/* Copyright (c) 2000, 2018, Oracle and/or its affiliates. All rights reserved.
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 St, Fifth Floor, Boston, MA 02110-1301 USA */
/**
@file
@brief
This file defines all compare functions
*/
#include
#include "sql_select.h"
#include "sql_optimizer.h" // JOIN_TAB
#include "sql_parse.h" // check_stack_overrun
#include "sql_time.h" // make_truncated_value_warning
#include "opt_trace.h"
#include "parse_tree_helpers.h"
#include "template_utils.h"
#include "item_json_func.h" // json_value, get_json_atom_wrapper
#include
using std::min;
using std::max;
#include "aggregate_check.h"
static bool convert_constant_item(THD *, Item_field *, Item **);
static longlong
get_year_value(THD *thd, Item ***item_arg, Item **cache_arg,
const Item *warn_item, bool *is_null);
static Item_result item_store_type(Item_result a, Item *item,
my_bool unsigned_flag)
{
Item_result b= item->result_type();
if (a == STRING_RESULT || b == STRING_RESULT)
return STRING_RESULT;
else if (a == REAL_RESULT || b == REAL_RESULT)
return REAL_RESULT;
else if (a == DECIMAL_RESULT || b == DECIMAL_RESULT ||
unsigned_flag != item->unsigned_flag)
return DECIMAL_RESULT;
else
return INT_RESULT;
}
static void agg_result_type(Item_result *type, my_bool *unsigned_flag,
Item **items, uint nitems)
{
Item **item, **item_end;
*type= STRING_RESULT;
*unsigned_flag= FALSE;
/* Skip beginning NULL items */
for (item= items, item_end= item + nitems; item < item_end; item++)
{
if ((*item)->type() != Item::NULL_ITEM)
{
*type= (*item)->result_type();
*unsigned_flag= (*item)->unsigned_flag;
item++;
break;
}
}
/* Combine result types. Note: NULL items don't affect the result */
for (; item < item_end; item++)
{
if ((*item)->type() != Item::NULL_ITEM)
{
*type= item_store_type(*type, *item, *unsigned_flag);
*unsigned_flag= *unsigned_flag && (*item)->unsigned_flag;
}
}
}
/*
Compare row signature of two expressions
SYNOPSIS:
cmp_row_type()
item1 the first expression
item2 the second expression
DESCRIPTION
The function checks that two expressions have compatible row signatures
i.e. that the number of columns they return are the same and that if they
are both row expressions then each component from the first expression has
a row signature compatible with the signature of the corresponding component
of the second expression.
RETURN VALUES
1 type incompatibility has been detected
0 otherwise
*/
static int cmp_row_type(Item* item1, Item* item2)
{
uint n= item1->cols();
if (item2->check_cols(n))
return 1;
for (uint i=0; ielement_index(i)->check_cols(item1->element_index(i)->cols()) ||
(item1->element_index(i)->result_type() == ROW_RESULT &&
cmp_row_type(item1->element_index(i), item2->element_index(i))))
return 1;
}
return 0;
}
/**
Aggregates result types from the array of items.
SYNOPSIS:
agg_cmp_type()
type [out] the aggregated type
items array of items to aggregate the type from
nitems number of items in the array
DESCRIPTION
This function aggregates result types from the array of items. Found type
supposed to be used later for comparison of values of these items.
Aggregation itself is performed by the item_cmp_type() function.
@param[out] type the aggregated type
@param items array of items to aggregate the type from
@param nitems number of items in the array
@retval
1 type incompatibility has been detected
@retval
0 otherwise
*/
static int agg_cmp_type(Item_result *type, Item **items, uint nitems)
{
uint i;
type[0]= items[0]->result_type();
for (i= 1 ; i < nitems ; i++)
{
type[0]= item_cmp_type(type[0], items[i]->result_type());
/*
When aggregating types of two row expressions we have to check
that they have the same cardinality and that each component
of the first row expression has a compatible row signature with
the signature of the corresponding component of the second row
expression.
*/
if (type[0] == ROW_RESULT && cmp_row_type(items[0], items[i]))
return 1; // error found: invalid usage of rows
}
return 0;
}
/**
@brief Aggregates field types from the array of items.
@param[in] items array of items to aggregate the type from
@paran[in] nitems number of items in the array
@details This function aggregates field types from the array of items.
Found type is supposed to be used later as the result field type
of a multi-argument function.
Aggregation itself is performed by the Field::field_type_merge()
function.
@note The term "aggregation" is used here in the sense of inferring the
result type of a function from its argument types.
@return aggregated field type.
*/
enum_field_types agg_field_type(Item **items, uint nitems)
{
uint i;
if (!nitems || items[0]->result_type() == ROW_RESULT )
return (enum_field_types)-1;
enum_field_types res= items[0]->field_type();
for (i= 1 ; i < nitems ; i++)
res= Field::field_type_merge(res, items[i]->field_type());
return real_type_to_type(res);
}
/*
Collects different types for comparison of first item with each other items
SYNOPSIS
collect_cmp_types()
items Array of items to collect types from
nitems Number of items in the array
skip_nulls Don't collect types of NULL items if TRUE
DESCRIPTION
This function collects different result types for comparison of the first
item in the list with each of the remaining items in the 'items' array.
RETURN
0 - if row type incompatibility has been detected (see cmp_row_type)
Bitmap of collected types - otherwise
*/
static uint collect_cmp_types(Item **items, uint nitems, bool skip_nulls= FALSE)
{
uint i;
uint found_types;
Item_result left_result= items[0]->result_type();
DBUG_ASSERT(nitems > 1);
found_types= 0;
for (i= 1; i < nitems ; i++)
{
if (skip_nulls && items[i]->type() == Item::NULL_ITEM)
continue; // Skip NULL constant items
if ((left_result == ROW_RESULT ||
items[i]->result_type() == ROW_RESULT) &&
cmp_row_type(items[0], items[i]))
return 0;
found_types|= 1U result_type());
}
/*
Even if all right-hand items are NULLs and we are skipping them all, we need
at least one type bit in the found_type bitmask.
*/
if (skip_nulls && !found_types)
found_types= 1U name,c1.derivation_name(),
c2.collation->name,c2.derivation_name(),
fname);
}
/**
This implementation of the factory method also implements flattening of
row constructors. Examples of flattening are:
- ROW(a, b) op ROW(x, y) => a op x P b op y.
- ROW(a, ROW(b, c) op ROW(x, ROW(y, z))) => a op x P b op y P c op z.
P is either AND or OR, depending on the comparison operation, and this
detail is left for combine().
The actual operator @i op is created by the concrete subclass in
create_scalar_predicate().
*/
Item_bool_func *Linear_comp_creator::create(Item *a, Item *b) const
{
/*
Test if the arguments are row constructors and thus can be flattened into
a list of ANDs or ORs.
*/
if (a->type() == Item::ROW_ITEM && b->type() == Item::ROW_ITEM)
{
if (a->cols() != b->cols())
{
my_error(ER_OPERAND_COLUMNS, MYF(0), a->cols());
return NULL;
}
DBUG_ASSERT(a->cols() > 1);
List list;
for (uint i= 0; i < a->cols(); ++i)
list.push_back(create(a->element_index(i), b->element_index(i)));
return combine(list);
}
return create_scalar_predicate(a, b);
}
Item_bool_func *Eq_creator::create_scalar_predicate(Item *a, Item *b) const
{
DBUG_ASSERT(a->type() != Item::ROW_ITEM || b->type() != Item::ROW_ITEM);
return new Item_func_eq(a, b);
}
Item_bool_func *Eq_creator::combine(List list) const
{
return new Item_cond_and(list);
}
Item_bool_func *Equal_creator::create_scalar_predicate(Item *a, Item *b)
const
{
DBUG_ASSERT(a->type() != Item::ROW_ITEM || b->type() != Item::ROW_ITEM);
return new Item_func_equal(a, b);
}
Item_bool_func *Equal_creator::combine(List list) const
{
return new Item_cond_and(list);
}
Item_bool_func* Ne_creator::create_scalar_predicate(Item *a, Item *b) const
{
DBUG_ASSERT(a->type() != Item::ROW_ITEM || b->type() != Item::ROW_ITEM);
return new Item_func_ne(a, b);
}
Item_bool_func *Ne_creator::combine(List list) const
{
return new Item_cond_or(list);
}
Item_bool_func* Gt_creator::create(Item *a, Item *b) const
{
return new Item_func_gt(a, b);
}
Item_bool_func* Lt_creator::create(Item *a, Item *b) const
{
return new Item_func_lt(a, b);
}
Item_bool_func* Ge_creator::create(Item *a, Item *b) const
{
return new Item_func_ge(a, b);
}
Item_bool_func* Le_creator::create(Item *a, Item *b) const
{
return new Item_func_le(a, b);
}
float Item_func_not::get_filtering_effect(table_map filter_for_table,
table_map read_tables,
const MY_BITMAP *fields_to_ignore,
double rows_in_table)
{
const float filter= args[0]->get_filtering_effect(filter_for_table,
read_tables,
fields_to_ignore,
rows_in_table);
/*
If the predicate that will be negated has COND_FILTER_ALLPASS
filtering it means that some dependent tables have not been
read, that the predicate is of a type that filtering effect is
not calculated for or something similar. In any case, the
filtering effect of the inverted predicate should also be
COND_FILTER_ALLPASS.
*/
if (filter == COND_FILTER_ALLPASS)
return COND_FILTER_ALLPASS;
return 1.0f - filter;
}
/*
Test functions
Most of these returns 0LL if false and 1LL if true and
NULL if some arg is NULL.
*/
longlong Item_func_not::val_int()
{
DBUG_ASSERT(fixed == 1);
bool value= args[0]->val_bool();
null_value=args[0]->null_value;
return ((!null_value && value == 0) ? 1 : 0);
}
/*
We put any NOT expression into parenthesis to avoid
possible problems with internal view representations where
any '!' is converted to NOT. It may cause a problem if
'!' is used in an expression together with other operators
whose precedence is lower than the precedence of '!' yet
higher than the precedence of NOT.
*/
void Item_func_not::print(String *str, enum_query_type query_type)
{
str->append('(');
Item_func::print(str, query_type);
str->append(')');
}
/**
special NOT for ALL subquery.
*/
longlong Item_func_not_all::val_int()
{
DBUG_ASSERT(fixed == 1);
bool value= args[0]->val_bool();
/*
return TRUE if there was records in underlying select in max/min
optimization (ALL subquery)
*/
if (empty_underlying_subquery())
return 1;
null_value= args[0]->null_value;
return ((!null_value && value == 0) ? 1 : 0);
}
bool Item_func_not_all::empty_underlying_subquery()
{
DBUG_ASSERT(subselect || !(test_sum_item || test_sub_item));
/*
When outer argument is NULL the subquery has not yet been evaluated, we
need to evaluate it to get to know whether it returns any rows to return
the correct result. 'ANY' subqueries are an exception because the
result would be false or null which for a top level item always mean false.
The subselect->unit->item->... chain should be used instead of
subselect->... to workaround subquery transformation which could make
subselect->engine unusable.
*/
if (subselect &&
subselect->substype() != Item_subselect::ANY_SUBS &&
!subselect->unit->item->is_evaluated())
subselect->unit->item->exec();
return ((test_sum_item && !test_sum_item->any_value()) ||
(test_sub_item && !test_sub_item->any_value()));
}
void Item_func_not_all::print(String *str, enum_query_type query_type)
{
if (show)
Item_func::print(str, query_type);
else
args[0]->print(str, query_type);
}
/**
Special NOP (No OPeration) for ALL subquery. It is like
Item_func_not_all.
@return
(return TRUE if underlying subquery do not return rows) but if subquery
returns some rows it return same value as argument (TRUE/FALSE).
*/
longlong Item_func_nop_all::val_int()
{
DBUG_ASSERT(fixed == 1);
longlong value= args[0]->val_int();
/*
return FALSE if there was records in underlying select in max/min
optimization (SAME/ANY subquery)
*/
if (empty_underlying_subquery())
return 0;
null_value= args[0]->null_value;
return (null_value || value == 0) ? 0 : 1;
}
/**
Convert a constant item to an int and replace the original item.
The function converts a constant expression or string to an integer.
On successful conversion the original item is substituted for the
result of the item evaluation.
This is done when comparing DATE/TIME of different formats and
also when comparing bigint to strings (in which case strings
are converted to bigints).
@param thd thread handle
@param field item will be converted using the type of this field
@param[in,out] item reference to the item to convert
@note
This function is called only at prepare stage.
As all derived tables are filled only after all derived tables
are prepared we do not evaluate items with subselects here because
they can contain derived tables and thus we may attempt to use a
table that has not been populated yet.
@retval
0 Can't convert item
@retval
1 Item was replaced with an integer version of the item
*/
static bool convert_constant_item(THD *thd, Item_field *field_item,
Item **item)
{
Field *field= field_item->field;
int result= 0;
if ((*item)->const_item() &&
/*
In case of GC it's possible that this func will be called on an
already converted constant. Don't convert it again.
*/
!((*item)->field_type() == field_item->field_type() &&
(*item)->basic_const_item()))
{
TABLE *table= field->table;
sql_mode_t orig_sql_mode= thd->variables.sql_mode;
enum_check_fields orig_count_cuted_fields= thd->count_cuted_fields;
my_bitmap_map *old_maps[2];
ulonglong orig_field_val= 0; /* original field value if valid */
old_maps[0]= NULL;
old_maps[1]= NULL;
if (table)
dbug_tmp_use_all_columns(table, old_maps,
table->read_set, table->write_set);
/* For comparison purposes allow invalid dates like 2000-01-32 */
thd->variables.sql_mode= (orig_sql_mode & ~MODE_NO_ZERO_DATE) |
MODE_INVALID_DATES;
thd->count_cuted_fields= CHECK_FIELD_IGNORE;
/*
Store the value of the field/constant if it references an outer field
because the call to save_in_field below overrides that value.
Don't save field value if no data has been read yet.
Outer constant values are always saved.
*/
bool save_field_value= (field_item->depended_from &&
(field_item->const_item() ||
!(field->table->status &
(STATUS_GARBAGE | STATUS_NOT_FOUND))));
if (save_field_value)
orig_field_val= field->val_int();
int rc;
if (!(*item)->is_null() &&
(((rc= (*item)->save_in_field(field, true)) == TYPE_OK) ||
rc == TYPE_NOTE_TIME_TRUNCATED)) // TS-TODO
{
int field_cmp= 0;
/*
If item is a decimal value, we must reject it if it was truncated.
TODO: consider doing the same for MYSQL_TYPE_YEAR,.
However: we have tests which assume that things '1999' and
'1991-01-01 01:01:01' can be converted to year.
Testing for MYSQL_TYPE_YEAR here, would treat such literals
as 'incorrect DOUBLE value'.
