/* Copyright (c) 2024, MariaDB This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; version 2 of the License. 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 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 Street, Fifth Floor, Boston, MA 02110-1335 USA */ /** @file @brief This file contains SargableLeft optimization */ #include "mariadb.h" #include "sql_priv.h" #include #include "sql_select.h" /* SargableLeft ============ This optimization makes conditions in forms like LEFT(key_col, N) = 'string_const' SUBSTRING(key_col, 1, N) = 'string_const' sargable. The conditions take the first N characters of key_col and compare them with a string constant. However, producing index lookup intervals for this collation is complex due to contractions. Contractions ------------ A contraction is a property of collation where a sequence of multiple characters is compared as some other character(s). For example, in utfmb4_danish_ci, 'AA' is compared as one character 'Å' which sorts after 'Z': MariaDB [test]> select a from t1 order by col1; +------+ | col1 | +------+ | BA1 | (1) | BC | | BZ | | BAA2 | (2) +------+ Now suppose we're producing lookup ranges for condition LEFT(col1, 2)='BA' In addition to looking near 'BA' (1), we need to look into the area right after 'BZ' (2), where we may find 'BAA'. Fortunately, this was already implemented for handling LIKE conditions in form 'key_col LIKE 'BA%'. Each collation provides like_range() call which produces lookup range in a collation-aware way. Differences between LIKE and LEFT= ---------------------------------- So can one reduce or even rewrite conditions with LEFT() into LIKE? No, there are differences. First, LIKE does character-by-character comparison, ignoring the collation's contractions: MariaDB [test]> select col1, col1='AA', col1 LIKE 'AA' from t1; +------+-----------+----------------+ | col1 | col1='AA' | col1 LIKE 'AA' | +------+-----------+----------------+ | AA | 1 | 1 | | Å | 1 | 0 | +------+-----------+----------------+ (However, index comparison function uses equality's comparison rules. my_like_range() will produce an index range 'AA' <= col1 <= 'AA'. Reading rows from it will return 'Å' as well) Second, LEFT imposes additional constraints on the length of both parts. For example: - LEFT(col,2)='string-longer-than-two-chars' - is false for any value of col. - LEFT(col,2)='A' is not equivalent to (col LIKE 'A%'), consider col='Ab'. Take-aways ---------- - SargableLeft makes use of my_like_range() to produce index intervals. - LEFT(col, N)='foo' - We ignore the value of N when producing the lookup range (this may make the range to include rows for which the predicate is false) = For the SUBSTRING form, we only need to check that M=1 in the SUBSTRING(col, M, N)='foo'. */ /* @brief Check if this condition is sargable LEFT(key_col, N)='foo', or similar condition with SUBSTRING(). @detail 'foo' here can be any constant we can compute during optimization, Only equality conditions are supported. See SargableLeft above for detals. @param field The first argument of LEFT or SUBSTRING if sargable, otherwise deferenced to NULL @param value_idx The index of argument that is the prefix string if sargable, otherwise dereferenced to -1 */ bool Item_bool_func::with_sargable_substr(Item_field **field, int *value_idx) const { int func_idx, val_idx= -1; Item **func_args, *real= NULL; bool ret= false; enum Functype type; if (functype() != EQ_FUNC) goto done; if (args[0]->type() == FUNC_ITEM) func_idx= 0; else if (args[1]->type() == FUNC_ITEM) func_idx= 1; else goto done; type= ((Item_func *) args[func_idx])->functype(); if (type != SUBSTR_FUNC && type != LEFT_FUNC) goto done; func_args= ((Item_func *) args[func_idx])->arguments(); real= func_args[0]->real_item(); val_idx= 1 - func_idx; if (real->type() == FIELD_ITEM && args[val_idx]->can_eval_in_optimize() && (type == LEFT_FUNC || func_args[1]->val_int() == 1)) { ret= true; goto done; } real= NULL; val_idx= -1; done: if (field != NULL) *field= (Item_field *) real; if (value_idx != NULL) *value_idx= val_idx; return ret; }