paths), and will give the correct value to the func() callback. It is only used by WalkAccessPaths() itself if the policy is ENTIRE_QUERY_BLOCK; if not, it is only used for the func() callback, and you can set it to nullptr if you wish. func() must have signature func(AccessPath *, const JOIN *), or it could be JOIN * if a non-const JOIN is given in. */ template void WalkAccessPaths(AccessPathPtr path, JoinPtr join, WalkAccessPathPolicy cross_query_blocks, Func &&func, bool post_order_traversal = false) { static_assert( std::is_convertible::value, "The “path” argument must be AccessPath * or const AccessPath *."); static_assert( std::is_convertible::value, "The “join” argument must be JOIN * or const JOIN * (or nullptr)."); if (cross_query_blocks == WalkAccessPathPolicy::ENTIRE_QUERY_BLOCK) { assert(join != nullptr); } if (!post_order_traversal) { if (func(path, join)) { // Stop recursing in this branch. return; } } switch (path->type) { case AccessPath::TABLE_SCAN: case AccessPath::SAMPLE_SCAN: case AccessPath::INDEX_SCAN: case AccessPath::INDEX_DISTANCE_SCAN: case AccessPath::REF: case AccessPath::REF_OR_NULL: case AccessPath::EQ_REF: case AccessPath::PUSHED_JOIN_REF: case AccessPath::FULL_TEXT_SEARCH: case AccessPath::CONST_TABLE: case AccessPath::MRR: case AccessPath::FOLLOW_TAIL: case AccessPath::INDEX_RANGE_SCAN: case AccessPath::INDEX_SKIP_SCAN: case AccessPath::GROUP_INDEX_SKIP_SCAN: case AccessPath::DYNAMIC_INDEX_RANGE_SCAN: case AccessPath::TABLE_VALUE_CONSTRUCTOR: case AccessPath::FAKE_SINGLE_ROW: case AccessPath::ZERO_ROWS: case AccessPath::ZERO_ROWS_AGGREGATED: case AccessPath::MATERIALIZED_TABLE_FUNCTION: case AccessPath::UNQUALIFIED_COUNT: // No children. break; case AccessPath::NESTED_LOOP_JOIN: WalkAccessPaths(path->nested_loop_join().outer, join, cross_query_blocks, std::forward(func), post_order_traversal); WalkAccessPaths(path->nested_loop_join().inner, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::NESTED_LOOP_SEMIJOIN_WITH_DUPLICATE_REMOVAL: WalkAccessPaths(path->nested_loop_semijoin_with_duplicate_removal().outer, join, cross_query_blocks, std::forward(func), post_order_traversal); WalkAccessPaths(path->nested_loop_semijoin_with_duplicate_removal().inner, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::BKA_JOIN: WalkAccessPaths(path->bka_join().outer, join, cross_query_blocks, std::forward(func), post_order_traversal); WalkAccessPaths(path->bka_join().inner, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::HASH_JOIN: WalkAccessPaths(path->hash_join().outer, join, cross_query_blocks, std::forward(func), post_order_traversal); WalkAccessPaths(path->hash_join().inner, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::FILTER: WalkAccessPaths(path->filter().child, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::SORT: WalkAccessPaths(path->sort().child, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::AGGREGATE: WalkAccessPaths(path->aggregate().child, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::TEMPTABLE_AGGREGATE: WalkAccessPaths(path->temptable_aggregate().subquery_path, join, cross_query_blocks, std::forward(func), post_order_traversal); WalkAccessPaths(path->temptable_aggregate().table_path, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::LIMIT_OFFSET: WalkAccessPaths(path->limit_offset().child, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::STREAM: if (cross_query_blocks == WalkAccessPathPolicy::ENTIRE_TREE || (cross_query_blocks == WalkAccessPathPolicy::ENTIRE_QUERY_BLOCK && path->stream().join == join)) { WalkAccessPaths(path->stream().child, path->stream().join, cross_query_blocks, std::forward(func), post_order_traversal); } break; case AccessPath::MATERIALIZE: WalkAccessPaths(path->materialize().table_path, join, cross_query_blocks, std::forward(func), post_order_traversal); for (const MaterializePathParameters::Operand &operand : path->materialize().param->m_operands) { if (cross_query_blocks == WalkAccessPathPolicy::ENTIRE_TREE || (cross_query_blocks == WalkAccessPathPolicy::ENTIRE_QUERY_BLOCK && operand.join == join)) { WalkAccessPaths(operand.subquery_path, operand.join, cross_query_blocks, std::forward(func), post_order_traversal); } } break; case AccessPath::MATERIALIZE_INFORMATION_SCHEMA_TABLE: WalkAccessPaths(path->materialize_information_schema_table().table_path, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::APPEND: if (cross_query_blocks == WalkAccessPathPolicy::ENTIRE_TREE) { for (const AppendPathParameters &child : *path->append().children) { WalkAccessPaths(child.path, child.join, cross_query_blocks, std::forward(func), post_order_traversal); } } break; case AccessPath::WINDOW: WalkAccessPaths(path->window().child, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::WEEDOUT: WalkAccessPaths(path->weedout().child, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::REMOVE_DUPLICATES: WalkAccessPaths(path->remove_duplicates().child, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::REMOVE_DUPLICATES_ON_INDEX: WalkAccessPaths(path->remove_duplicates_on_index().child, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::ALTERNATIVE: WalkAccessPaths(path->alternative().child, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::CACHE_INVALIDATOR: WalkAccessPaths(path->cache_invalidator().child, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::INDEX_MERGE: for (AccessPath *child : *path->index_merge().children) { WalkAccessPaths(child, join, cross_query_blocks, std::forward(func), post_order_traversal); } break; case AccessPath::ROWID_INTERSECTION: for (AccessPath *child : *path->rowid_intersection().children) { WalkAccessPaths(child, join, cross_query_blocks, std::forward(func), post_order_traversal); } break; case AccessPath::ROWID_UNION: for (AccessPath *child : *path->rowid_union().children) { WalkAccessPaths(child, join, cross_query_blocks, std::forward(func), post_order_traversal); } break; case AccessPath::DELETE_ROWS: WalkAccessPaths(path->delete_rows().child, join, cross_query_blocks, std::forward(func), post_order_traversal); break; case AccessPath::UPDATE_ROWS: WalkAccessPaths(path->update_rows().child, join, cross_query_blocks, std::forward(func), post_order_traversal); break; } if (post_order_traversal) { if (func(path, join)) { // Stop recursing in this branch. In practice a no-op, since we are // already done with this branch. return; } } } /** A wrapper around WalkAccessPaths() that collects all tables under “root_path” and calls the given functor, stopping at materializations. This is typically used to know which tables to sort or the like. func() must have signature func(TABLE *), and return true upon error. */ template void WalkTablesUnderAccessPath(AccessPath *root_path, Func &&func, bool include_pruned_tables) { WalkAccessPaths( root_path, /*join=*/nullptr, WalkAccessPathPolicy::STOP_AT_MATERIALIZATION, [&](AccessPath *path, const JOIN *) { switch (path->type) { case AccessPath::TABLE_SCAN: return func(path->table_scan().table); case AccessPath::SAMPLE_SCAN: // SampleScan can be executed only in the secondary engine. return false; /* LCOV_EXCL_LINE */ case AccessPath::INDEX_SCAN: return func(path->index_scan().table); case AccessPath::INDEX_DISTANCE_SCAN: return func(path->index_distance_scan().table); case AccessPath::REF: return func(path->ref().table); case AccessPath::REF_OR_NULL: return func(path->ref_or_null().table); case AccessPath::EQ_REF: return func(path->eq_ref().table); case AccessPath::PUSHED_JOIN_REF: return func(path->pushed_join_ref().table); case AccessPath::FULL_TEXT_SEARCH: return func(path->full_text_search().table); case AccessPath::CONST_TABLE: return func(path->const_table().table); case AccessPath::MRR: return func(path->mrr().table); case AccessPath::FOLLOW_TAIL: return func(path->follow_tail().table); case AccessPath::INDEX_RANGE_SCAN: return func(path->index_range_scan().used_key_part[0].field->table); case AccessPath::INDEX_SKIP_SCAN: return func(path->index_skip_scan().table); case AccessPath::GROUP_INDEX_SKIP_SCAN: return func(path->group_index_skip_scan().table); case AccessPath::DYNAMIC_INDEX_RANGE_SCAN: return func(path->dynamic_index_range_scan().table); case AccessPath::STREAM: return func(path->stream().table); case AccessPath::MATERIALIZED_TABLE_FUNCTION: return func(path->materialized_table_function().table); case AccessPath::ALTERNATIVE: return func( path->alternative().table_scan_path->table_scan().table); case AccessPath::UNQUALIFIED_COUNT: // Should never be below anything that needs // WalkTablesUnderAccessPath(). assert(false); return true; case AccessPath::ZERO_ROWS: if (include_pruned_tables && path->zero_rows().child != nullptr) { WalkTablesUnderAccessPath(path->zero_rows().child, func, include_pruned_tables); } return false; case AccessPath::WINDOW: return func(path->window().temp_table); case AccessPath::AGGREGATE: case AccessPath::APPEND: case AccessPath::BKA_JOIN: case AccessPath::CACHE_INVALIDATOR: case AccessPath::FAKE_SINGLE_ROW: case AccessPath::FILTER: case AccessPath::HASH_JOIN: case AccessPath::LIMIT_OFFSET: case AccessPath::MATERIALIZE: case AccessPath::MATERIALIZE_INFORMATION_SCHEMA_TABLE: case AccessPath::NESTED_LOOP_JOIN: case AccessPath::NESTED_LOOP_SEMIJOIN_WITH_DUPLICATE_REMOVAL: case AccessPath::REMOVE_DUPLICATES: case AccessPath::REMOVE_DUPLICATES_ON_INDEX: case AccessPath::SORT: case AccessPath::TABLE_VALUE_CONSTRUCTOR: case AccessPath::TEMPTABLE_AGGREGATE: case AccessPath::WEEDOUT: case AccessPath::ZERO_ROWS_AGGREGATED: case AccessPath::INDEX_MERGE: case AccessPath::ROWID_INTERSECTION: case AccessPath::ROWID_UNION: case AccessPath::DELETE_ROWS: case AccessPath::UPDATE_ROWS: return false; } assert(false); return true; }); } #endif // SQL_JOIN_OPTIMIZER_WALK_ACCESS_PATHS_H