See Bug#13580652 YEAR COLUMN CAN BE EQUAL TO 1999.1
*/
if (field->type() == MYSQL_TYPE_LONGLONG)
{
field_cmp= stored_field_cmp_to_item(thd, field, *item);
DBUG_PRINT("info", ("convert_constant_item %d", field_cmp));
}
if (0 == field_cmp)
{
Item *tmp= field->type() == MYSQL_TYPE_TIME ?
#define OLD_CMP
#ifdef OLD_CMP
new Item_time_with_ref(field->decimals(),
field->val_time_temporal(), *item) :
#else
new Item_time_with_ref(max((*item)->time_precision(),
field->decimals()),
(*item)->val_time_temporal(),
*item) :
#endif
field->is_temporal_with_date() ?
#ifdef OLD_CMP
new Item_datetime_with_ref(field->type(),
field->decimals(),
field->val_date_temporal(),
*item) :
#else
new Item_datetime_with_ref(field->type(),
max((*item)->datetime_precision(),
field->decimals()),
(*item)->val_date_temporal(),
*item) :
#endif
new Item_int_with_ref(field->type(), field->val_int(), *item,
MY_TEST(field->flags & UNSIGNED_FLAG));
if (tmp)
thd->change_item_tree(item, tmp);
result= 1; // Item was replaced
}
}
/* Restore the original field value. */
if (save_field_value)
{
result= field->store(orig_field_val, TRUE);
/* orig_field_val must be a valid value that can be restored back. */
DBUG_ASSERT(!result);
}
thd->variables.sql_mode= orig_sql_mode;
thd->count_cuted_fields= orig_count_cuted_fields;
if (table)
dbug_tmp_restore_column_maps(table->read_set, table->write_set, old_maps);
}
return result;
}
bool Item_bool_func2::convert_constant_arg(THD *thd, Item *field, Item **item)
{
if (field->real_item()->type() != FIELD_ITEM)
return false;
Item_field *field_item= (Item_field*) (field->real_item());
if (field_item->field->can_be_compared_as_longlong() &&
!(field_item->is_temporal_with_date() &&
(*item)->result_type() == STRING_RESULT))
{
if (convert_constant_item(thd, field_item, item))
{
cmp.set_cmp_func(this, tmp_arg, tmp_arg + 1, INT_RESULT);
field->cmp_context= (*item)->cmp_context= INT_RESULT;
return true;
}
}
return false;
}
void Item_bool_func2::fix_length_and_dec()
{
max_length= 1; // Function returns 0 or 1
THD *thd;
/*
As some compare functions are generated after sql_yacc,
we have to check for out of memory conditions here
*/
if (!args[0] || !args[1])
return;
DBUG_ENTER("Item_bool_func2::fix_length_and_dec");
/*
See agg_item_charsets() in item.cc for comments
on character set and collation aggregation.
*/
if (args[0]->result_type() == STRING_RESULT &&
args[1]->result_type() == STRING_RESULT &&
agg_arg_charsets_for_comparison(cmp.cmp_collation, args, 2))
DBUG_VOID_RETURN;
args[0]->cmp_context= args[1]->cmp_context=
item_cmp_type(args[0]->result_type(), args[1]->result_type());
// Make a special case of compare with fields to get nicer DATE comparisons
if (functype() == LIKE_FUNC) // Disable conversion in case of LIKE function.
{
set_cmp_func();
DBUG_VOID_RETURN;
}
/*
Geometry item cannot participate in an arithmetic or string comparison or
a full text search, except in equal/not equal comparison.
We allow geometry arguments in equal/not equal, since such
comparisons are used now and are meaningful, although it simply compares
the GEOMETRY byte string rather than doing a geometric equality comparison.
*/
const Functype func_type= functype();
if (func_type == LT_FUNC || func_type == LE_FUNC || func_type == GE_FUNC ||
func_type == GT_FUNC || func_type == FT_FUNC)
reject_geometry_args(arg_count, args, this);
thd= current_thd;
if (!thd->lex->is_ps_or_view_context_analysis())
{
if (convert_constant_arg(thd, args[0], &args[1]) ||
convert_constant_arg(thd, args[1], &args[0]))
DBUG_VOID_RETURN;
}
set_cmp_func();
DBUG_VOID_RETURN;
}
void Arg_comparator::cleanup()
{
if (comparators != NULL)
{
/*
We cannot rely on (*a)->cols(), since *a may be deallocated
at this point, so use comparator_count to loop.
*/
for (size_t i= 0; i < comparator_count; i++)
{
comparators[i].cleanup();
}
}
delete[] comparators;
comparators= 0;
delete_json_scalar_holder(json_scalar);
json_scalar= 0;
}
int Arg_comparator::set_compare_func(Item_result_field *item, Item_result type)
{
owner= item;
func= comparator_matrix[type]
[is_owner_equal_func()];
switch (type) {
case ROW_RESULT:
{
uint n= (*a)->cols();
if (n != (*b)->cols())
{
my_error(ER_OPERAND_COLUMNS, MYF(0), n);
comparators= 0;
return 1;
}
if (!(comparators= new Arg_comparator[n]))
return 1;
comparator_count= n;
for (uint i=0; i < n; i++)
{
if ((*a)->element_index(i)->cols() != (*b)->element_index(i)->cols())
{
my_error(ER_OPERAND_COLUMNS, MYF(0), (*a)->element_index(i)->cols());
return 1;
}
if (comparators[i].set_cmp_func(owner, (*a)->addr(i), (*b)->addr(i),
set_null))
return 1;
}
break;
}
case STRING_RESULT:
{
/*
We must set cmp_charset here as we may be called from for an automatic
generated item, like in natural join
*/
if (cmp_collation.set((*a)->collation, (*b)->collation) ||
cmp_collation.derivation == DERIVATION_NONE)
{
my_coll_agg_error((*a)->collation, (*b)->collation,
owner->func_name());
return 1;
}
if (cmp_collation.collation == &my_charset_bin)
{
/*
We are using BLOB/BINARY/VARBINARY, change to compare byte by byte,
without removing end space
*/
if (func == &Arg_comparator::compare_string)
func= &Arg_comparator::compare_binary_string;
else if (func == &Arg_comparator::compare_e_string)
func= &Arg_comparator::compare_e_binary_string;
/*
As this is binary compassion, mark all fields that they can't be
transformed. Otherwise we would get into trouble with comparisons
like:
WHERE col= 'j' AND col LIKE BINARY 'j'
which would be transformed to:
WHERE col= 'j'
*/
(*a)->walk(&Item::set_no_const_sub, Item::WALK_POSTFIX, NULL);
(*b)->walk(&Item::set_no_const_sub, Item::WALK_POSTFIX, NULL);
}
break;
}
case INT_RESULT:
{
if ((*a)->is_temporal() && (*b)->is_temporal())
{
func= is_owner_equal_func() ?
&Arg_comparator::compare_e_time_packed :
&Arg_comparator::compare_time_packed;
}
else if (func == &Arg_comparator::compare_int_signed)
{
if ((*a)->unsigned_flag)
func= (((*b)->unsigned_flag)?
&Arg_comparator::compare_int_unsigned :
&Arg_comparator::compare_int_unsigned_signed);
else if ((*b)->unsigned_flag)
func= &Arg_comparator::compare_int_signed_unsigned;
}
else if (func== &Arg_comparator::compare_e_int)
{
if ((*a)->unsigned_flag ^ (*b)->unsigned_flag)
func= &Arg_comparator::compare_e_int_diff_signedness;
}
break;
}
case DECIMAL_RESULT:
break;
case REAL_RESULT:
{
if ((*a)->decimals < NOT_FIXED_DEC && (*b)->decimals < NOT_FIXED_DEC)
{
precision= 5 / log_10[max((*a)->decimals, (*b)->decimals) + 1];
if (func == &Arg_comparator::compare_real)
func= &Arg_comparator::compare_real_fixed;
else if (func == &Arg_comparator::compare_e_real)
func= &Arg_comparator::compare_e_real_fixed;
}
break;
}
default:
DBUG_ASSERT(0);
}
return 0;
}
/**
Parse date provided in a string to a MYSQL_TIME.
@param[in] thd Thread handle
@param[in] str A string to convert
@param[in] warn_type Type of the timestamp for issuing the warning
@param[in] warn_name Field name for issuing the warning
@param[out] l_time The MYSQL_TIME objects is initialized.
Parses a date provided in the string str into a MYSQL_TIME object. If the
string contains an incorrect date or doesn't correspond to a date at all
then a warning is issued. The warn_type and the warn_name arguments are used
as the name and the type of the field when issuing the warning. If any input
was discarded (trailing or non-timestamp-y characters), return value will be
TRUE.
@return Status flag
@retval FALSE Success.
@retval True Indicates failure.
*/
bool get_mysql_time_from_str(THD *thd, String *str, timestamp_type warn_type,
const char *warn_name, MYSQL_TIME *l_time)
{
bool value;
MYSQL_TIME_STATUS status;
my_time_flags_t flags= TIME_FUZZY_DATE | TIME_INVALID_DATES;
if (thd->variables.sql_mode & MODE_NO_ZERO_IN_DATE)
flags|= TIME_NO_ZERO_IN_DATE;
if (thd->variables.sql_mode & MODE_NO_ZERO_DATE)
flags|= TIME_NO_ZERO_DATE;
if (!str_to_datetime(str, l_time, flags, &status) &&
(l_time->time_type == MYSQL_TIMESTAMP_DATETIME ||
l_time->time_type == MYSQL_TIMESTAMP_DATE))
/*
Do not return yet, we may still want to throw a "trailing garbage"
warning.
*/
value= FALSE;
else
{
value= TRUE;
status.warnings= MYSQL_TIME_WARN_TRUNCATED; /* force warning */
}
if (status.warnings > 0)
make_truncated_value_warning(thd, Sql_condition::SL_WARNING,
ErrConvString(str), warn_type, warn_name);
return value;
}
/**
@brief Convert date provided in a string
to its packed temporal int representation.
@param[in] thd thread handle
@param[in] str a string to convert
@param[in] warn_type type of the timestamp for issuing the warning
@param[in] warn_name field name for issuing the warning
@param[out] error_arg could not extract a DATE or DATETIME
@details Convert date provided in the string str to the int
representation. If the string contains wrong date or doesn't
contain it at all then a warning is issued. The warn_type and
the warn_name arguments are used as the name and the type of the
field when issuing the warning.
@return
converted value. 0 on error and on zero-dates -- check 'failure'
*/
static ulonglong get_date_from_str(THD *thd, String *str,
timestamp_type warn_type,
const char *warn_name, bool *error_arg)
{
MYSQL_TIME l_time;
*error_arg= get_mysql_time_from_str(thd, str, warn_type, warn_name, &l_time);
if (*error_arg)
return 0;
return TIME_to_longlong_datetime_packed(&l_time);
}
/**
Check if str_arg is a constant and convert it to datetime packed value.
Note, const_value may stay untouched, so the caller is responsible to
initialize it.
@param date_arg date argument, it's name is used for error reporting.
@param str_arg string argument to get datetime value from.
@param[out] const_value the converted value is stored here, if not NULL.
@return true on error, false on success, false if str_arg is not a const.
*/
bool Arg_comparator::get_date_from_const(Item *date_arg,
Item *str_arg,
ulonglong *const_value)
{
THD *thd= current_thd;
/*
Do not cache GET_USER_VAR() function as its const_item() may return TRUE
for the current thread but it still may change during the execution.
Don't use cache while in the context analysis mode only (i.e. for
EXPLAIN/CREATE VIEW and similar queries). Cache is useless in such
cases and can cause problems. For example evaluating subqueries can
confuse storage engines since in context analysis mode tables
aren't locked.
*/
if (!thd->lex->is_ps_or_view_context_analysis() &&
str_arg->const_item() &&
(str_arg->type() != Item::FUNC_ITEM ||
((Item_func*) str_arg)->functype() != Item_func::GUSERVAR_FUNC))
{
ulonglong value;
if (str_arg->field_type() == MYSQL_TYPE_TIME)
{
// Convert from TIME to DATETIME
value= str_arg->val_date_temporal();
if (str_arg->null_value)
return true;
}
else
{
// Convert from string to DATETIME
DBUG_ASSERT(str_arg->result_type() == STRING_RESULT);
bool error;
String tmp, *str_val= 0;
timestamp_type t_type= (date_arg->field_type() == MYSQL_TYPE_DATE ?
MYSQL_TIMESTAMP_DATE : MYSQL_TIMESTAMP_DATETIME);
str_val= str_arg->val_str(&tmp);
if (str_arg->null_value)
return true;
value= get_date_from_str(thd, str_val, t_type,
date_arg->item_name.ptr(), &error);
if (error)
return true;
}
if (const_value)
*const_value= value;
}
return false;
}
/*
Check whether compare_datetime() can be used to compare items.
SYNOPSIS
Arg_comparator::can_compare_as_dates()
a, b [in] items to be compared
const_value [out] converted value of the string constant, if any
DESCRIPTION
Check several cases when the DATE/DATETIME comparator should be used.
The following cases are checked:
1. Both a and b is a DATE/DATETIME field/function returning string or
int result.
2. Only a or b is a DATE/DATETIME field/function returning string or
int result and the other item (b or a) is an item with string result.
If the second item is a constant one then it's checked to be
convertible to the DATE/DATETIME type. If the constant can't be
converted to a DATE/DATETIME then the compare_datetime() comparator
isn't used and the warning about wrong DATE/DATETIME value is issued.
In all other cases (date-[int|real|decimal]/[int|real|decimal]-date)
the comparison is handled by other comparators.
If the datetime comparator can be used and one the operands of the
comparison is a string constant that was successfully converted to a
DATE/DATETIME type then the result of the conversion is returned in the
const_value if it is provided. If there is no constant or
compare_datetime() isn't applicable then the *const_value remains
unchanged.
@return true if can compare as dates, false otherwise.
*/
bool
Arg_comparator::can_compare_as_dates(Item *a, Item *b, ulonglong *const_value)
{
if (a->type() == Item::ROW_ITEM || b->type() == Item::ROW_ITEM)
return false;
if (a->is_temporal_with_date())
{
if (b->is_temporal_with_date()) // date[time] + date
{
return true;
}
else if (b->result_type() == STRING_RESULT) // date[time] + string
{
return !get_date_from_const(a, b, const_value);
}
else
return false; // date[time] + number
}
else if (b->is_temporal_with_date() &&
a->result_type() == STRING_RESULT) // string + date[time]
{
return !get_date_from_const(b, a, const_value);
}
else
return false; // No date[time] items found
}
/*
Retrieves correct TIME value from the given item.
SYNOPSIS
get_time_value()
thd thread handle
item_arg [in/out] item to retrieve TIME value from
cache_arg [in/out] pointer to place to store the cache item to
warn_item [in] unused
is_null [out] TRUE the item_arg is null
DESCRIPTION
Retrieves the correct TIME value from given item for comparison by the
compare_datetime() function.
If item's result can be compared as longlong then its int value is used
and a value returned by get_time function is used otherwise.
If an item is a constant one then its value is cached and it isn't
get parsed again. An Item_cache_int object is used for for cached values.
