From 2531f722905e9c377d758cba19fe0171f031fdd1 Mon Sep 17 00:00:00 2001 From: cnnblike Date: Wed, 26 Feb 2025 17:38:46 +0800 Subject: [PATCH] Revert "fix(DDL): improve nested functional index support" This reverts commit 12ebf3541025ac39f608faa7232b583069bfcb35. --- mysql-test/mysql-source-code-meta.patch | 52 +--- .../suite/ctc/r/ctc_ddl_func_index.result | 40 --- .../suite/ctc/t/ctc_ddl_func_index.test | 15 +- storage/ctc/ha_ctc_ddl.cc | 250 ++++-------------- storage/ctc/ha_ctc_ddl.h | 18 -- 5 files changed, 57 insertions(+), 318 deletions(-) diff --git a/mysql-test/mysql-source-code-meta.patch b/mysql-test/mysql-source-code-meta.patch index b0c1936..5794762 100644 --- a/mysql-test/mysql-source-code-meta.patch +++ b/mysql-test/mysql-source-code-meta.patch @@ -1120,56 +1120,6 @@ index 0530ca15..efeaf90e 100644 enum_binlog_command binlog_command, const char *query, size_t query_length, const char *db, const char *table_name); -diff --git a/sql/item_json_func.cc b/sql/item_json_func.cc -index 31aade1c..8a1012e4 100644 ---- a/sql/item_json_func.cc -+++ b/sql/item_json_func.cc -@@ -3656,7 +3656,7 @@ bool Item_func_array_cast::fix_fields(THD *thd, Item **ref) { - @param item the Item in which the cast operation is performed - @param[out] str the string to print to - */ --static void print_cast_type(Cast_target cast_type, const Item *item, -+void print_cast_type(Cast_target cast_type, const Item *item, - String *str) { - const unsigned decimals = item->decimals; - switch (cast_type) { -diff --git a/sql/item_json_func.h b/sql/item_json_func.h -index 791e9b3e..7bc26c04 100644 ---- a/sql/item_json_func.h -+++ b/sql/item_json_func.h -@@ -1103,6 +1103,9 @@ class Item_func_json_value final : public Item_func { - my_decimal *val_decimal(my_decimal *value) override; - bool get_date(MYSQL_TIME *ltime, my_time_flags_t flags) override; - bool get_time(MYSQL_TIME *ltime) override; -+ Json_on_response_type m_on_empty; -+ Json_on_response_type m_on_error; -+ Cast_target m_cast_target; - - private: - /// Represents a default value given in JSON_VALUE's DEFAULT xxx ON EMPTY or -@@ -1112,15 +1115,12 @@ class Item_func_json_value final : public Item_func { - /// Parsed path. - Json_path m_path_json; - /// Type of the ON EMPTY clause. -- Json_on_response_type m_on_empty; - /// Type of the ON ERROR clause. -- Json_on_response_type m_on_error; - /// The default value for ON EMPTY (if not ERROR or NULL ON EMPTY). - unique_ptr_destroy_only m_default_empty; - /// The default value for ON EMPTY (if not ERROR or NULL ON EMPTY). - unique_ptr_destroy_only m_default_error; - /// The target data type. -- Cast_target m_cast_target; - - /** - Creates a Json_value_default object representing the default value given in -@@ -1244,4 +1244,6 @@ bool sort_and_remove_dups(const Json_wrapper &orig, Sorted_index_array *v); - - bool save_json_to_field(THD *thd, Field *field, const Json_wrapper *w, - bool no_error); -+void print_cast_type(Cast_target cast_type, const Item *item, -+ String *str); - #endif /* ITEM_JSON_FUNC_INCLUDED */ diff --git a/sql/locking_service.cc b/sql/locking_service.cc index 9ca8e21a..ef215f3d 100644 --- a/sql/locking_service.cc @@ -3942,6 +3892,6 @@ index 19110c7a..131a4c69 100644 ELSE() # This is a fall-back. FILE(WRITE ${INFO_SRC} "\nMySQL source ${VERSION}\n") -+ FILE(APPEND ${INFO_SRC} "\nCantian patch source 1.0.2\n") ++ FILE(APPEND ${INFO_SRC} "\nCantian patch source 1.0.1\n") ENDIF() ENDMACRO(CREATE_INFO_SRC) \ No newline at end of file diff --git a/mysql-test/suite/ctc/r/ctc_ddl_func_index.result b/mysql-test/suite/ctc/r/ctc_ddl_func_index.result index c61b0af..d8e34ff 100644 --- a/mysql-test/suite/ctc/r/ctc_ddl_func_index.result +++ b/mysql-test/suite/ctc/r/ctc_ddl_func_index.result @@ -49,48 +49,8 @@ id select_type table partitions type possible_keys key key_len ref rows filtered 1 SIMPLE t1 NULL ref func_index_2 func_index_2 7 const 4 100.00 NULL Warnings: Note 1003 /* select#1 */ select `db1`.`t1`.`c1` AS `c1`,`db1`.`t1`.`c2` AS `c2`,`db1`.`t1`.`c3` AS `c3`,`db1`.`t1`.`c4` AS `c4` from `db1`.`t1` where (substr(`c4`,1,1) = 'a') -create index func_index_3 on t1 ((substr(upper(c4), 1, 1))); -select * from t1 where substr(upper(c4), 1, 1) = 'a'; -c1 c2 c3 c4 -1 1 aaa aaa -2 2 aaA aBB -4 4 aaBa Aaa -explain select * from t1 where substr(upper(c4), 1, 1) = 'a'; -id select_type table partitions type possible_keys key key_len ref rows filtered Extra -1 SIMPLE t1 NULL ref func_index_3 func_index_3 7 const 1 100.00 NULL -Warnings: -Note 1003 /* select#1 */ select `db1`.`t1`.`c1` AS `c1`,`db1`.`t1`.`c2` AS `c2`,`db1`.`t1`.`c3` AS `c3`,`db1`.`t1`.`c4` AS `c4` from `db1`.`t1` where (substr(upper(`c4`),1,1) = 'a') -analyze table t1; -Table Op Msg_type Msg_text -db1.t1 analyze status OK -explain select * from t1 where substr(upper(c4), 1, 1) = 'a'; -id select_type table partitions type possible_keys key key_len ref rows filtered Extra -1 SIMPLE t1 NULL ref func_index_3 func_index_3 7 const 4 100.00 NULL -Warnings: -Note 1003 /* select#1 */ select `db1`.`t1`.`c1` AS `c1`,`db1`.`t1`.`c2` AS `c2`,`db1`.`t1`.`c3` AS `c3`,`db1`.`t1`.`c4` AS `c4` from `db1`.`t1` where (substr(upper(`c4`),1,1) = 'a') -create index func_index_4 on t1 ((upper(substr(c4, 1, 1)))); -select * from t1 where upper(substr(c4, 1, 1)) = 'a'; -c1 c2 c3 c4 -1 1 aaa aaa -2 2 aaA aBB -4 4 aaBa Aaa -explain select * from t1 where upper(substr(c4, 1, 1)) = 'a'; -id select_type table partitions type possible_keys key key_len ref rows filtered Extra -1 SIMPLE t1 NULL ref func_index_4 func_index_4 7 const 1 100.00 NULL -Warnings: -Note 1003 /* select#1 */ select `db1`.`t1`.`c1` AS `c1`,`db1`.`t1`.`c2` AS `c2`,`db1`.`t1`.`c3` AS `c3`,`db1`.`t1`.`c4` AS `c4` from `db1`.`t1` where (upper(substr(`c4`,1,1)) = 'a') -analyze table t1; -Table Op Msg_type Msg_text -db1.t1 analyze status OK -explain select * from t1 where upper(substr(c4, 1, 1)) = 'a'; -id select_type table partitions type possible_keys key key_len ref rows filtered Extra -1 SIMPLE t1 NULL ref func_index_4 func_index_4 7 const 4 100.00 NULL -Warnings: -Note 1003 /* select#1 */ select `db1`.`t1`.`c1` AS `c1`,`db1`.`t1`.`c2` AS `c2`,`db1`.`t1`.`c3` AS `c3`,`db1`.`t1`.`c4` AS `c4` from `db1`.`t1` where (upper(substr(`c4`,1,1)) = 'a') alter table t1 drop index func_index_1; drop index func_index_2 on t1; -drop index func_index_3 on t1; -drop index func_index_4 on t1; create table t2 (c1 int, c2 int, c3 varchar(10)); insert into t2 values (1, 1, 'aaa'), (2, 2, 'aaA'), (3, 3, 'AAA'), (4, 4, 'aaBa'); select * from t2; diff --git a/mysql-test/suite/ctc/t/ctc_ddl_func_index.test b/mysql-test/suite/ctc/t/ctc_ddl_func_index.test index df44be2..879c0c2 100644 --- a/mysql-test/suite/ctc/t/ctc_ddl_func_index.test +++ b/mysql-test/suite/ctc/t/ctc_ddl_func_index.test @@ -20,23 +20,10 @@ explain select * from t1 where substr(c4, 1, 1) = 'a'; analyze table t1; explain select * from t1 where substr(c4, 1, 1) = 'a'; -# 创建嵌套函数索引场景 -create index func_index_3 on t1 ((substr(upper(c4), 1, 1))); -select * from t1 where substr(upper(c4), 1, 1) = 'a'; -explain select * from t1 where substr(upper(c4), 1, 1) = 'a'; -analyze table t1; -explain select * from t1 where substr(upper(c4), 1, 1) = 'a'; -create index func_index_4 on t1 ((upper(substr(c4, 1, 1)))); -select * from t1 where upper(substr(c4, 1, 1)) = 'a'; -explain select * from t1 where upper(substr(c4, 1, 1)) = 'a'; -analyze table t1; -explain select * from t1 where upper(substr(c4, 1, 1)) = 'a'; - # 删除函数索引 alter table t1 drop index func_index_1; drop index func_index_2 on t1; -drop index func_index_3 on t1; -drop index func_index_4 on t1; + # 创建函数索引异常场景 create table t2 (c1 int, c2 int, c3 varchar(10)); diff --git a/storage/ctc/ha_ctc_ddl.cc b/storage/ctc/ha_ctc_ddl.cc index f922c6f..30205b9 100644 --- a/storage/ctc/ha_ctc_ddl.cc +++ b/storage/ctc/ha_ctc_ddl.cc @@ -48,9 +48,6 @@ #include "sql/sql_table.h" // primary_key_name #include "sql/sql_partition.h" #include "sql/item_func.h" -#include "sql/item_json_func.h" -#include "sql/table_function.h" - #include "my_time.h" #include "decimal.h" @@ -1250,7 +1247,7 @@ static int ctc_check_func_name(std::string func_name) const std::string mysql_func_name = mysql_func_name_to_ctc_map[i].mysql_func_name; if (func_name_len >= mysql_func_name.length() && func_name.compare(0, mysql_func_name.length(), mysql_func_name) == 0) { - my_printf_error(ER_DISALLOWED_OPERATION, "Function %s is not indexable", MYF(0), + my_printf_error(ER_DISALLOWED_OPERATION, "Function %s is not indexable", MYF(0), mysql_func_name_to_ctc_map[i].ctc_func_name.c_str()); return CT_ERROR; } @@ -1258,204 +1255,67 @@ static int ctc_check_func_name(std::string func_name) return CT_SUCCESS; } -static int recursively_get_dependency_item(TABLE *form, Item_func *item_func, TcDb__CtcDDLTableKeyPart *req_key_part, - uint32_t *col_item_count) -{ - for (uint32_t i = 0; i < item_func->arg_count; i++) { - if (item_func->get_arg(i)->type() == Item::FUNC_ITEM) { - // nested function, would go recursively to do detection - int result = recursively_get_dependency_item(form, (Item_func *) item_func->get_arg(i), - req_key_part, col_item_count); - if (result != CT_SUCCESS) { - return result; - } - } - - if (item_func->get_arg(i)->type() == Item::FIELD_ITEM) { - // the field* args[i] contains don't have proper m_field_index when it's a alter table scenario. - // thus we have to look up by name through metadata on the new TABLE* - Item_field *arg_item_field = (Item_field*) item_func->get_arg(i); - Field *field = ctc_get_field_by_name(form, arg_item_field->field->field_name); - if (field && field->is_gcol()) { - my_printf_error(ER_DISALLOWED_OPERATION, "%s", MYF(0), - "Cantian does not support index on generated column."); - return CT_ERROR; - } - - if (*col_item_count >= 1) { - my_printf_error(ER_DISALLOWED_OPERATION, "%s", MYF(0), - "Cantian does not support function indexes with multiple columns of arguments."); - return CT_ERROR; - } - req_key_part->name = const_cast(item_func->get_arg(i)->item_name.ptr()); - (*col_item_count)++; - } - } - return CT_SUCCESS; -} - -void ctc_rewrite_expression_clause(TABLE *form, const