From 9345e9ffa5abfcf9a427b57cfc8b8084b65820e1 Mon Sep 17 00:00:00 2001 From: Boone Gorges Date: Fri, 17 Oct 2014 20:21:18 +0000 Subject: [PATCH] Overhaul SQL generating logic in WP_Meta_Query to avoid unnecessary table joins. The logic used to generate clause SQL in `WP_Meta_Query` is somewhat arcane, stemming mostly from an ongoing effort to eliminate costly table joins when they are not necessary. By systematizing the process of looking for shareable joins - as was done in `WP_Tax_Query` [29902] - it becomes possible to simplify the construction of SQL queries in `get_sql_for_clause()`. Moreover, the simplified logic is actually considerably better at identifying shareable joins, such that certain uses of `WP_Meta_Query` will see joins reduced by 50% or more. Includes integration tests for a representative cross-section of the query clause combinations that result in shared table aliases. Props boonebgorges, sc0ttkclark. See #24093. Built from https://develop.svn.wordpress.org/trunk@29940 git-svn-id: http://core.svn.wordpress.org/trunk@29691 1a063a9b-81f0-0310-95a4-ce76da25c4cd --- wp-includes/meta.php | 207 +++++++++++++++++++++++++++++-------------- 1 file changed, 142 insertions(+), 65 deletions(-) diff --git a/wp-includes/meta.php b/wp-includes/meta.php index afbf954a1c..9779391024 100644 --- a/wp-includes/meta.php +++ b/wp-includes/meta.php @@ -987,6 +987,10 @@ class WP_Meta_Query { // First-order clause. } else if ( $this->is_first_order_clause( $query ) ) { + if ( isset( $query['value'] ) && array() === $query['value'] ) { + unset( $query['value'] ); + } + $clean_queries[] = $query; // Otherwise, it's a nested query, so we recurse. @@ -1167,7 +1171,12 @@ class WP_Meta_Query { * } */ protected function get_sql_clauses() { - $sql = $this->get_sql_for_query( $this->queries ); + /* + * $queries are passed by reference to get_sql_for_query() for recursion. + * To keep $this->queries unaltered, pass a copy. + */ + $queries = $this->queries; + $sql = $this->get_sql_for_query( $queries ); if ( ! empty( $sql['where'] ) ) { $sql['where'] = ' AND ' . $sql['where']; @@ -1195,7 +1204,7 @@ class WP_Meta_Query { * @type string $where SQL fragment to append to the main WHERE clause. * } */ - protected function get_sql_for_query( $query, $depth = 0 ) { + protected function get_sql_for_query( &$query, $depth = 0 ) { $sql_chunks = array( 'join' => array(), 'where' => array(), @@ -1211,7 +1220,7 @@ class WP_Meta_Query { $indent .= " "; } - foreach ( $query as $key => $clause ) { + foreach ( $query as $key => &$clause ) { if ( 'relation' === $key ) { $relation = $query['relation']; } else if ( is_array( $clause ) ) { @@ -1278,7 +1287,7 @@ class WP_Meta_Query { * @type string $where SQL fragment to append to the main WHERE clause. * } */ - public function get_sql_for_clause( $clause, $parent_query ) { + public function get_sql_for_clause( &$clause, $parent_query ) { global $wpdb; $sql_chunks = array( @@ -1286,16 +1295,13 @@ class WP_Meta_Query { 'join' => array(), ); - $i = count( $this->table_aliases ); - $alias = $i ? 'mt' . $i : $this->meta_table; - if ( isset( $clause['compare'] ) ) { - $meta_compare = strtoupper( $clause['compare'] ); + $clause['compare'] = strtoupper( $clause['compare'] ); } else { - $meta_compare = isset( $clause['value'] ) && is_array( $clause['value'] ) ? 'IN' : '='; + $clause['compare'] = isset( $clause['value'] ) && is_array( $clause['value'] ) ? 'IN' : '='; } - if ( ! in_array( $meta_compare, array( + if ( ! in_array( $clause['compare'], array( '=', '!=', '>', '>=', '<', '<=', 'LIKE', 'NOT LIKE', 'IN', 'NOT IN', @@ -1303,58 +1309,62 @@ class WP_Meta_Query { 'EXISTS', 'NOT EXISTS', 'REGEXP', 'NOT REGEXP', 'RLIKE' ) ) ) { - $meta_compare = '='; + $clause['compare'] = '='; } - /* - * There are a number of different query structures that get - * built in different ways. - * 1. Key-only clauses - (a) clauses without a 'value' key that - * appear in the context of an OR relation and do not use - * 'NOT EXISTS' as the 'compare', or (b) clauses with an - * empty array for 'value'. - */ - if ( ! empty( $clause['key'] ) && ( - ( ! array_key_exists( 'value', $clause ) && 'NOT EXISTS' !