It seamlessly substitutes the original item. The cache item is marked as
non-constant to prevent re-caching it again.
RETURN
obtained value
*/
longlong
get_time_value(THD *thd, Item ***item_arg, Item **cache_arg,
const Item *warn_item, bool *is_null)
{
longlong value;
Item *item= **item_arg;
/*
Note, it's wrong to assume that we always get
a TIME expression or NULL here:
DBUG_ASSERT(item->field_type() == MYSQL_TYPE_TIME ||
item->field_type() == MYSQL_TYPE_NULL);
because when this condition is optimized:
WHERE time_column=DATE(NULL) AND time_column=TIME(NULL);
rhe first AND part is eliminated and DATE(NULL) is substituted
to the second AND part like this:
WHERE DATE(NULL) = TIME(NULL) // as TIME
whose Arg_comparator has already get_time_value set for both arguments.
Therefore, get_time_value is executed for DATE(NULL).
This condition is further evaluated as impossible condition.
TS-TODO: perhaps such cases should be evaluated without
calling get_time_value at all.
See a similar comment in Arg_comparator::compare_time_packed.
*/
value= item->val_time_temporal();
*is_null= item->null_value;
/*
Do not cache GET_USER_VAR() function as its const_item() may return TRUE
for the current thread but it still may change during the execution.
*/
if (item->const_item() && cache_arg &&
item->type() != Item::CACHE_ITEM &&
(item->type() != Item::FUNC_ITEM ||
((Item_func*)item)->functype() != Item_func::GUSERVAR_FUNC))
{
Item_cache_datetime *cache= new Item_cache_datetime(item->field_type());
/* Mark the cache as non-const to prevent re-caching. */
cache->set_used_tables(1);
cache->store(item, value);
*cache_arg= cache;
*item_arg= cache_arg;
}
return value;
}
/**
Sets compare functions for various datatypes.
NOTE
The result type of a comparison is chosen by item_cmp_type().
Here we override the chosen result type for certain expression
containing date or time or decimal expressions.
*/
int Arg_comparator::set_cmp_func(Item_result_field *owner_arg,
Item **a1, Item **a2,
Item_result type)
{
ulonglong const_value= (ulonglong)-1;
owner= owner_arg;
set_null= set_null && owner_arg;
a= a1;
b= a2;
if (type != ROW_RESULT &&
(((*a)->result_type() == STRING_RESULT &&
(*a)->field_type() == MYSQL_TYPE_JSON) ||
((*b)->result_type() == STRING_RESULT &&
(*b)->field_type() == MYSQL_TYPE_JSON)))
{
// Use the JSON comparator if at least one of the arguments is JSON.
is_nulls_eq= is_owner_equal_func();
func= &Arg_comparator::compare_json;
return 0;
}
if (can_compare_as_dates(*a, *b, &const_value))
{
a_cache= 0;
b_cache= 0;
if (const_value != (ulonglong)-1)
{
/*
cache_converted_constant can't be used here because it can't
correctly convert a DATETIME value from string to int representation.
*/
Item_cache_datetime *cache= new Item_cache_datetime(MYSQL_TYPE_DATETIME);
/* Mark the cache as non-const to prevent re-caching. */
cache->set_used_tables(1);
if (!(*a)->is_temporal_with_date())
{
cache->store((*a), const_value);
a_cache= cache;
a= &a_cache;
}
else
{
cache->store((*b), const_value);
b_cache= cache;
b= &b_cache;
}
}
is_nulls_eq= is_owner_equal_func();
func= &Arg_comparator::compare_datetime;
get_value_a_func= &get_datetime_value;
get_value_b_func= &get_datetime_value;
cmp_collation.set(&my_charset_numeric);
set_cmp_context_for_datetime();
return 0;
}
else if ((type == STRING_RESULT ||
// When comparing time field and cached/converted time constant
type == REAL_RESULT) &&
(*a)->field_type() == MYSQL_TYPE_TIME &&
(*b)->field_type() == MYSQL_TYPE_TIME)
{
/* Compare TIME values as integers. */
a_cache= 0;
b_cache= 0;
is_nulls_eq= is_owner_equal_func();
func= &Arg_comparator::compare_datetime;
get_value_a_func= &get_time_value;
get_value_b_func= &get_time_value;
set_cmp_context_for_datetime();
return 0;
}
else if (type == STRING_RESULT &&
(*a)->result_type() == STRING_RESULT &&
(*b)->result_type() == STRING_RESULT)
{
DTCollation coll;
coll.set((*a)->collation.collation);
if (agg_item_set_converter(coll, owner->func_name(),
b, 1, MY_COLL_CMP_CONV, 1))
return 1;
}
else if (try_year_cmp_func(type))
{
return 0;
}
else if (type == REAL_RESULT &&
(((*a)->result_type() == DECIMAL_RESULT && !(*a)->const_item() &&
(*b)->result_type() == STRING_RESULT && (*b)->const_item()) ||
((*b)->result_type() == DECIMAL_RESULT && !(*b)->const_item() &&
(*a)->result_type() == STRING_RESULT && (*a)->const_item())))
{
/*
or
Do comparison as decimal rather than float, in order not to lose precision.
*/
type= DECIMAL_RESULT;
}
THD *thd= current_thd;
a= cache_converted_constant(thd, a, &a_cache, type);
b= cache_converted_constant(thd, b, &b_cache, type);
return set_compare_func(owner_arg, type);
}
int Arg_comparator::set_cmp_func(Item_result_field *owner_arg,
Item **a1, Item **a2, bool set_null_arg)
{
set_null= set_null_arg;
const Item_result item_result=
item_cmp_type((*a1)->result_type(),
(*a2)->result_type());
return set_cmp_func(owner_arg, a1, a2, item_result);
}
/*
Helper function to call from Arg_comparator::set_cmp_func()
*/
bool Arg_comparator::try_year_cmp_func(Item_result type)
{
if (type == ROW_RESULT)
return FALSE;
bool a_is_year= (*a)->field_type() == MYSQL_TYPE_YEAR;
bool b_is_year= (*b)->field_type() == MYSQL_TYPE_YEAR;
if (!a_is_year && !b_is_year)
return FALSE;
if (a_is_year && b_is_year)
{
get_value_a_func= &get_year_value;
get_value_b_func= &get_year_value;
}
else if (a_is_year && (*b)->is_temporal_with_date())
{
get_value_a_func= &get_year_value;
get_value_b_func= &get_datetime_value;
}
else if (b_is_year && (*a)->is_temporal_with_date())
{
get_value_b_func= &get_year_value;
get_value_a_func= &get_datetime_value;
}
else
return FALSE;
is_nulls_eq= is_owner_equal_func();
func= &Arg_comparator::compare_datetime;
set_cmp_context_for_datetime();
return TRUE;
}
/**
Convert and cache a constant.
@param value [in] An item to cache
@param cache_item [out] Placeholder for the cache item
@param type [in] Comparison type
@details
When given item is a constant and its type differs from comparison type
then cache its value to avoid type conversion of this constant on each
evaluation. In this case the value is cached and the reference to the cache
is returned.
Original value is returned otherwise.
@return cache item or original value.
*/
Item** Arg_comparator::cache_converted_constant(THD *thd_arg, Item **value,
Item **cache_item,
Item_result type)
{
/* Don't need cache if doing context analysis only. */
if (!thd_arg->lex->is_ps_or_view_context_analysis() &&
(*value)->const_item() && type != (*value)->result_type())
{
Item_cache *cache= Item_cache::get_cache(*value, type);
cache->setup(*value);
*cache_item= cache;
return cache_item;
}
return value;
}
void Arg_comparator::set_datetime_cmp_func(Item_result_field *owner_arg,
Item **a1, Item **b1)
{
owner= owner_arg;
a= a1;
b= b1;
a_cache= 0;
b_cache= 0;
is_nulls_eq= FALSE;
func= &Arg_comparator::compare_datetime;
get_value_a_func= &get_datetime_value;
get_value_b_func= &get_datetime_value;
set_cmp_context_for_datetime();
}
/*
Retrieves correct DATETIME value from given item.
SYNOPSIS
get_datetime_value()
thd thread handle
item_arg [in/out] item to retrieve DATETIME value from
cache_arg [in/out] pointer to place to store the caching item to
warn_item [in] item for issuing the conversion warning
is_null [out] TRUE the item_arg is null
DESCRIPTION
Retrieves the correct DATETIME value from given item for comparison by the
compare_datetime() function.
If item's result can be compared as longlong then its int value is used
and its string value is used otherwise. Strings are always parsed and
converted to int values by the get_date_from_str() function.
This allows us to compare correctly string dates with missed insignificant
zeros. If an item is a constant one then its value is cached and it isn't
get parsed again. An Item_cache_int object is used for caching values. It
seamlessly substitutes the original item. The cache item is marked as
non-constant to prevent re-caching it again. In order to compare
correctly DATE and DATETIME items the result of the former are treated as
a DATETIME with zero time (00:00:00).
RETURN
obtained value
*/
longlong
get_datetime_value(THD *thd, Item ***item_arg, Item **cache_arg,
const Item *warn_item, bool *is_null)
{
longlong value= 0;
String buf, *str= 0;
Item *item= **item_arg;
if (item->is_temporal())
{
value= item->val_date_temporal();
*is_null= item->null_value;
}
else
{
str= item->val_str(&buf);
*is_null= item->null_value;
}
if (*is_null)
return ~(ulonglong) 0;
/*
Convert strings to the integer DATE/DATETIME representation.
Even if both dates provided in strings we can't compare them directly as
strings as there is no warranty that they are correct and do not miss
some insignificant zeros.
*/
if (str)
{
bool error;
enum_field_types f_type= warn_item->field_type();
timestamp_type t_type= f_type ==
MYSQL_TYPE_DATE ? MYSQL_TIMESTAMP_DATE : MYSQL_TIMESTAMP_DATETIME;
value= (longlong) get_date_from_str(thd, str, t_type, warn_item->item_name.ptr(), &error);
/*
If str did not contain a valid date according to the current
SQL_MODE, get_date_from_str() has already thrown a warning,
and we don't want to throw NULL on invalid date (see 5.2.6
"SQL modes" in the manual), so we're done here.
*/
}
/*
Do not cache GET_USER_VAR() function as its const_item() may return TRUE
for the current thread but it still may change during the execution.
*/
if (item->const_item() && cache_arg &&
item->type() != Item::CACHE_ITEM &&
(item->type() != Item::FUNC_ITEM ||
((Item_func*)item)->functype() != Item_func::GUSERVAR_FUNC))
{
Item_cache_datetime *cache= new Item_cache_datetime(MYSQL_TYPE_DATETIME);
/* Mark the cache as non-const to prevent re-caching. */
cache->set_used_tables(1);
cache->store(item, value);
*cache_arg= cache;
*item_arg= cache_arg;
}
return value;
}
/*
Retrieves YEAR value of 19XX-00-00 00:00:00 form from given item.
SYNOPSIS
get_year_value()
thd thread handle
item_arg [in/out] item to retrieve YEAR value from
cache_arg [in/out] pointer to place to store the caching item to
warn_item [in] item for issuing the conversion warning
is_null [out] TRUE the item_arg is null
DESCRIPTION
Retrieves the YEAR value of 19XX form from given item for comparison by the
compare_datetime() function.
Converts year to DATETIME of form YYYY-00-00 00:00:00 for the compatibility
with the get_datetime_value function result.
RETURN
obtained value
*/
static longlong
get_year_value(THD *thd, Item ***item_arg, Item **cache_arg,
const Item *warn_item, bool *is_null)
{
longlong value= 0;
Item *item= **item_arg;
value= item->val_int();
*is_null= item->null_value;
if (*is_null)
return ~(ulonglong) 0;
/* Convert year to DATETIME packed format */
return year_to_longlong_datetime_packed(static_cast(value));
}
/*
Compare items values as dates.
SYNOPSIS
Arg_comparator::compare_datetime()
DESCRIPTION
Compare items values as DATE/DATETIME for both EQUAL_FUNC and from other
comparison functions. The correct DATETIME values are obtained
with help of the get_datetime_value() function.
RETURN
If is_nulls_eq is TRUE:
1 if items are equal or both are null
0 otherwise
If is_nulls_eq is FALSE:
-1 a < b or at least one item is null
0 a == b
1 a > b
See the table:
is_nulls_eq | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 0 |
a_is_null | 1 | 0 | 1 | 0 | 1 | 0 | 1 | 0 |
b_is_null | 1 | 1 | 0 | 0 | 1 | 1 | 0 | 0 |
result | 1 | 0 | 0 |0/1|-1 |-1 |-1 |-1/0/1|
*/
int Arg_comparator::compare_datetime()
{
bool a_is_null, b_is_null;
longlong a_value, b_value;
THD *thd= current_thd;
/* Get DATE/DATETIME/TIME value of the 'a' item. */
a_value= (*get_value_a_func)(thd, &a, &a_cache, *b, &a_is_null);
if (!is_nulls_eq && a_is_null)
{
if (set_null)
owner->null_value= 1;
return -1;
}
/* Get DATE/DATETIME/TIME value of the 'b' item. */
b_value= (*get_value_b_func)(thd, &b, &b_cache, *a, &b_is_null);
if (a_is_null || b_is_null)
{
if (set_null)
owner->null_value= is_nulls_eq ? 0 : 1;
return is_nulls_eq ? (a_is_null == b_is_null) : -1;
}
/* Here we have two not-NULL values. */
if (set_null)
owner->null_value= 0;
/* Compare values. */
if (is_nulls_eq)
return (a_value == b_value);
return a_value < b_value ? -1 : (a_value > b_value ? 1 : 0);
}
/**
Get one of the arguments to the comparator as a JSON value.
@param[in] arg pointer to the argument
@param[in,out] value buffer used for reading the JSON value
@param[in,out] tmp buffer used for converting string values to the
correct charset, if necessary
@param[out] result where to store the result
@param[in,out] scalar pointer to a location with pre-allocated memory
used for JSON scalars that are converted from
SQL scalars
@retval false on success
@retval true on failure
*/
static bool get_json_arg(Item* arg, String *value, String *tmp,
Json_wrapper *result, Json_scalar_holder **scalar)
{
Json_scalar_holder *holder= NULL;
/*
If the argument is a non-JSON type, it gets converted to a JSON
scalar. Use the pre-allocated memory passed in via the "scalar"
argument. Note, however, that geometry types are not converted
to scalars. They are converted to JSON objects by get_json_atom_wrapper().
*/
if ((arg->field_type() != MYSQL_TYPE_JSON) &&
(arg->field_type() != MYSQL_TYPE_GEOMETRY))
{
/*
If it's a constant item, and we've already read it, just return
the value that's cached in the pre-allocated memory.
*/
if (*scalar && arg->const_item())
{
Json_wrapper tmp(get_json_scalar_from_holder(*scalar));
tmp.set_alias();
result->steal(&tmp);
return false;
}
/*
Allocate memory to hold the scalar, if we haven't already done
so. Otherwise, we reuse the previously allocated memory.