THD *thd, Item *item, String *out); - -void ctc_print_on_empty_or_error(TABLE *form, const THD *thd, String *out, bool on_empty, - Json_on_response_type response_type, Item *default_string) -{ - switch (response_type) { - case Json_on_response_type::ERROR: - out->append(STRING_WITH_LEN(" error")); - break; - case Json_on_response_type::NULL_VALUE: - out->append(STRING_WITH_LEN(" null")); - break; - case Json_on_response_type::DEFAULT: - out->append(STRING_WITH_LEN(" default ")); - ctc_rewrite_expression_clause(form, thd, default_string, out); - break; - case Json_on_response_type::IMPLICIT: - // Nothing to print when the clause was implicit. - return; - }; - - if (on_empty) { - out->append(STRING_WITH_LEN(" on empty")); - } else { - out->append(STRING_WITH_LEN(" on error")); - } -} - -#define JSON_VALUE_ON_EMPTY_ARG_IDX 2 -#define JSON_VALUE_ON_ERROR_ARG_IDX 3 - -// equivalent to Item_func_json_value::print, but stripped charset info -void ctc_item_print_json_value(TABLE *form, const THD *thd, Item* item, String *out) -{ - Item_func_json_value *item_func = (Item_func_json_value *) item; - out->append(STRING_WITH_LEN("json_value(")); - ctc_rewrite_expression_clause(form, thd, item_func->get_arg(0), out); - out->append(STRING_WITH_LEN(", ")); - ctc_rewrite_expression_clause(form, thd, item_func->get_arg(1), out); - out->append(STRING_WITH_LEN(" returning ")); - if (item_func->m_cast_target == ITEM_CAST_CHAR && item_func->collation.collation != &my_charset_bin) { - // don't add char, use CLOB instead - out->append(STRING_WITH_LEN("CLOB")); - } else { - print_cast_type(item_func->m_cast_target, item_func, out); - } - ctc_print_on_empty_or_error(form, thd, out, true, item_func->m_on_empty, - item_func->get_arg(JSON_VALUE_ON_EMPTY_ARG_IDX)); - ctc_print_on_empty_or_error(form, thd, out, false, item_func->m_on_error, - item_func->get_arg(JSON_VALUE_ON_ERROR_ARG_IDX)); - out->append(')'); -}; - -std::map item_func_printer = { - {"json_value", (ctc_item_print_t) ctc_item_print_json_value} -}; - -static void ctc_print_op(TABLE *form, const THD *thd, Item_func *item_func, String *out) -{ - out->append('('); - for (uint i = 0; i < item_func->arg_count - 1; i++) { - ctc_rewrite_expression_clause(form, thd, item_func->get_arg(i), out); - out->append(' '); - out->append(item_func->func_name()); - out->append(' '); - } - ctc_rewrite_expression_clause(form, thd, item_func->get_arg(item_func->arg_count - 1), out); - out->append(')'); -} - -static void ctc_print_func(TABLE *form, const THD *thd, Item_func *item_func, String *out) -{ - out->append(item_func->func_name()); - out->append('('); - for (uint i = 0; i < item_func->arg_count; i++) { - if (i != 0) out->append(','); - ctc_rewrite_expression_clause(form, thd, item_func->get_arg(i), out); - } - out->append(')'); -} - -// the origin implementation of Item_field::print would add extra back quote (`) when it's called, -// due to append_identifier() would explicitly use get_quote_char_for_identifier() to get quote char. -// meanwhile ctsql don't have support for that, ctc_print_field shall be used instead. - -static void ctc_print_field(TABLE *form MY_ATTRIBUTE((unused)), const THD *thd MY_ATTRIBUTE((unused)), - Item_ident *item_ident, String *out) +static uint32_t ctc_fill_func_key_part(TABLE *form, THD *thd, TcDb__CtcDDLTableKeyPart *req_key_part, Value_generator *gcol_info) { - // equivalent to append_identifier(thd, out, item_ident->field_name, strlen(item_ident->field_name), NULL, NULL) - // but without quote_char support - out->append(item_ident->field_name, strlen(item_ident->field_name), system_charset_info); -} - -// this expression rewrite process may only happen once during DDL, performance is not major concern -void ctc_rewrite_expression_clause(TABLE *form, const THD *thd, Item *item, String *out) -{ - Item::Type item_type = item->type(); - if (item_type == Item::Type::FIELD_ITEM) { - ctc_print_field(form, thd, (Item_ident*) item, out); - return; - } - - // for function item, in order to intercept all print() call to its args, a ctc_print_func is used instead of - // the built-in print() - if (item_type == Item::Type::FUNC_ITEM) { - std::string func_name_string = std::string(((Item_func *)item)->func_name()); - if (print_op_func_name.find(func_name_string) != print_op_func_name.end()) { - ctc_print_op(form, thd, (Item_func*) item, out); - } else if (item_func_printer.find(func_name_string) != item_func_printer.end()) { - ctc_item_print_t printer = item_func_printer[func_name_string]; - printer(form, thd, (Item_func *)item, out); - } else { - ctc_print_func(form, thd, (Item_func *)item, out); - } - return; - } - - // otherwise, it should be int literals / string literals etc, use the built-in print - // but with QT_WITHOUT_INTRODUCERS so we don't have introducers like _utf8mb4, which is not supported by ctsql - item->print(thd, out, enum_query_type(QT_WITHOUT_INTRODUCERS | QT_NO_DB | QT_NO_TABLE)); -} - -static int ctc_fill_func_key_part(TABLE *form, THD *thd, - TcDb__CtcDDLTableKeyPart *req_key_part, Value_generator *gcol_info) -{ - Item_func *func_expr_item = dynamic_cast(gcol_info->expr_item); - if (func_expr_item == nullptr) { - my_printf_error(ER_DISALLOWED_OPERATION, "%s", MYF(0), - "[CTC_CREATE_TABLE]: CTC do not support this functional index."); - return CT_ERROR; - } + Item_func *func_expr_item = dynamic_cast(gcol_info->expr_item); + if (func_expr_item == nullptr) { + my_printf_error(ER_DISALLOWED_OPERATION, "%s", MYF(0), + "[CTC_CREATE_TABLE]: CTC do not support this functional index."); + return CT_ERROR; + } - uint32_t arg_count = func_expr_item->arg_count; - if (arg_count == 0) { - my_printf_error(ER_DISALLOWED_OPERATION, "%s", MYF(0), - "[CTC_CREATE_TABLE]: There is no functional index."); - return CT_ERROR; - } + uint32_t arg_count = func_expr_item->arg_count; + if (arg_count == 0) { + my_printf_error(ER_DISALLOWED_OPERATION, "%s", MYF(0), + "[CTC_CREATE_TABLE]: There is no functional index."); + return CT_ERROR; + } - if (ctc_check_func_name(std::string(func_expr_item->func_name())) != CT_SUCCESS) { - return CT_ERROR; + if (ctc_check_func_name(std::string(func_expr_item->func_name())) != CT_SUCCESS) { + return CT_ERROR; + } + + req_key_part->is_func = true; + req_key_part->func_name = const_cast(func_expr_item->func_name()); + Item **args = func_expr_item->arguments(); + uint32_t col_item_count = 0; + Field *field = nullptr; + for (uint32_t i = 0; i < arg_count; i++) { + field = ctc_get_field_by_name(form, const_cast(args[i]->item_name.ptr())); + if (field && field->is_gcol()) { + my_printf_error(ER_DISALLOWED_OPERATION, "%s", MYF(0), + "Cantian does not support index on generated column."); + return CT_ERROR; } - - req_key_part->is_func = true; - req_key_part->func_name = const_cast(func_expr_item->func_name()); - - uint32_t col_item_count = 0; - // recursively_get_dependency_item fill req_key_part->name - int result = recursively_get_dependency_item(form, func_expr_item, req_key_part, &col_item_count); - if (result != CT_SUCCESS || col_item_count != 1) { + if (args[i]->type() == Item::FIELD_ITEM) { + if (col_item_count >= 1) { my_printf_error(ER_DISALLOWED_OPERATION, "%s", MYF(0), - "[CTC_CREATE_TABLE]: CTC do not support this functional index."); + "Cantian does not support function indexes with multiple columns of arguments."); return CT_ERROR; + } + req_key_part->name = const_cast(args[i]->item_name.ptr()); + col_item_count++; } - - char buffer[FUNC_TEXT_MAX_LEN] = {0}; - String gc_expr(buffer, sizeof(buffer), &my_charset_bin); - - gc_expr.length(0); - ctc_rewrite_expression_clause(form, thd, gcol_info->expr_item, &gc_expr); - strncpy(req_key_part->func_text, gc_expr.c_ptr(), FUNC_TEXT_MAX_LEN - 1); - return CT_SUCCESS; + } + + char buffer[FUNC_TEXT_MAX_LEN] = {0}; + String gc_expr(buffer, sizeof(buffer), &my_charset_bin); + gcol_info->print_expr(thd, &gc_expr); + string expr_str(buffer); + expr_str.erase(remove(expr_str.begin(), expr_str.end(), '`'), expr_str.end()); + // 处理json_value建索引，只允许returning char + // 不带returning默认char512 + if (strcmp(req_key_part->func_name, "json_value") == 0) { + std::regex reg_char("returning[ ]char[(]\\d+[)]"); + std::regex reg_charset("[_][a-z]+[0-9]*[a-z]*[0-9]*['$]"); + std::regex reg_charset2("[ ]character[ ]set[ ][a-z]+[0-9]*[a-z]*[0-9]"); + expr_str =std::regex_replace(expr_str, reg_char, "returning CLOB"); + expr_str =std::regex_replace(expr_str, reg_charset, "'"); + //处理char带charset设置 + expr_str =std::regex_replace(expr_str, reg_charset2, ""); + } + strncpy(req_key_part->func_text, expr_str.c_str(), FUNC_TEXT_MAX_LEN - 1); + return CT_SUCCESS; } static inline longlong get_session_level_create_index_parallelism(THD *thd) diff --git a/storage/ctc/ha_ctc_ddl.h b/storage/ctc/ha_ctc_ddl.h index af3bc48..8d6bc1a 100644 --- a/storage/ctc/ha_ctc_ddl.h +++ b/storage/ctc/ha_ctc_ddl.h @@ -21,8 +21,6 @@ #include #include #include "storage/ctc/ha_ctcpart.h" -#include "sql/item_json_func.h" -#include "sql/table_function.h" #define UN_SUPPORT_DDL "ddl statement" /** Max table name length as defined in CT_MAX_NAME_LEN */ #define CTC_MAX_TABLE_NAME_LEN 64 @@ -174,22 +172,6 @@ static map mysql_collate_num_to_ctc_type = { {76, COLLATE_UTF8_TOLOWER_CI}, }; -typedef void (*ctc_item_print_t) (TABLE *form, const THD *thd, Item* item, String *out); -// the mapping relationship between rewroten Item sub-class print() - -// as for MySQL 8.0.26, only following sub class of Item would be using print_op -// 1. Item_func_int_div ("DIV"), -// 2. sub class of Item_num_op, -// * Item_func_div: / ; Item_func_mul: * ; Item_func_mod: % ; -// * Item_func_additive_op: Item_func_plus: + ; Item_func_minus: - -// 3. sub class of Item_func_bit, -// * Item_func_bit_two_param: Item_func_bit_or: | ; Item_func_bit_and: & ; Item_func_bit_xor: ^ -// * sub class of Item_func_shift: Item_func_shift_left: << ; Item_func_shift_right: >> -// * Item_func_bit_neg: ~ -static std::set print_op_func_name = { - "DIV", "/", "*", "%", "+", "-", "|", "&", "^", "<<", ">>", "~" -}; - typedef struct { char base_name[SMALL_RECORD_SIZE]; char var_name[SMALL_RECORD_SIZE]; -- Gitee