== $meta_compare && 'OR' === $parent_query['relation'] ) || - ( isset( $clause['value'] ) && is_array( $clause['value'] ) && empty( $clause['value'] ) ) - ) ) { + $meta_compare = $clause['compare']; - $alias = $this->meta_table; - $sql_chunks['join'][] = " INNER JOIN $this->meta_table ON ($this->primary_table.$this->primary_id_column = $alias.$this->meta_id_column)"; - $sql_chunks['where'][] = $wpdb->prepare( "$this->meta_table.meta_key = %s", trim( $clause['key'] ) ); + // First build the JOIN clause, if one is required. + $join = ''; - // 2. NOT EXISTS. - } else if ( 'NOT EXISTS' === $meta_compare ) { - $join = " LEFT JOIN $this->meta_table"; - $join .= $i ? " AS $alias" : ''; - $join .= $wpdb->prepare( " ON ($this->primary_table.$this->primary_id_column = $alias.$this->meta_id_column AND $alias.meta_key = %s )", $clause['key'] ); - $sql_chunks['join'][] = $join; + // We prefer to avoid joins if possible. Look for an existing join compatible with this clause. + $alias = $this->find_compatible_table_alias( $clause, $parent_query ); + if ( false === $alias ) { + $i = count( $this->table_aliases ); + $alias = $i ? 'mt' . $i : $this->meta_table; - $sql_chunks['where'][] = $alias . '.' . $this->meta_id_column . ' IS NULL'; + // JOIN clauses for NOT EXISTS have their own syntax. + if ( 'NOT EXISTS' === $meta_compare ) { + $join .= " LEFT JOIN $this->meta_table"; + $join .= $i ? " AS $alias" : ''; + $join .= $wpdb->prepare( " ON ($this->primary_table.$this->primary_id_column = $alias.$this->meta_id_column AND $alias.meta_key = %s )", $clause['key'] ); + $sql_chunks['join'][] = $join; - // 3. EXISTS and other key-only queries. - } else if ( 'EXISTS' === $meta_compare || ( ! empty( $clause['key'] ) && ! array_key_exists( 'value', $clause ) ) ) { - $join = " INNER JOIN $this->meta_table"; - $join .= $i ? " AS $alias" : ''; - $join .= " ON ( $this->primary_table.$this->primary_id_column = $alias.$this->meta_id_column )"; - $sql_chunks['join'][] = $join; + // All other JOIN clauses. + } else { + $alias = $this->find_compatible_table_alias( $clause, $parent_query ); + if ( false === $alias ) { + $alias = $i ? 'mt' . $i : $this->meta_table; - $sql_chunks['where'][] = $wpdb->prepare( $alias . '.meta_key = %s', trim( $clause['key'] ) ); - - // 4. Clauses that have a value. - } else if ( array_key_exists( 'value', $clause ) ) { - $join = " INNER JOIN $this->meta_table"; - $join .= $i ? " AS $alias" : ''; - $join .= " ON ($this->primary_table.$this->primary_id_column = $alias.$this->meta_id_column)"; - $sql_chunks['join'][] = $join; - - if ( ! empty( $clause['key'] ) ) { - $sql_chunks['where'][] = $wpdb->prepare( "$alias.meta_key = %s", trim( $clause['key'] ) ); + $join .= " INNER JOIN $this->meta_table"; + $join .= $i ? " AS $alias" : ''; + $join .= " ON ( $this->primary_table.$this->primary_id_column = $alias.$this->meta_id_column )"; + } } - $meta_type = $this->get_cast_for_type( isset( $clause['type'] ) ? $clause['type'] : '' ); + $this->table_aliases[] = $alias; + $sql_chunks['join'][] = $join; + } - $meta_value = isset( $clause['value'] ) ? $clause['value'] : ''; + // Save the alias to this clause, for future siblings to find. + $clause['alias'] = $alias; + + // Next, build the WHERE clause. + $where = ''; + + // meta_key. + if ( array_key_exists( 'key', $clause ) ) { + if ( 'NOT EXISTS' === $meta_compare ) { + $sql_chunks['where'][] = $alias . '.' . $this->meta_id_column . ' IS NULL'; + } else { + $sql_chunks['where'][] = $wpdb->prepare( "$alias.meta_key = %s", trim( $clause['key'] ) ); + } + } + + // meta_value. + if ( array_key_exists( 'value', $clause ) ) { + $meta_value = $clause['value']; + $meta_type = $this->get_cast_for_type( isset( $clause['type'] ) ? $clause['type'] : '' ); if ( in_array( $meta_compare, array( 'IN', 'NOT IN', 'BETWEEN', 'NOT BETWEEN' ) ) ) { if ( ! is_array( $meta_value ) ) { @@ -1364,19 +1374,34 @@ class WP_Meta_Query { $meta_value = trim( $meta_value ); } - if ( 'IN' == substr( $meta_compare, -2 ) ) { - $meta_compare_string = '(' . substr( str_repeat( ',%s', count( $meta_value ) ), 1 ) . ')'; - } elseif ( 'BETWEEN' == substr( $meta_compare, -7 ) ) { - $meta_value = array_slice( $meta_value, 0, 2 ); - $meta_compare_string = '%s AND %s'; - } elseif ( 'LIKE' == $meta_compare || 'NOT LIKE' == $meta_compare ) { - $meta_value = '%' . $wpdb->esc_like( $meta_value ) . '%'; - $meta_compare_string = '%s'; - } else { - $meta_compare_string = '%s'; + switch ( $meta_compare ) { + case 'IN' : + case 'NOT IN' : + $meta_compare_string = '(' . substr( str_repeat( ',%s', count( $meta_value ) ), 1 ) . ')'; + $where = $wpdb->prepare( $meta_compare_string, $meta_value ); + break; + + case 'BETWEEN' : + case 'NOT BETWEEN' : + $meta_value = array_slice( $meta_value, 0, 2 ); + $where = $wpdb->prepare( '%s AND %s', $meta_value ); + break; + + case 'LIKE' : + case 'NOT LIKE' : + $meta_value = '%' . $wpdb->esc_like( $meta_value ) . '%'; + $where = $wpdb->prepare( '%s', $meta_value ); + break; + + default : + $where = $wpdb->prepare( '%s', $meta_value ); + break; + } - $sql_chunks['where'][] = $wpdb->prepare( "CAST($alias.meta_value AS {$meta_type}) {$meta_compare} {$meta_compare_string}", $meta_value ); + if ( $where ) { + $sql_chunks['where'][] = "CAST($alias.meta_value AS {$meta_type}) {$meta_compare} {$where}"; + } } /* @@ -1387,10 +1412,62 @@ class WP_Meta_Query { $sql_chunks['where'] = array( '( ' . implode( ' AND ', $sql_chunks['where'] ) . ' )' ); } - $this->table_aliases[] = $alias; - return $sql_chunks; } + + /** + * Identify an existing table alias that is compatible with the current query clause. + * + * We avoid unnecessary table joins by allowing each clause to look for + * an existing table alias that is compatible with the query that it + * needs to perform. An existing alias is compatible if (a) it is a + * sibling of $clause (ie, it's under the scope of the same relation), + * and (b) the combination of operator and relation between the clauses + * allows for a shared table join. In the case of WP_Meta_Query, this + * only applies to IN clauses that are connected by the relation OR. + * + * @since 4.1.0 + * @access protected + * + * @param array $clause Query clause. + * @param array $parent_query Parent query of $clause. + * @return string|bool Table alias if found, otherwise false. + */ + protected function find_compatible_table_alias( $clause, $parent_query ) { + $alias = false; + + foreach ( $parent_query as $sibling ) { + // If the sibling has no alias yet, there's nothing to check. + if ( empty( $sibling['alias'] ) ) { + continue; + } + + // We're only interested in siblings that are first-order clauses. + if ( ! is_array( $sibling ) || ! $this->is_first_order_clause( $sibling ) ) { + continue; + } + + $compatible_compares = array(); + + // Clauses connected by OR can share joins as long as they have "positive" operators. + if ( 'OR' === $parent_query['relation'] ) { + $compatible_compares = array( '=', 'IN', 'BETWEEN', 'LIKE', 'REGEXP', 'RLIKE', '>', '>=', '<', '<=' ); + + // Clauses joined by AND with "negative" operators share a join only if they also share a key. + } else if ( isset( $sibling['key'] ) && isset( $clause['key'] ) && $sibling['key'] === $clause['key'] ) { + $compatible_compares = array( '!=', 'NOT IN', 'NOT LIKE' ); + } + + $clause_compare = strtoupper( $clause['compare'] ); + $sibling_compare = strtoupper( $sibling['compare'] ); + if ( in_array( $clause_compare, $compatible_compares ) && in_array( $sibling_compare, $compatible_compares ) ) { + $alias = $sibling['alias']; + break; + } + } + + return apply_filters( 'meta_query_find_compatible_table_alias', $alias, $clause, $parent_query, $this ) ; + } } /**