*/
if (!*scalar)
*scalar= create_json_scalar_holder();
holder= *scalar;
}
return get_json_atom_wrapper(&arg, 0, "is_temporal())
return new cmp_item_datetime(item);
else
return new cmp_item_string(cs);
case INT_RESULT:
return new cmp_item_int;
case REAL_RESULT:
return new cmp_item_real;
case ROW_RESULT:
return new cmp_item_row;
case DECIMAL_RESULT:
return new cmp_item_decimal;
default:
DBUG_ASSERT(0);
break;
}
return 0; // to satisfy compiler :)
}
cmp_item* cmp_item_string::make_same()
{
return new cmp_item_string(cmp_charset);
}
cmp_item* cmp_item_int::make_same()
{
return new cmp_item_int();
}
cmp_item* cmp_item_real::make_same()
{
return new cmp_item_real();
}
cmp_item* cmp_item_row::make_same()
{
return new cmp_item_row();
}
cmp_item_row::~cmp_item_row()
{
DBUG_ENTER("~cmp_item_row");
DBUG_PRINT("enter",("this: 0x%lx", (long) this));
if (comparators)
{
for (uint i= 0; i < n; i++)
{
if (comparators[i])
delete comparators[i];
}
}
DBUG_VOID_RETURN;
}
void cmp_item_row::alloc_comparators(Item *item)
{
n= item->cols();
DBUG_ASSERT(comparators == NULL);
if (!comparators)
comparators= (cmp_item **) current_thd->mem_calloc(sizeof(cmp_item *)*n);
if (comparators)
{
for (uint i= 0; i < n; i++)
{
DBUG_ASSERT(comparators[i] == NULL);
Item *item_i= item->element_index(i);
if (!(comparators[i]=
cmp_item::get_comparator(item_i->result_type(), item_i,
item_i->collation.collation)))
break; // new failed
if (item_i->result_type() == ROW_RESULT)
static_cast(comparators[i])->alloc_comparators(item_i);
}
}
}
void cmp_item_row::store_value(Item *item)
{
DBUG_ENTER("cmp_item_row::store_value");
DBUG_ASSERT(comparators);
if (comparators)
{
item->bring_value();
item->null_value= 0;
for (uint i= 0; i < n; i++)
{
comparators[i]->store_value(item->element_index(i));
item->null_value|= item->element_index(i)->null_value;
}
}
DBUG_VOID_RETURN;
}
void cmp_item_row::store_value_by_template(cmp_item *t, Item *item)
{
cmp_item_row *tmpl= (cmp_item_row*) t;
if (tmpl->n != item->cols())
{
my_error(ER_OPERAND_COLUMNS, MYF(0), tmpl->n);
return;
}
n= tmpl->n;
if ((comparators= (cmp_item **) sql_alloc(sizeof(cmp_item *)*n)))
{
item->bring_value();
item->null_value= 0;
for (uint i=0; i < n; i++)
{
if (!(comparators[i]= tmpl->comparators[i]->make_same()))
break; // new failed
comparators[i]->store_value_by_template(tmpl->comparators[i],
item->element_index(i));
item->null_value|= item->element_index(i)->null_value;
}
}
}
int cmp_item_row::cmp(Item *arg)
{
arg->null_value= 0;
if (arg->cols() != n)
{
my_error(ER_OPERAND_COLUMNS, MYF(0), n);
return 1;
}
bool was_null= 0;
arg->bring_value();
for (uint i=0; i < n; i++)
{
const int rc= comparators[i]->cmp(arg->element_index(i));
switch (rc)
{
case UNKNOWN:
was_null= true;
break;
case TRUE:
return TRUE;
case FALSE:
break; // elements #i are equal
}
arg->null_value|= arg->element_index(i)->null_value;
}
return was_null ? UNKNOWN : FALSE;
}
int cmp_item_row::compare(const cmp_item *c) const
{
const cmp_item_row *l_cmp= down_cast(c);
for (uint i=0; i < n; i++)
{
int res;
if ((res= comparators[i]->compare(l_cmp->comparators[i])))
return res;
}
return 0;
}
void cmp_item_decimal::store_value(Item *item)
{
my_decimal *val= item->val_decimal(&value);
/* val may be zero if item is nnull */
if (val && val != &value)
my_decimal2decimal(val, &value);
set_null_value(item->null_value);
}
int cmp_item_decimal::cmp(Item *arg)
{
my_decimal tmp_buf, *tmp= arg->val_decimal(&tmp_buf);
return (m_null_value || arg->null_value) ?
UNKNOWN : (my_decimal_cmp(&value, tmp) != 0);
}
int cmp_item_decimal::compare(const cmp_item *arg) const
{
const cmp_item_decimal *l_cmp= down_cast(arg);
return my_decimal_cmp(&value, &l_cmp->value);
}
cmp_item* cmp_item_decimal::make_same()
{
return new cmp_item_decimal();
}
cmp_item_datetime::cmp_item_datetime(const Item *warn_item_arg)
:warn_item(warn_item_arg), lval_cache(0),
has_date(warn_item_arg->is_temporal_with_date())
{}
void cmp_item_datetime::store_value(Item *item)
{
bool is_null;
Item **tmp_item= lval_cache ? &lval_cache : &item;
if (has_date)
value= get_datetime_value(current_thd, &tmp_item, &lval_cache, warn_item,
&is_null);
else
value= get_time_value(current_thd, &tmp_item, &lval_cache, warn_item,
&is_null);
set_null_value(item->null_value);
}
int cmp_item_datetime::cmp(Item *arg)
{
bool is_null;
Item **tmp_item= &arg;
longlong value2= 0;
if (has_date)
value2= get_datetime_value(current_thd, &tmp_item, 0, warn_item, &is_null);
else
value2= get_time_value(current_thd, &tmp_item, 0, warn_item, &is_null);
const bool rc= (value != value2);
return (m_null_value || arg->null_value) ? UNKNOWN : rc;
}
int cmp_item_datetime::compare(const cmp_item *ci) const
{
const cmp_item_datetime *l_cmp= down_cast(ci);
return (value < l_cmp->value) ? -1 : ((value == l_cmp->value) ? 0 : 1);
}
cmp_item *cmp_item_datetime::make_same()
{
return new cmp_item_datetime(warn_item);
}
float
Item_func_in::get_single_col_filtering_effect(Item_ident *fieldref,
table_map filter_for_table,
const MY_BITMAP *fields_to_ignore,
double rows_in_table)
{
/*
Does not contribute to filtering effect if
1) This field belongs to another table.
2) Filter effect for this field has already been taken into
account. 'fieldref' may be a field or a reference to a field
(through a view, to an outer table etc)
*/
if ((fieldref->used_tables() != filter_for_table) || // 1)
bitmap_is_set(fields_to_ignore,
static_cast
(fieldref->real_item())->field->field_index)) // 2)
return COND_FILTER_ALLPASS;
const Item_field *fld= (Item_field*)fieldref->real_item();
return fld->get_cond_filter_default_probability(rows_in_table,
COND_FILTER_EQUALITY);
}
float Item_func_in::get_filtering_effect(table_map filter_for_table,
table_map read_tables,
const MY_BITMAP *fields_to_ignore,
double rows_in_table)
{
DBUG_ASSERT((read_tables & filter_for_table) == 0);
/*
To contribute to filtering effect, the condition must refer to
exactly one unread table: the table filtering is currently
calculated for.
Dependent subqueries are not considered available values and no
filtering should be calculated for this item if the IN list
contains one. dep_subq_in_list is 'true' if the IN list contains a
dependent subquery.
*/
if ((used_tables() & ~read_tables) != filter_for_table ||
dep_subq_in_list)
return COND_FILTER_ALLPASS;
/*
No matter how many row values are input the filtering effect
shall not be higher than in_max_filter (currently 0.5).
*/
const float in_max_filter= 0.5f;
float filter= COND_FILTER_ALLPASS;
if (args[0]->type() == Item::ROW_ITEM)
{
/*
This is a row value IN predicate:
"WHERE (col1, col2, ...) IN ((1,2,..), ...)"
which can be rewritten to:
"WHERE (col1=1 AND col2=2...) OR (col1=.. AND col2=...) OR ..."
The filtering effect is:
filter= #row_values * filter()
where filter() = filter(col1) * filter(col2) * ...
In other words, we ignore the fact that there could be identical
row values since writing "WHERE (a,b) IN ((1,1), (1,1), ...)" is
not expected input from a user.
*/
Item_row* lhs_row= static_cast(args[0]);
// For all items in the left row
float single_rowval_filter= COND_FILTER_ALLPASS;
for (uint i= 0; i < lhs_row->cols(); i++)
{
/*
May contribute to condition filtering only if
lhs_row->element_index(i) is a field or a reference to a field
(through a view, to an outer table etc)
*/
if (lhs_row->element_index(i)->real_item()->type() == Item::FIELD_ITEM)
{
Item_ident *fieldref=
static_cast(lhs_row->element_index(i));
const float tmp_filt= get_single_col_filtering_effect(fieldref,
filter_for_table,
fields_to_ignore,
rows_in_table);
single_rowval_filter*= tmp_filt;
}
}
/*
If single_rowval_filter == COND_FILTER_ALLPASS, the filtering
effect of this field should be ignored. If not, selectivity
should not be higher than 'in_max_filter' even if there are a
lot of values on the right hand side
arg_count includes the left hand side item
*/
if (single_rowval_filter != COND_FILTER_ALLPASS)
filter= min((arg_count - 1) * single_rowval_filter, in_max_filter);
}
else if (args[0]->real_item()->type() == Item::FIELD_ITEM)
{
/*
This is a single-column IN predicate:
"WHERE col IN (1, 2, ...)"
which can be rewritten to:
"WHERE col=1 OR col1=2 OR ..."
The filtering effect is: #values_right_hand_side * selectivity(=)
As for row values, it is assumed that no values on the right
hand side are identical.
*/
DBUG_ASSERT(args[0]->type() == FIELD_ITEM || args[0]->type() == REF_ITEM);
Item_ident *fieldref= static_cast(args[0]);
const float tmp_filt= get_single_col_filtering_effect(fieldref,
filter_for_table,
fields_to_ignore,
rows_in_table);
/*
If tmp_filt == COND_FILTER_ALLPASS, the filtering effect of this
field should be ignored. If not, selectivity should not be
higher than 'in_max_filter' even if there are a lot of values on
the right hand side
arg_count includes the left hand side item
*/
if (tmp_filt != COND_FILTER_ALLPASS)
filter= min((arg_count - 1) * tmp_filt, in_max_filter);
}
if (negated && filter != COND_FILTER_ALLPASS)
filter= 1.0f - filter;
DBUG_ASSERT(filter >= 0.0f && filter null_inside())
return 1;
}
return 0;
}
/**
Perform context analysis of an IN item tree.
This function performs context analysis (name resolution) and calculates
various attributes of the item tree with Item_func_in as its root.
The function saves in ref the pointer to the item or to a newly created
item that is considered as a replacement for the original one.
@param thd reference to the global context of the query thread
@param ref pointer to Item* variable where pointer to resulting "fixed"
item is to be assigned
@note
Let T0(e)/T1(e) be the value of not_null_tables(e) when e is used on
a predicate/function level. Then it's easy to show that:
@verbatim
T0(e IN(e1,...,en)) = union(T1(e),intersection(T1(ei)))
T1(e IN(e1,...,en)) = union(T1(e),intersection(T1(ei)))
T0(e NOT IN(e1,...,en)) = union(T1(e),union(T1(ei)))
T1(e NOT IN(e1,...,en)) = union(T1(e),intersection(T1(ei)))
@endverbatim
@retval
0 ok
@retval
1 got error
*/
bool Item_func_in::fix_fields(THD *thd, Item **ref)
{
if (Item_func_opt_neg::fix_fields(thd, ref))
return true;
// not_null_tables_cache == union(T1(e),union(T1(ei)))
if (pred_level && negated)
return false;
// not_null_tables_cache = union(T1(e),intersection(T1(ei)))
not_null_tables_cache= ~(table_map) 0;
Item **arg_end= args + arg_count;
for (Item **arg= args + 1; arg != arg_end; arg++)
not_null_tables_cache&= (*arg)->not_null_tables();
not_null_tables_cache|= (*args)->not_null_tables();
return false;
}
void Item_func_in::fix_after_pullout(st_select_lex *parent_select,
st_select_lex *removed_select)
{
Item_func_opt_neg::fix_after_pullout(parent_select, removed_select);
// not_null_tables_cache == union(T1(e),union(T1(ei)))
if (pred_level && negated)
return;
// not_null_tables_cache = union(T1(e),intersection(T1(ei)))
not_null_tables_cache= ~(table_map) 0;
Item **arg_end= args + arg_count;
for (Item **arg= args + 1; arg != arg_end; arg++)
not_null_tables_cache&= (*arg)->not_null_tables();
not_null_tables_cache|= (*args)->not_null_tables();
}
static int srtcmp_in(CHARSET_INFO *cs, const String *x,const String *y)
{
return cs->coll->strnncollsp(cs,
(uchar *) x->ptr(),x->length(),
(uchar *) y->ptr(),y->length(), 0);
}
void Item_func_in::fix_length_and_dec()
{
Item **arg, **arg_end;
bool const_itm= 1;
THD *thd= current_thd;
bool datetime_found= FALSE;
/* TRUE arguments values will be compared as DATETIMEs. */
bool compare_as_datetime= FALSE;
Item *date_arg= 0;
uint found_types= 0;
uint type_cnt= 0, i;
Item_result cmp_type= STRING_RESULT;
left_result_type= args[0]->result_type();
if (!(found_types= collect_cmp_types(args, arg_count, true)))
return;
for (arg= args + 1, arg_end= args + arg_count; arg != arg_end ; arg++)
{
if (!arg[0]->const_item())
{
const_itm= 0;
if (arg[0]->real_item()->type() == Item::SUBSELECT_ITEM)
dep_subq_in_list= true;
break;
}
}
for (i= 0; i tmp;
}
else
{
if (!(cmp= new cmp_item_row))
return;
cmp_items[ROW_RESULT]= cmp;
}
cmp->n= args[0]->cols();
cmp->alloc_comparators(args[0]);
}
/* All DATE/DATETIME fields/functions has the STRING result type. */
if (cmp_type == STRING_RESULT || cmp_type == ROW_RESULT)
{
uint col, cols= args[0]->cols();
for (col= 0; col < cols; col++)
{
bool skip_column= FALSE;
/*
Check that all items to be compared has the STRING result type and at
least one of them is a DATE/DATETIME item.
*/
for (arg= args, arg_end= args + arg_count; arg != arg_end ; arg++)
{
Item *itm= ((cmp_type == STRING_RESULT) ? arg[0] :
arg[0]->element_index(col));
if (itm->result_type() != STRING_RESULT)
{
skip_column= TRUE;
break;
}
else if (itm->is_temporal_with_date())
{
datetime_found= TRUE;
/*
Internally all DATE/DATETIME values are converted to the DATETIME
type. So try to find a DATETIME item to issue correct warnings.
*/
if (!date_arg)
date_arg= itm;
else if (itm->field_type() == MYSQL_TYPE_DATETIME)
{
date_arg= itm;
/* All arguments are already checked to have the STRING result. */
if (cmp_type == STRING_RESULT)
break;
}
}
}
if (skip_column)
continue;
if (datetime_found)
{
if (cmp_type == ROW_RESULT)
{
cmp_item **cmp= 0;
if (array)
cmp= ((in_row*)array)->tmp.comparators + col;
else
cmp= ((cmp_item_row*)cmp_items[ROW_RESULT])->comparators + col;
*cmp= new cmp_item_datetime(date_arg);
/* Reset variables for the next column. */
date_arg= 0;
datetime_found= FALSE;
}
else
compare_as_datetime= TRUE;
}
}
}
}
if (bisection_possible)
{
if (compare_as_datetime)
array= new in_datetime(thd, date_arg, arg_count - 1);
else
{
/*
IN must compare INT columns and constants as int values (the same
way as equality does).
So we must check here if the column on the left and all the constant
values on the right can be compared as integers and adjust the
comparison type accordingly.
*/
bool datetime_as_longlong= false;
if (args[0]->real_item()->type() == FIELD_ITEM &&
thd->lex->sql_command != SQLCOM_CREATE_VIEW &&
thd->lex->sql_command != SQLCOM_SHOW_CREATE &&
cmp_type != INT_RESULT)
{
Item_field *field_item= (Item_field*) (args[0]->real_item());
if (field_item->field->can_be_compared_as_longlong())
{
bool all_converted= true;
for (arg=args + 1, arg_end= args + arg_count; arg != arg_end ; arg++)
{
if (!convert_constant_item (thd, field_item, &arg[0]))
all_converted= false;
}
if (all_converted)
{
cmp_type= INT_RESULT;
datetime_as_longlong= field_item->is_temporal();
}
}
}
switch (cmp_type) {
case STRING_RESULT:
array=new in_string(thd, arg_count-1, (qsort2_cmp) srtcmp_in,
cmp_collation.collation);
break;
case INT_RESULT:
array= datetime_as_longlong ?
args[0]->field_type() == MYSQL_TYPE_TIME ?
(in_vector*) new in_time_as_longlong(thd, arg_count - 1) :
(in_vector*) new in_datetime_as_longlong(thd, arg_count - 1) :
(in_vector*) new in_longlong(thd, arg_count - 1);
break;
case REAL_RESULT:
array= new in_double(thd, arg_count-1);
break;
case ROW_RESULT:
/*
The row comparator was created at the beginning.
*/
break;
case DECIMAL_RESULT:
array= new in_decimal(thd, arg_count - 1);
break;
default:
DBUG_ASSERT(0);
return;
}
}
if (!array || thd->is_fatal_error) // OOM
return;
uint j=0;
for (uint i=1 ; i < arg_count ; i++)
{
array->set(j,args[i]);
if (!args[i]->null_value)
j++; // include this cell in the array.
else
{
/*
We don't put NULL values in array, to avoid erronous matches in
bisection.
*/
have_null= 1;
}
}
array->used_count= j;
DBUG_ASSERT(array->used_count count);
if (array->used_count < array->count)
array->shrink_array(j);
if (array->used_count)
array->sort();
}
else
{
if (compare_as_datetime)
cmp_items[STRING_RESULT]= new cmp_item_datetime(date_arg);
else
{
for (i= 0; i cmp_context= item_cmp_type(left_result_type, arg[0]->result_type());
}
max_length= 1;
}
void Item_func_in::print(String *str, enum_query_type query_type)
{
str->append('(');
args[0]->print(str, query_type);
if (negated)
str->append(STRING_WITH_LEN(" not"));
str->append(STRING_WITH_LEN(" in ("));
print_args(str, 1, query_type);
str->append(STRING_WITH_LEN("))"));
}
/*
Evaluate the function and return its value.
SYNOPSIS
val_int()
DESCRIPTION
Evaluate the function and return its value.
IMPLEMENTATION
If the array object is defined then the value of the function is
calculated by means of this array.
Otherwise several cmp_item objects are used in order to do correct
comparison of left expression and an expression from the values list.
One cmp_item object correspond to one used comparison type. Left
expression can be evaluated up to number of different used comparison
types. A bit mapped variable value_added_map is used to check whether
the left expression already was evaluated for a particular result type.
Result types are mapped to it according to their integer values i.e.
STRING_RESULT is mapped to bit 0, REAL_RESULT to bit 1, so on.
RETURN
Value of the function
*/
longlong Item_func_in::val_int()
{
cmp_item *in_item;
DBUG_ASSERT(fixed == 1);
uint value_added_map= 0;
if (array)
{
bool tmp=array->find_item(args[0]);
/*
NULL on left -> UNKNOWN.
Found no match, and NULL on right -> UNKNOWN.
NULL on right can never give a match, as it is not stored in
array.
See also the 'bisection_possible' variable in fix_length_and_dec().
*/
null_value=args[0]->null_value || (!tmp && have_null);
return (longlong) (!null_value && tmp != negated);
}
if ((null_value= args[0]->real_item()->type() == NULL_ITEM))
return 0;
have_null= 0;
for (uint i= 1 ; i < arg_count ; i++)
{
if (args[i]->real_item()->type() == NULL_ITEM)
{
have_null= TRUE;
continue;
}
Item_result cmp_type= item_cmp_type(left_result_type, args[i]->result_type());
in_item= cmp_items[(uint)cmp_type];
DBUG_ASSERT(in_item);
if (!(value_added_map & (1U store_value(args[0]);
value_added_map|= 1U cmp(args[i]);
if (rc == FALSE)
return (longlong) (!negated);
have_null|= (rc == UNKNOWN);
}
null_value= have_null;
return (longlong) (!null_value && negated);
}
void Item::check_deprecated_bin_op(const Item *a, const Item *b)
{
/*
We want to warn about cases which will likely change behaviour in future
versions. The conditions to emit a warning are:
1. If there's only one argument, the item should be a [VAR]BINARY
argument (1) and it should be different from the hex/bit/NULL literal (2).
2. If there are two arguments, both should be [VAR]BINARY (3) and at least
one of them should be different from the hex/bit/NULL literal (4)
*/
if (a->result_type() == STRING_RESULT &&
a->collation.collation == &my_charset_bin && // (1), (3)
(!b ||
(b->result_type() == STRING_RESULT &&
b->collation.collation == &my_charset_bin)) && // (3)
((a->type() != Item::VARBIN_ITEM &&
a->type() != Item::NULL_ITEM) || // (2), (4)
(b && b->type() != Item::VARBIN_ITEM &&
b->type() != Item::NULL_ITEM))) // (4)
{
push_warning_printf(current_thd, Sql_condition::SL_WARNING,
ER_WARN_DEPRECATED_SYNTAX,
"Bitwise operations on BINARY will change behavior"
" in a future version, check the 'Bit functions'"
" section in the manual.");
}
}
longlong Item_func_bit_or::val_int()
{
DBUG_ASSERT(fixed == 1);
ulonglong arg1= (ulonglong) args[0]->val_int();
if (args[0]->null_value)
{
null_value=1; /* purecov: inspected */
return 0; /* purecov: inspected */
}
ulonglong arg2= (ulonglong) args[1]->val_int();
if (args[1]->null_value)
{
null_value=1;
return 0;
}
null_value=0;
return (longlong) (arg1 | arg2);
}
longlong Item_func_bit_and::val_int()
{
DBUG_ASSERT(fixed == 1);
ulonglong arg1= (ulonglong) args[0]->val_int();
if (args[0]->null_value)
{
null_value=1; /* purecov: inspected */
return 0; /* purecov: inspected */
}
ulonglong arg2= (ulonglong) args[1]->val_int();
if (args[1]->null_value)
{
null_value=1; /* purecov: inspected */
return 0; /* purecov: inspected */
}
null_value=0;
return (longlong) (arg1 & arg2);
}
Item_cond::Item_cond(THD *thd, Item_cond *item)
:Item_bool_func(thd, item),
abort_on_null(item->abort_on_null)
{
/*
item->list will be copied by copy_andor_arguments() call
*/
}
/**
Contextualization for Item_cond functional items
Item_cond successors use Item_cond::list instead of Item_func::args
and Item_func::arg_count, so we can't itemize parse-time Item_cond
objects by forwarding a contextualization process to the parent Item_func
class: we need to overload this function to run a contextualization
the Item_cond::list items.
*/
bool Item_cond::itemize(Parse_context *pc, Item **res)
{
if (skip_itemize(res))
return false;
if (super::itemize(pc, res))
return true;
List_iterator li(list);
Item *item;
while ((item= li++))
{
if (item->itemize(pc, &item))
return true;
li.replace(item);
}
return false;
}
void Item_cond::copy_andor_arguments(THD *thd, Item_cond *item)
{
List_iterator_fast li(item->list);
while (Item *it= li++)
{
DBUG_ASSERT(it->real_item()); // Sanity check (no dangling 'ref')
list.push_back(it->copy_andor_structure(thd));
}
}
bool
Item_cond::fix_fields(THD *thd, Item **ref)
{
DBUG_ASSERT(fixed == 0);
List_iterator li(list);
Item *item;
/*
Semi-join flattening should only be performed for predicates on
the AND-top-level. Disable it if this condition is not an AND.
*/
Disable_semijoin_flattening DSF(thd->lex->current_select(),
functype() != COND_AND_FUNC);
uchar buff[sizeof(char*)]; // Max local vars in function
used_tables_cache= 0;
const_item_cache= true;
if (functype() == COND_AND_FUNC && abort_on_null)
not_null_tables_cache= 0;
else
not_null_tables_cache= ~(table_map) 0;
if (check_stack_overrun(thd, STACK_MIN_SIZE, buff))
return TRUE; // Fatal error flag is set!
/*
The following optimization reduces the depth of an AND-OR tree.
E.g. a WHERE clause like
F1 AND (F2 AND (F2 AND F4))
is parsed into a tree with the same nested structure as defined
by braces. This optimization will transform such tree into
AND (F1, F2, F3, F4).
Trees of OR items are flattened as well:
((F1 OR F2) OR (F3 OR F4)) => OR (F1, F2, F3, F4)
Items for removed AND/OR levels will dangle until the death of the
entire statement.
The optimization is currently prepared statements and stored procedures
friendly as it doesn't allocate any memory and its effects are durable
(i.e. do not depend on PS/SP arguments).
*/
while ((item=li++))
{
while (item->type() == Item::COND_ITEM &&
((Item_cond*) item)->functype() == functype() &&
!((Item_cond*) item)->list.is_empty())
{ // Identical function
li.replace(((Item_cond*) item)->list);
((Item_cond*) item)->list.empty();
item= *li.ref(); // new current item
}
if (abort_on_null)
item->top_level_item();
// item can be substituted in fix_fields
if ((!item->fixed &&
item->fix_fields(thd, li.ref())) ||
(item= *li.ref())->check_cols(1))
return TRUE; /* purecov: inspected */
used_tables_cache|= item->used_tables();
const_item_cache&= item->const_item();
if (functype() == COND_AND_FUNC && abort_on_null)
not_null_tables_cache|= item->not_null_tables();
else
not_null_tables_cache&= item->not_null_tables();
with_sum_func|= item->with_sum_func;
with_subselect|= item->has_subquery();
with_stored_program|= item->has_stored_program();
maybe_null|= item->maybe_null;
}
thd->lex->current_select()->cond_count+= list.elements;
fix_length_and_dec();
fixed= true;
return false;
}
void Item_cond::fix_after_pullout(st_select_lex *parent_select,
st_select_lex *removed_select)
{
List_iterator li(list);
Item *item;
used_tables_cache= get_initial_pseudo_tables();
const_item_cache= true;
if (functype() == COND_AND_FUNC && abort_on_null)
not_null_tables_cache= 0;
else
not_null_tables_cache= ~(table_map) 0;
while ((item=li++))
{
item->fix_after_pullout(parent_select, removed_select);
used_tables_cache|= item->used_tables();
const_item_cache&= item->const_item();
if (functype() == COND_AND_FUNC && abort_on_null)
not_null_tables_cache|= item->not_null_tables();
else
not_null_tables_cache&= item->not_null_tables();
}
}
bool Item_cond::eq(const Item *item, bool binary_cmp) const
{
if (this == item)
return true;
if (item->type() != COND_ITEM)
return false;
const Item_cond *item_cond= down_cast(item);
if (functype() != item_cond->functype() ||
list.elements != item_cond->list.elements ||
func_name() != item_cond->func_name())
return false;
// Item_cond never uses "args". Inspect "list" instead.
DBUG_ASSERT(arg_count == 0 && item_cond->arg_count == 0);
List_iterator_fast it1(const_cast(this)->list);
List_iterator_fast it2(const_cast(item_cond)->list);
Item *i;
while ((i= it1++))
if (!i->eq(it2++, binary_cmp))
return false;
return true;
}
bool Item_cond::walk(Item_processor processor, enum_walk walk, uchar *arg)
{
if ((walk & WALK_PREFIX) && (this->*processor)(arg))
return true;
List_iterator_fast li(list);
Item *item;
while ((item= li++))
{
if (item->walk(processor, walk, arg))
return true;
}
return (walk & WALK_POSTFIX) && (this->*processor)(arg);
}
/**
Transform an Item_cond object with a transformer callback function.
The function recursively applies the transform method to each
member item of the condition list.
If the call of the method for a member item returns a new item
the old item is substituted for a new one.
After this the transformer is applied to the root node
of the Item_cond object.
@param transformer the transformer callback function to be applied to
the nodes of the tree of the object
@param arg parameter to be passed to the transformer
@return
Item returned as the result of transformation of the root node
*/
Item *Item_cond::transform(Item_transformer transformer, uchar *arg)
{
DBUG_ASSERT(!current_thd->stmt_arena->is_stmt_prepare());
List_iterator li(list);
Item *item;
while ((item= li++))
{
Item *new_item= item->transform(transformer, arg);
if (!new_item)
return 0;
/*
THD::change_item_tree() should be called only if the tree was
really transformed, i.e. when a new item has been created.
Otherwise we'll be allocating a lot of unnecessary memory for
change records at each execution.
*/
if (new_item != item)
current_thd->change_item_tree(li.ref(), new_item);
}
return Item_func::transform(transformer, arg);
}
/**
Compile Item_cond object with a processor and a transformer
callback functions.
First the function applies the analyzer to the root node of
the Item_func object. Then if the analyzer succeeeds (returns TRUE)
the function recursively applies the compile method to member
item of the condition list.
If the call of the method for a member item returns a new item
the old item is substituted for a new one.
After this the transformer is applied to the root node
of the Item_cond object.
@param analyzer the analyzer callback function to be applied to the
nodes of the tree of the object
@param[in,out] arg_p parameter to be passed to the analyzer
@param transformer the transformer callback function to be applied to the
nodes of the tree of the object
@param arg_t parameter to be passed to the transformer
@return Item returned as result of transformation of the node,
the same item if no transformation applied, or NULL if
transformation caused an error.
*/
Item *Item_cond::compile(Item_analyzer analyzer, uchar **arg_p,
Item_transformer transformer, uchar *arg_t)
{
if (!(this->*analyzer)(arg_p))
return this;
List_iterator li(list);
Item *item;
while ((item= li++))
{
/*
The same parameter value of arg_p must be passed
to analyze any argument of the condition formula.
*/
uchar *arg_v= *arg_p;
Item *new_item= item->compile(analyzer, &arg_v, transformer, arg_t);
if (new_item == NULL)
return NULL;
if (new_item != item)
current_thd->change_item_tree(li.ref(), new_item);
}
// strange to call transform(): each argument will thus have the transformer
// called twice on it (in compile() above and Item_func::transform below)??
return Item_func::transform(transformer, arg_t);
}
void Item_cond::traverse_cond(Cond_traverser traverser,
void *arg, traverse_order order)
{
List_iterator li(list);
Item *item;
switch(order) {
case(PREFIX):
(*traverser)(this, arg);
while ((item= li++))
{
item->traverse_cond(traverser, arg, order);
}
(*traverser)(NULL, arg);
break;
case(POSTFIX):
while ((item= li++))
{
item->traverse_cond(traverser, arg, order);
}
(*traverser)(this, arg);
}
}
/**
Move SUM items out from item tree and replace with reference.
The split is done to get an unique item for each SUM function
so that we can easily find and calculate them.
(Calculation done by update_sum_func() and copy_sum_funcs() in
sql_select.cc)
@param thd Thread handler
@param ref_pointer_array Pointer to array of reference fields
@param fields All fields in select
@note
This function is run on all expression (SELECT list, WHERE, HAVING etc)
that have or refer (HAVING) to a SUM expression.
*/
void Item_cond::split_sum_func(THD *thd, Ref_ptr_array ref_pointer_array,
List &fields)
{
List_iterator li(list);
Item *item;
while ((item= li++))
item->split_sum_func2(thd, ref_pointer_array, fields, li.ref(), TRUE);
}
void Item_cond::update_used_tables()
{
List_iterator_fast li(list);
Item *item;
used_tables_cache=0;
const_item_cache=1;
with_subselect= false;
with_stored_program= false;
while ((item=li++))
{
item->update_used_tables();
used_tables_cache|= item->used_tables();
const_item_cache&= item->const_item();
with_subselect|= item->has_subquery();
with_stored_program|= item->has_stored_program();
}
}
void Item_cond::print(String *str, enum_query_type query_type)
{
str->append('(');
List_iterator_fast li(list);
Item *item;
if ((item=li++))
item->print(str, query_type);
while ((item=li++))
{
str->append(' ');
str->append(func_name());
str->append(' ');
item->print(str, query_type);
}
str->append(')');
}
void Item_cond::neg_arguments(THD *thd)
{
List_iterator li(list);
Item *item;
while ((item= li++)) /* Apply not transformation to the arguments */
{
Item *new_item= item->neg_transformer(thd);
if (!new_item)
{
if (!(new_item= new Item_func_not(item)))
return; // Fatal OEM error
}
(void) li.replace(new_item);
}
}
float Item_cond_and::get_filtering_effect(table_map filter_for_table,
table_map read_tables,
const MY_BITMAP *fields_to_ignore,
double rows_in_table)
{
if (!(used_tables() & filter_for_table))
return COND_FILTER_ALLPASS; // No conditions below this apply to the table
float filter= COND_FILTER_ALLPASS;
List_iterator it(list);
Item *item;
/*
Calculated as "Conjunction of independent events":
P(A and B ...) = P(A) * P(B) * ...
*/
while ((item= it++))
filter*= item->get_filtering_effect(filter_for_table,
read_tables,
fields_to_ignore,
rows_in_table);
return filter;
}
/**
Evaluation of AND(expr, expr, expr ...).
@note
abort_if_null is set for AND expressions for which we don't care if the
result is NULL or 0. This is set for:
- WHERE clause
- HAVING clause
- IF(expression)
@retval
1 If all expressions are true
@retval
0 If all expressions are false or if we find a NULL expression and
'abort_on_null' is set.
@retval
NULL if all expression are either 1 or NULL
*/
longlong Item_cond_and::val_int()
{
DBUG_ASSERT(fixed == 1);
List_iterator_fast li(list);
Item *item;
null_value= 0;
while ((item=li++))
{
if (!item->val_bool())
{
if (abort_on_null || !(null_value= item->null_value))
return 0; // return FALSE
}
}
return null_value ? 0 : 1;
}
float Item_cond_or::get_filtering_effect(table_map filter_for_table,
table_map read_tables,
const MY_BITMAP *fields_to_ignore,
double rows_in_table)
{
if (!(used_tables() & filter_for_table))
return COND_FILTER_ALLPASS; // No conditions below this apply to the table
float filter= 0.0f;
List_iterator it(list);
Item *item;
while ((item= it++))
{
const float cur_filter=
item->get_filtering_effect(filter_for_table, read_tables,
fields_to_ignore, rows_in_table);
/*
Calculated as "Disjunction of independent events":
P(A or B) = P(A) + P(B) - P(A) * P(B)
If any of the ORed predicates has a filtering effect of
COND_FILTER_ALLPASS, the end result is COND_FILTER_ALLPASS. This is as
expected since COND_FILTER_ALLPASS means that a) the predicate has
no filtering effect at all, or b) the predicate's filtering
effect is unknown. In both cases, the only meaningful result is
for OR to produce COND_FILTER_ALLPASS.
*/
filter= filter + cur_filter - (filter * cur_filter);
}
return filter;
}
longlong Item_cond_or::val_int()
{
DBUG_ASSERT(fixed == 1);
List_iterator_fast li(list);
Item *item;
null_value=0;
while ((item=li++))
{
if (item->val_bool())
{
null_value=0;
return 1;
}
if (item->null_value)
null_value=1;
}
return 0;
}
/**
Create an AND expression from two expressions.
@param a expression or NULL
@param b expression.
@param org_item Don't modify a if a == *org_item.
If a == NULL, org_item is set to point at b,
to ensure that future calls will not modify b.
@note
This will not modify item pointed to by org_item or b
The idea is that one can call this in a loop and create and
'and' over all items without modifying any of the original items.
@retval
NULL Error
@retval
Item
*/
Item *and_expressions(Item *a, Item *b, Item **org_item)
{
if (!a)
return (*org_item= b);
if (a == *org_item)
{
Item_cond *res;
if ((res= new Item_cond_and(a, b)))
{
res->set_used_tables(a->used_tables() | b->used_tables());
res->set_not_null_tables(a->not_null_tables() | b->not_null_tables());
}
return res;
}
if (((Item_cond_and*) a)->add(b))
return 0;
((Item_cond_and*) a)->set_used_tables(a->used_tables() | b->used_tables());
((Item_cond_and*) a)->set_not_null_tables(a->not_null_tables() |
b->not_null_tables());
return a;
}
float Item_func_isnull::get_filtering_effect(table_map filter_for_table,
table_map read_tables,
const MY_BITMAP *fields_to_ignore,
double rows_in_table)
{
const Item_field* fld=
contributes_to_filter(read_tables, filter_for_table, fields_to_ignore);
if (!fld)
return COND_FILTER_ALLPASS;
return fld->get_cond_filter_default_probability(rows_in_table,
COND_FILTER_EQUALITY);
}
longlong Item_func_isnull::val_int()
{
DBUG_ASSERT(fixed == 1);
/*
Handle optimization if the argument can't be null
This has to be here because of the test in update_used_tables().
*/
if (const_item_cache)
return cached_value;
return args[0]->is_null() ? 1: 0;
}
longlong Item_is_not_null_test::val_int()
{
DBUG_ASSERT(fixed == 1);
DBUG_ENTER("Item_is_not_null_test::val_int");
if (!used_tables_cache && !with_subselect && !with_stored_program)
{
/*
TODO: Currently this branch never executes, since used_tables_cache
is never equal to 0 -- it always contains RAND_TABLE_BIT,
see get_initial_pseudo_tables().
*/
owner->was_null|= (!cached_value);
DBUG_PRINT("info", ("cached: %ld", (long) cached_value));
DBUG_RETURN(cached_value);
}
if (args[0]->is_null())
{
DBUG_PRINT("info", ("null"));
owner->was_null|= 1;
DBUG_RETURN(0);
}
else
DBUG_RETURN(1);
}
/**
Optimize case of not_null_column IS NULL.
*/
void Item_is_not_null_test::update_used_tables()
{
const table_map initial_pseudo_tables= get_initial_pseudo_tables();
used_tables_cache= initial_pseudo_tables;
if (!args[0]->maybe_null)
{
cached_value= 1;
return;
}
args[0]->update_used_tables();
with_subselect= args[0]->has_subquery();
with_stored_program= args[0]->has_stored_program();
used_tables_cache|= args[0]->used_tables();
if (used_tables_cache == initial_pseudo_tables && !with_subselect &&
!with_stored_program)
/* Remember if the value is always NULL or never NULL */
cached_value= !args[0]->is_null();
}
float
Item_func_isnotnull::get_filtering_effect(table_map filter_for_table,
table_map read_tables,
const MY_BITMAP *fields_to_ignore,
double rows_in_table)
{
const Item_field* fld=
contributes_to_filter(read_tables, filter_for_table, fields_to_ignore);
if (!fld)
return COND_FILTER_ALLPASS;
return 1.0f - fld->get_cond_filter_default_probability(rows_in_table,
COND_FILTER_EQUALITY);
}
longlong Item_func_isnotnull::val_int()
{
DBUG_ASSERT(fixed == 1);
return args[0]->is_null() ? 0 : 1;
}
void Item_func_isnotnull::print(String *str, enum_query_type query_type)
{
str->append('(');
args[0]->print(str, query_type);
str->append(STRING_WITH_LEN(" is not null)"));
}
float Item_func_like::get_filtering_effect(table_map filter_for_table,
table_map read_tables,
const MY_BITMAP *fields_to_ignore,
double rows_in_table)
{
const Item_field* fld=
contributes_to_filter(read_tables, filter_for_table, fields_to_ignore);
if (!fld)
return COND_FILTER_ALLPASS;
/*
Filtering effect is similar to that of BETWEEN because
* "col like abc%" is similar to
"col between abc and abd"
The same applies for 'abc_'
* "col like %abc" can be seen as
"reverse(col) like cba%"" (see above)
* "col like "abc%def%..." is also similar
Now we're left with "col like " which should
have filtering effect like equality, but it would be costly to
look through the whole string searching for wildcards and since
LIKE is mostly used for wildcards this isn't checked.
*/
return fld->get_cond_filter_default_probability(rows_in_table,
COND_FILTER_BETWEEN);
}
bool Item_func_like::itemize(Parse_context *pc, Item **res)
{
if (skip_itemize(res))
return false;
if (super::itemize(pc, res) ||
(escape_item != NULL && escape_item->itemize(pc, &escape_item)))
return true;
if (escape_item == NULL)
{
THD *thd= pc->thd;
escape_item=
((thd->variables.sql_mode & MODE_NO_BACKSLASH_ESCAPES) ?
new (pc->mem_root) Item_string("", 0, &my_charset_latin1) :
new (pc->mem_root) Item_string("\\", 1, &my_charset_latin1));
}
return escape_item == NULL;
}
longlong Item_func_like::val_int()
{
DBUG_ASSERT(fixed == 1);
if (!escape_evaluated && eval_escape_clause(current_thd))
return error_int();
String* res = args[0]->val_str(&cmp.value1);
if (args[0]->null_value)
{
null_value=1;
return 0;
}
String* res2 = args[1]->val_str(&cmp.value2);
if (args[1]->null_value)
{
null_value=1;
return 0;
}
null_value=0;
if (can_do_bm)
return bm_matches(res->ptr(), res->length()) ? 1 : 0;
return my_wildcmp(cmp.cmp_collation.collation,
res->ptr(),res->ptr()+res->length(),
res2->ptr(),res2->ptr()+res2->length(),
escape,wild_one,wild_many) ? 0 : 1;
}
/**
We can optimize a where if first character isn't a wildcard
*/
Item_func::optimize_type Item_func_like::select_optimize() const
{
if (!args[1]->const_item())
return OPTIMIZE_NONE;
String* res2= args[1]->val_str((String *)&cmp.value2);
if (!res2)
return OPTIMIZE_NONE;
if (!res2->length()) // Can optimize empty wildcard: column LIKE ''
return OPTIMIZE_OP;
DBUG_ASSERT(res2->ptr());
char first= res2->ptr()[0];
return (first == wild_many || first == wild_one) ?
OPTIMIZE_NONE : OPTIMIZE_OP;
}
bool Item_func_like::fix_fields(THD *thd, Item **ref)
{
DBUG_ASSERT(fixed == 0);
Disable_semijoin_flattening DSF(thd->lex->current_select(), true);
if (Item_bool_func2::fix_fields(thd, ref) ||
escape_item->fix_fields(thd, &escape_item) ||
escape_item->check_cols(1))
return true;
if (!escape_item->const_during_execution())
{
my_error(ER_WRONG_ARGUMENTS,MYF(0),"ESCAPE");
return true;
}
if (escape_item->const_item())
{
/*
We need to know the escape character in order to apply Boyer-Moore. Since
it is const, it is safe to evaluate it now at the resolution stage.
*/
if (eval_escape_clause(thd))
return true;
/*
We could also do boyer-more for non-const items, but as we would have to
recompute the tables for each row it's not worth it.
*/
if (args[1]->const_item() && !use_strnxfrm(collation.collation) &&
!(specialflag & SPECIAL_NO_NEW_FUNC))
{
String* res2 = args[1]->val_str(&cmp.value2);
if (thd->is_error())
return true;
if (!res2)
return false; // Null argument
const size_t len = res2->length();
const char* first = res2->ptr();
const char* last = first + len - 1;
/*
Minimum length pattern before Boyer-Moore is used
for SELECT "text" LIKE "%pattern%" including the two
wildcards in class Item_func_like.
*/
const size_t min_bm_pattern_len= 5;
if (len > min_bm_pattern_len &&
*first == wild_many &&
*last == wild_many)
{
const char* tmp = first + 1;
for (; *tmp != wild_many && *tmp != wild_one && *tmp != escape; tmp++) ;
can_do_bm= (tmp == last) && !use_mb(args[0]->collation.collation);
}
if (can_do_bm)
{
pattern_len = (int) len - 2;
pattern = thd->strmake(first + 1, pattern_len);
DBUG_PRINT("info", ("Initializing pattern: '%s'", first));
int *suff = (int*) thd->alloc((int) (sizeof(int)*
((pattern_len + 1)*2+
alphabet_size)));
bmGs = suff + pattern_len + 1;
bmBc = bmGs + pattern_len + 1;
bm_compute_good_suffix_shifts(suff);
bm_compute_bad_character_shifts();
DBUG_PRINT("info",("done"));
}
}
}
return false;
}
void Item_func_like::cleanup()
{
can_do_bm= false;
escape_evaluated= false;
Item_bool_func2::cleanup();
}
/**
Evaluate the expression in the escape clause.
@param thd thread handler
@return false on success, true on failure
*/
bool Item_func_like::eval_escape_clause(THD *thd)
{
DBUG_ASSERT(!escape_evaluated);
String buf;
String *escape_str= escape_item->val_str(&buf);
if (escape_str)
{
const char *escape_str_ptr= escape_str->ptr();
if (escape_used_in_parsing && (
(((thd->variables.sql_mode & MODE_NO_BACKSLASH_ESCAPES) &&
escape_str->numchars() != 1) ||
escape_str->numchars() > 1)))
{
my_error(ER_WRONG_ARGUMENTS,MYF(0),"ESCAPE");
return true;
}
if (use_mb(cmp.cmp_collation.collation))
{
const CHARSET_INFO *cs= escape_str->charset();
my_wc_t wc;
int rc= cs->cset->mb_wc(cs, &wc,
(const uchar*) escape_str_ptr,
(const uchar*) escape_str_ptr +
escape_str->length());
escape= (int) (rc > 0 ? wc : '\\');
}
else
{
/*
In the case of 8bit character set, we pass native
code instead of Unicode code as "escape" argument.
Convert to "cs" if charset of escape differs.
*/
const CHARSET_INFO *cs= cmp.cmp_collation.collation;
size_t unused;
if (escape_str->needs_conversion(escape_str->length(),
escape_str->charset(), cs, &unused))
{
char ch;
uint errors;
size_t cnvlen= copy_and_convert(&ch, 1, cs, escape_str_ptr,
escape_str->length(),
escape_str->charset(), &errors);
escape= cnvlen ? ch : '\\';
}
else
escape= escape_str_ptr ? *escape_str_ptr : '\\';
}
}
else
escape= '\\';
escape_evaluated= true;
return false;
}
/**
@brief Compile regular expression.
@param[in] send_error send error message if any.
@details Make necessary character set conversion then
compile regular expression passed in the args[1].
@retval 0 success.
@retval 1 error occurred.
@retval -1 given null regular expression.
*/
int Item_func_regex::regcomp(bool send_error)
{
char buff[MAX_FIELD_WIDTH];
String tmp(buff,sizeof(buff),&my_charset_bin);
String *res= args[1]->val_str(&tmp);
int error;
if (args[1]->null_value)
return -1;
if (regex_compiled)
{
if (!stringcmp(res, &prev_regexp))
return 0;
prev_regexp.copy(*res);
my_regfree(&preg);
regex_compiled= 0;
}
if (cmp_collation.collation != regex_lib_charset)
{
/* Convert UCS2 strings to UTF8 */
uint dummy_errors;
if (conv.copy(res->ptr(), res->length(), res->charset(),
regex_lib_charset, &dummy_errors))
return 1;
res= &conv;
}
if ((error= my_regcomp(&preg, res->c_ptr_safe(),
regex_lib_flags, regex_lib_charset)))
{
if (send_error)
{
(void) my_regerror(error, &preg, buff, sizeof(buff));
my_error(ER_REGEXP_ERROR, MYF(0), buff);
}
return 1;
}
regex_compiled= 1;
return 0;
}
bool
Item_func_regex::fix_fields(THD *thd, Item **ref)
{
DBUG_ASSERT(fixed == 0);
Disable_semijoin_flattening DSF(thd->lex->current_select(), true);
if ((!args[0]->fixed &&
args[0]->fix_fields(thd, args)) || args[0]->check_cols(1) ||
(!args[1]->fixed &&
args[1]->fix_fields(thd, args + 1)) || args[1]->check_cols(1))
return TRUE; /* purecov: inspected */
with_sum_func=args[0]->with_sum_func || args[1]->with_sum_func;
with_subselect= args[0]->has_subquery() || args[1]->has_subquery();
with_stored_program= args[0]->has_stored_program() ||
args[1]->has_stored_program();
max_length= 1;
decimals= 0;
if (agg_arg_charsets_for_comparison(cmp_collation, args, 2))
return TRUE;
regex_lib_flags= (cmp_collation.collation->state &
(MY_CS_BINSORT | MY_CS_CSSORT)) ?
MY_REG_EXTENDED | MY_REG_NOSUB :
MY_REG_EXTENDED | MY_REG_NOSUB | MY_REG_ICASE;
/*
If the case of UCS2 and other non-ASCII character sets,
we will convert patterns and strings to UTF8.
*/
regex_lib_charset= (cmp_collation.collation->mbminlen > 1) ?
&my_charset_utf8_general_ci :
cmp_collation.collation;
used_tables_cache=args[0]->used_tables() | args[1]->used_tables();
not_null_tables_cache= (args[0]->not_null_tables() |
args[1]->not_null_tables());
const_item_cache=args[0]->const_item() && args[1]->const_item();
if (!regex_compiled && args[1]->const_item())
{
int comp_res= regcomp(TRUE);
if (comp_res == -1)
{ // Will always return NULL
maybe_null=1;
fixed= 1;
return FALSE;
}
else if (comp_res)
return TRUE;
regex_is_const= 1;
maybe_null= args[0]->maybe_null;
}
else
maybe_null=1;
fixed= 1;
return FALSE;
}
longlong Item_func_regex::val_int()
{
DBUG_ASSERT(fixed == 1);
char buff[MAX_FIELD_WIDTH];
String tmp(buff,sizeof(buff),&my_charset_bin);
String *res= args[0]->val_str(&tmp);
if ((null_value= (args[0]->null_value ||
(!regex_is_const && regcomp(FALSE)))))
return 0;
if (cmp_collation.collation != regex_lib_charset)
{
/* Convert UCS2 strings to UTF8 */
uint dummy_errors;
if (conv.copy(res->ptr(), res->length(), res->charset(),
regex_lib_charset, &dummy_errors))
{
null_value= 1;
return 0;
}
res= &conv;
}
return my_regexec(&preg,res->c_ptr_safe(),0,(my_regmatch_t*) 0,0) ? 0 : 1;
}
void Item_func_regex::cleanup()
{
DBUG_ENTER("Item_func_regex::cleanup");
Item_bool_func::cleanup();
if (regex_compiled)
{
my_regfree(&preg);
regex_compiled=0;
prev_regexp.length(0);
}
DBUG_VOID_RETURN;
}
#define likeconv(cs,A) (uchar) (cs)->sort_order[(uchar) (A)]
/**
Precomputation dependent only on pattern_len.
*/
void Item_func_like::bm_compute_suffixes(int *suff)
{
const int plm1 = pattern_len - 1;
int f = 0;
int g = plm1;
int *const splm1 = suff + plm1;
const CHARSET_INFO *cs= cmp.cmp_collation.collation;
*splm1 = pattern_len;
if (!cs->sort_order)
{
int i;
for (i = pattern_len - 2; i >= 0; i--)
{
int tmp = *(splm1 + i - f);
if (g < i && tmp < i - g)
suff[i] = tmp;
else
{
if (i < g)
g = i; // g = min(i, g)
f = i;
while (g >= 0 && pattern[g] == pattern[g + plm1 - f])
g--;
suff[i] = f - g;
}
}
}
else
{
int i;
for (i = pattern_len - 2; 0 = 0 &&
likeconv(cs, pattern[g]) == likeconv(cs, pattern[g + plm1 - f]))
g--;
suff[i] = f - g;
}
}
}
}
/**
Precomputation dependent only on pattern_len.
*/
void Item_func_like::bm_compute_good_suffix_shifts(int *suff)
{
bm_compute_suffixes(suff);
int *end = bmGs + pattern_len;
int *k;
for (k = bmGs; k < end; k++)
*k = pattern_len;
int tmp;
int i;
int j = 0;
const int plm1 = pattern_len - 1;
for (i = plm1; i > -1; i--)
{
if (suff[i] == i + 1)
{
for (tmp = plm1 - i; j < tmp; j++)
{
int *tmp2 = bmGs + j;
if (*tmp2 == pattern_len)
*tmp2 = tmp;
}
}
}
int *tmp2;
for (tmp = plm1 - i; j < tmp; j++)
{
tmp2 = bmGs + j;
if (*tmp2 == pattern_len)
*tmp2 = tmp;
}
tmp2 = bmGs + plm1;
for (i = 0; i sort_order)
{
for (j = 0; j < plm1; j++)
bmBc[(uchar) pattern[j]] = plm1 - j;
}
else
{
for (j = 0; j < plm1; j++)
bmBc[likeconv(cs,pattern[j])] = plm1 - j;
}
}
/**
Search for pattern in text.
@return
returns true/false for match/no match
*/
bool Item_func_like::bm_matches(const char* text, size_t text_len) const
{
int bcShift;
int shift = pattern_len;
int j = 0;
const CHARSET_INFO *cs= cmp.cmp_collation.collation;
const int plm1= pattern_len - 1;
const int tlmpl= text_len - pattern_len;
/* Searching */
if (!cs->sort_order)
{
while (j = 0) && (pattern[i] == text[i + j]) ;--i) {}
if (i < 0)
return true;
else
{
bcShift= bmBc[(uchar) text[i + j]] - plm1 + i;
shift= max(bcShift, bmGs[i]);
}
j+= shift;
}
return false;
}
else
{
while (j = 0) && likeconv(cs,pattern[i]) == likeconv(cs,text[i + j]);
--i) {}
if (i < 0)
return true;
else
{
bcShift= bmBc[likeconv(cs, text[i + j])] - plm1 + i;
shift= max(bcShift, bmGs[i]);
}
j+= shift;
}
return false;
}
}
float Item_func_xor::get_filtering_effect(table_map filter_for_table,
table_map read_tables,
const MY_BITMAP *fields_to_ignore,
double rows_in_table)
{
DBUG_ASSERT(arg_count == 2);
const float filter0=
args[0]->get_filtering_effect(filter_for_table, read_tables,
fields_to_ignore, rows_in_table);
if (filter0 == COND_FILTER_ALLPASS)
return COND_FILTER_ALLPASS;
const float filter1=
args[1]->get_filtering_effect(filter_for_table, read_tables,
fields_to_ignore, rows_in_table);
if (filter1 == COND_FILTER_ALLPASS)
return COND_FILTER_ALLPASS;
/*
Calculated as "exactly one of independent events":
P(A and not B) + P(B and not A) = P(A) + P(B) - 2 * P(A) * P(B)
*/
return filter0 + filter1 - (2 * filter0 * filter1);
}
/**
Make a logical XOR of the arguments.
If either operator is NULL, return NULL.
@todo
(low priority) Change this to be optimized as: @n
A XOR B -> (A) == 1 AND (B) 1) OR (A 1 AND (B) == 1) @n
To be able to do this, we would however first have to extend the MySQL
range optimizer to handle OR better.
@note
As we don't do any index optimization on XOR this is not going to be
very fast to use.
*/
longlong Item_func_xor::val_int()
{
DBUG_ASSERT(fixed == 1);
int result= 0;
null_value= false;
for (uint i= 0; i < arg_count; i++)
{
result^= (args[i]->val_int() != 0);
if (args[i]->null_value)
{
null_value= true;
return 0;
}
}
return result;
}
/**
Apply NOT transformation to the item and return a new one.
Transform the item using next rules:
@verbatim
a AND b AND ... -> NOT(a) OR NOT(b) OR ...
a OR b OR ... -> NOT(a) AND NOT(b) AND ...
NOT(a) -> a
a = b -> a != b
a != b -> a = b
a < b -> a >= b
a >= b -> a < b
a > b -> a b
IS NULL(a) -> IS NOT NULL(a)
IS NOT NULL(a) -> IS NULL(a)
@endverbatim
@param thd thread handler
@return
New item or
NULL if we cannot apply NOT transformation (see Item::neg_transformer()).
*/
Item *Item_func_not::neg_transformer(THD *thd) /* NOT(x) -> x */
{
return args[0];
}
Item *Item_bool_rowready_func2::neg_transformer(THD *thd)
{
Item *item= negated_item();
return item;
}
/**
XOR can be negated by negating one of the operands:
NOT (a XOR b) => (NOT a) XOR b
=> a XOR (NOT b)
@param thd Thread handle
@return New negated item
*/
Item *Item_func_xor::neg_transformer(THD *thd)
{
Item *neg_operand;
Item_func_xor *new_item;
if ((neg_operand= args[0]->neg_transformer(thd)))
// args[0] has neg_tranformer
new_item= new(thd->mem_root) Item_func_xor(neg_operand, args[1]);
else if ((neg_operand= args[1]->neg_transformer(thd)))
// args[1] has neg_tranformer
new_item= new(thd->mem_root) Item_func_xor(args[0], neg_operand);
else
{
neg_operand= new(thd->mem_root) Item_func_not(args[0]);
new_item= new(thd->mem_root) Item_func_xor(neg_operand, args[1]);
}
return new_item;
}
/**
a IS NULL -> a IS NOT NULL.
*/
Item *Item_func_isnull::neg_transformer(THD *thd)
{
Item *item= new Item_func_isnotnull(args[0]);
return item;
}
/**
a IS NOT NULL -> a IS NULL.
*/
Item *Item_func_isnotnull::neg_transformer(THD *thd)
{
Item *item= new Item_func_isnull(args[0]);
return item;
}
Item *Item_cond_and::neg_transformer(THD *thd) /* NOT(a AND b AND ...) -> */
/* NOT a OR NOT b OR ... */
{
neg_arguments(thd);
Item *item= new Item_cond_or(list);
return item;
}
Item *Item_cond_or::neg_transformer(THD *thd) /* NOT(a OR b OR ...) -> */
/* NOT a AND NOT b AND ... */
{
neg_arguments(thd);
Item *item= new Item_cond_and(list);
return item;
}
Item *Item_func_nop_all::neg_transformer(THD *thd)
{
/* "NOT (e $cmp$ ANY (SELECT ...)) -> e $rev_cmp$" ALL (SELECT ...) */
Item_func_not_all *new_item= new Item_func_not_all(args[0]);
Item_allany_subselect *allany= (Item_allany_subselect*)args[0];
allany->func= allany->func_creator(FALSE);
allany->all= !allany->all;
allany->upper_item= new_item;
return new_item;
}
Item *Item_func_not_all::neg_transformer(THD *thd)
{
/* "NOT (e $cmp$ ALL (SELECT ...)) -> e $rev_cmp$" ANY (SELECT ...) */
Item_func_nop_all *new_item= new Item_func_nop_all(args[0]);
Item_allany_subselect *allany= (Item_allany_subselect*)args[0];
allany->all= !allany->all;
allany->func= allany->func_creator(TRUE);
allany->upper_item= new_item;
return new_item;
}
Item *Item_func_eq::negated_item() /* a = b -> a != b */
{
return new Item_func_ne(args[0], args[1]);
}
Item *Item_func_ne::negated_item() /* a != b -> a = b */
{
return new Item_func_eq(args[0], args[1]);
}
Item *Item_func_lt::negated_item() /* a < b -> a >= b */
{
return new Item_func_ge(args[0], args[1]);
}
Item *Item_func_ge::negated_item() /* a >= b -> a < b */
{
return new Item_func_lt(args[0], args[1]);
}
Item *Item_func_gt::negated_item() /* a > b -> a b */
{
return new Item_func_gt(args[0], args[1]);
}
/**
just fake method, should never be called.
*/
Item *Item_bool_rowready_func2::negated_item()
{
DBUG_ASSERT(0);
return 0;
}
Item_equal::Item_equal(Item_field *f1, Item_field *f2)
: Item_bool_func(), const_item(0), eval_item(0), cond_false(0),
compare_as_dates(FALSE)
{
const_item_cache= 0;
fields.push_back(f1);
fields.push_back(f2);
}
Item_equal::Item_equal(Item *c, Item_field *f)
: Item_bool_func(), eval_item(0), cond_false(0)
{
const_item_cache= 0;
fields.push_back(f);
const_item= c;
compare_as_dates= f->is_temporal_with_date();
}
Item_equal::Item_equal(Item_equal *item_equal)
: Item_bool_func(), eval_item(0), cond_false(0)
{
const_item_cache= 0;
List_iterator_fast li(item_equal->fields);
Item_field *item;
while ((item= li++))
{
fields.push_back(item);
}
const_item= item_equal->const_item;
compare_as_dates= item_equal->compare_as_dates;
cond_false= item_equal->cond_false;
}
bool Item_equal::compare_const(THD *thd, Item *c)
{
if (compare_as_dates)
{
cmp.set_datetime_cmp_func(this, &c, &const_item);
cond_false= cmp.compare();
}
else
{
Item_func_eq *func= new Item_func_eq(c, const_item);
if (func == NULL)
return true;
if (func->set_cmp_func())
return true;
func->quick_fix_field();
cond_false= !func->val_int();
}
if (thd->is_error())
return true;
if (cond_false)
const_item_cache= 1;
return false;
}
bool Item_equal::add(THD *thd, Item *c, Item_field *f)
{
if (cond_false)
return false;
if (!const_item)
{
DBUG_ASSERT(f);
const_item= c;
compare_as_dates= f->is_temporal_with_date();
return false;
}
return compare_const(thd, c);
}
bool Item_equal::add(THD *thd, Item *c)
{
if (cond_false)
return false;
if (!const_item)
{
const_item= c;
return false;
}
return compare_const(thd, c);
}
void Item_equal::add(Item_field *f)
{
fields.push_back(f);
}
uint Item_equal::members()
{
return fields.elements;
}
/**
Check whether a field is referred in the multiple equality.
The function checks whether field is occurred in the Item_equal object .
@param field field whose occurrence is to be checked
@retval
1 if nultiple equality contains a reference to field
@retval
0 otherwise
*/
bool Item_equal::contains(Field *field)
{
List_iterator_fast it(fields);
Item_field *item;
while ((item= it++))
{
if (field->eq(item->field))
return 1;
}
return 0;
}
/**
Join members of another Item_equal object.
The function actually merges two multiple equalities.
After this operation the Item_equal object additionally contains
the field items of another item of the type Item_equal.
If the optional constant items are not equal the cond_false flag is
set to 1.
@param thd thread handler
@param item multiple equality whose members are to be joined
@returns false if success, true if error
*/
bool Item_equal::merge(THD *thd, Item_equal *item)
{
fields.concat(&item->fields);
Item *c= item->const_item;
if (c)
{
/*
The flag cond_false will be set to 1 after this, if
the multiple equality already contains a constant and its
value is not equal to the value of c.
*/
if (add(thd, c))
return true;
}
cond_false|= item->cond_false;
return false;
}
/**
Order field items in multiple equality according to a sorting criteria.
The function perform ordering of the field items in the Item_equal
object according to the criteria determined by the cmp callback parameter.
If cmp(item_field1,item_field2,arg)const_item() &&
/*
Don't propagate constant status of outer-joined column.
Such a constant status here is a result of:
a) empty outer-joined table: in this case such a column has a
value of NULL; but at the same time other arguments of
Item_equal don't have to be NULLs and the value of the whole
multiple equivalence expression doesn't have to be NULL or FALSE
because of the outer join nature;
or
b) outer-joined table contains only 1 row: the result of
this column is equal to a row field value *or* NULL.
Both values are inacceptable as Item_equal constants.
*/
!item->is_outer_field())
{
it.remove();
if (add(thd, item))
return true;
}
}
return false;
}
bool Item_equal::fix_fields(THD *thd, Item **ref)
{
List_iterator_fast li(fields);
Item *item;
not_null_tables_cache= used_tables_cache= 0;
const_item_cache= 0;
while ((item= li++))
{
used_tables_cache|= item->used_tables();
not_null_tables_cache|= item->not_null_tables();
maybe_null|= item->maybe_null;
}
fix_length_and_dec();
fixed= 1;
return 0;
}
/**
Get filtering effect for multiple equalities, i.e.
"tx.col = value_1 = ... = value_n" where value_i may be a
constant, a column etc.
The multiple equality only contributes to the filtering effect for
'filter_for_table' if
a) A column in 'filter_for_table' is referred to
b) at least one value_i is a constant or a column in a table
already read
If this multiple equality refers to more than one column in
'filter_for_table', the predicates on all these fields will
contribute to the filtering effect.
*/
float Item_equal::get_filtering_effect(table_map filter_for_table,
table_map read_tables,
const MY_BITMAP *fields_to_ignore,
double rows_in_table)
{
// This predicate does not refer to a column in 'filter_for_table'
if (!(used_tables() & filter_for_table))
return COND_FILTER_ALLPASS;
float filter= COND_FILTER_ALLPASS;
/*
Keep track of whether or not a usable value that is either a
constant or a column in an already read table has been found.
*/
bool found_comparable= false;
// Is there a constant that this multiple equality is equal to?
if (const_item)
found_comparable= true;
List_iterator it(fields);
Item_field *cur_field;
/*
Calculate filtering effect for all applicable fields. If this
item has multiple fields from 'filter_for_table', each of these
fields will contribute to the filtering effect.
*/
while ((cur_field= it++))
{
if (cur_field->used_tables() & read_tables)
{
// cur_field is a field in a table earlier in the join sequence.
found_comparable= true;
}
else if (cur_field->used_tables() == filter_for_table)
{
if (bitmap_is_set(fields_to_ignore, cur_field->field->field_index))
{
/*
cur_field is a field in 'filter_for_table', but it is a
field which already contributes to the filtering effect.
Its value can still be used as a constant if another column
in the same table is referred to in this multiple equality.
*/
found_comparable= true;
}
else
{
/*
cur_field is a field in 'filter_for_table', and it's not one
of the fields that must be ignored
*/
float cur_filter=
cur_field->get_cond_filter_default_probability(rows_in_table,
COND_FILTER_EQUALITY);
// Use index statistics if available for this field
if (!cur_field->field->key_start.is_clear_all())
{
// cur_field is indexed - there may be statistics for it.
const TABLE *tab= cur_field->field->table;
for (uint j= 0; j < tab->s->keys; j++)
{
if (cur_field->field->key_start.is_set(j) &&
tab->key_info[j].has_records_per_key(0))
{
cur_filter= static_cast(tab->key_info[j].records_per_key(0) / rows_in_table);
break;
}
}
/*
Since rec_per_key and rows_per_table are calculated at
different times, their values may not be in synch and thus
it is possible that cur_filter is greater than 1.0 if
rec_per_key is outdated. Force the filter to 1.0 in such
cases.
*/
if (cur_filter >= 1.0)
cur_filter= 1.0f;
}
filter*= cur_filter;
}
}
}
return found_comparable ? filter : COND_FILTER_ALLPASS;
}
void Item_equal::update_used_tables()
{
List_iterator_fast li(fields);
Item *item;
not_null_tables_cache= used_tables_cache= 0;
if ((const_item_cache= cond_false))
return;
with_subselect= false;
with_stored_program= false;
while ((item=li++))
{
item->update_used_tables();
used_tables_cache|= item->used_tables();
not_null_tables_cache|= item->not_null_tables();
/* see commentary at Item_equal::update_const() */
const_item_cache&= item->const_item() && !item->is_outer_field();
with_subselect|= item->has_subquery();
with_stored_program|= item->has_stored_program();
}
}
longlong Item_equal::val_int()
{
Item_field *item_field;
if (cond_false)
return 0;
List_iterator_fast it(fields);
Item *item= const_item ? const_item : it++;
eval_item->store_value(item);
if ((null_value= item->null_value))
return 0;
while ((item_field= it++))
{
/* Skip fields of non-const tables. They haven't been read yet */
if (item_field->field->table->const_table)
{
const int rc= eval_item->cmp(item_field);
if ((rc == TRUE) || (null_value= (rc == UNKNOWN)))
return 0;
}
}
return 1;
}
void Item_equal::fix_length_and_dec()
{
Item *item= get_first();
eval_item= cmp_item::get_comparator(item->result_type(), item,
item->collation.collation);
}
bool Item_equal::walk(Item_processor processor, enum_walk walk, uchar *arg)
{
if ((walk & WALK_PREFIX) && (this->*processor)(arg))
return true;
List_iterator_fast it(fields);
Item *item;
while ((item= it++))
{
if (item->walk(processor, walk, arg))
return true;
}
return (walk & WALK_POSTFIX) && (this->*processor)(arg);
}
Item *Item_equal::transform(Item_transformer transformer, uchar *arg)
{
DBUG_ASSERT(!current_thd->stmt_arena->is_stmt_prepare());
List_iterator it(fields);
Item *item;
while ((item= it++))
{
Item *new_item= item->transform(transformer, arg);
if (!new_item)
return 0;
/*
THD::change_item_tree() should be called only if the tree was
really transformed, i.e. when a new item has been created.
Otherwise we'll be allocating a lot of unnecessary memory for
change records at each execution.
*/
if (new_item != item)
current_thd->change_item_tree((Item **) it.ref(), new_item);
}
return Item_func::transform(transformer, arg);
}
void Item_equal::print(String *str, enum_query_type query_type)
{
str->append(func_name());
str->append('(');
List_iterator_fast it(fields);
Item *item;
if (const_item)
const_item->print(str, query_type);
else
{
item= it++;
item->print(str, query_type);
}
while ((item= it++))
{
str->append(',');
str->append(' ');
item->print(str, query_type);
}
str->append(')');
}
longlong Item_func_trig_cond::val_int()
{
if (trig_var == NULL)
{
DBUG_ASSERT(m_join != NULL && m_idx >= 0);
switch(trig_type)
{
case IS_NOT_NULL_COMPL:
trig_var= &m_join->qep_tab[m_idx].not_null_compl;
break;
case FOUND_MATCH:
trig_var= &m_join->qep_tab[m_idx].found;
break;
default:
DBUG_ASSERT(false); /* purecov: inspected */
return 0;
}
}
return *trig_var ? args[0]->val_int() : 1;
}
void Item_func_trig_cond::print(String *str, enum_query_type query_type)
{
/*
Print:
(, condition, TRUE)
which means: if a certain property () is true, then return
the value of , else return TRUE. If source tables are
present, they are the owner of the property.
*/
str->append(func_name());
str->append("(");
switch(trig_type)
{
case IS_NOT_NULL_COMPL:
str->append("is_not_null_compl");
break;
case FOUND_MATCH:
str->append("found_match");
break;
case OUTER_FIELD_IS_NOT_NULL:
str->append("outer_field_is_not_null");
break;
default:
DBUG_ASSERT(0);
}
if (m_join != NULL)
{
/*
Item printing is done at various stages of optimization, so there can
be a JOIN_TAB or a QEP_TAB.
*/
TABLE *table, *last_inner_table;
plan_idx last_inner;
if (m_join->qep_tab)
{
QEP_TAB *qep_tab= &m_join->qep_tab[m_idx];
table= qep_tab->table();
last_inner= qep_tab->last_inner();
last_inner_table= m_join->qep_tab[last_inner].table();
}
else
{
JOIN_TAB *join_tab= m_join->best_ref[m_idx];
table= join_tab->table();
last_inner= join_tab->last_inner();
last_inner_table= m_join->best_ref[last_inner]->table();
}
str->append("(");
str->append(table->alias);
if (last_inner != m_idx)
{
/* case of t1 LEFT JOIN (t2,t3,...): print range of inner tables */
str->append("..");
str->append(last_inner_table->alias);
}
str->append(")");
}
str->append(", ");
args[0]->print(str, query_type);
str->append(", true)");
}
/**
Get item that can be substituted for the supplied item.
@param field field item to get substitution field for, which must be
present within the multiple equality itself.
@retval Found substitution item in the multiple equality.
@details Get the first item of multiple equality that can be substituted
for the given field item. In order to make semijoin materialization strategy
work correctly we can't propagate equal fields between a materialized
semijoin and the outer query (or any other semijoin) unconditionally.
Thus the field is returned according to the following rules:
1) If the given field belongs to a materialized semijoin then the
first field in the multiple equality which belongs to the same semijoin
is returned.
2) If the given field doesn't belong to a materialized semijoin then
the first field in the multiple equality is returned.
*/
Item_field* Item_equal::get_subst_item(const Item_field *field)
{
DBUG_ASSERT(field != NULL);
const JOIN_TAB *field_tab= field->field->table->reginfo.join_tab;
/*
field_tab is NULL if this function was not called from
JOIN::optimize() but from e.g. mysql_delete() or mysql_update().
In these cases there is only one table and no semijoin
*/
if (field_tab &&
sj_is_materialize_strategy(field_tab->get_sj_strategy()))
{
/*
It's a field from a materialized semijoin. We can substitute it only
with a field from the same semijoin.
Example: suppose we have a join_tab order:
ot1 ot2 ot3 SJM(it1 it2 it3)
is the temporary table that is materialized from the join
of it1, it2 and it3.
and equality ot2.col = .col = it1.col = it2.col
If we're looking for best substitute for 'it2.col', we must pick it1.col
and not ot2.col. it2.col is evaluated while performing materialization,
when the outer tables are not available in the execution.
Note that subquery materialization does not have the same problem:
even though IN->EXISTS has injected equalities involving outer query's
expressions, it has wrapped those expressions in variants of Item_ref,
never Item_field, so they can be part of an Item_equal only if they are
constant (in which case there is no problem with choosing them below);
@see check_simple_equality().
*/
List_iterator it(fields);
Item_field *item;
plan_idx first= field_tab->first_sj_inner(), last= field_tab->last_sj_inner();
while ((item= it++))
{
plan_idx idx= item->field->table->reginfo.join_tab->idx();
if (idx >= first && idx