Ignoring error codes from type conversion allows default (wrong) values to
go unnoticed in the formation of index search conditions.
Fixed by correctly checking for conversion errors.
The bug could cause choosing a sub-optimal execution plan for
a single-table query if a unique index with many null keys were
defined for the table.
It happened because the code of the check_quick_keys function
made an assumption that any key may occur in an unique index
only once. Yet this is not true for keys with nulls that may
have multiple occurrences in the index.
We use INT_RESULT type if all arguments are of type INT for 'if', 'case',
'coalesce' functions regardless of arguments' unsigned flag, so sometimes we can
exceed the INT bounds.
After fix for bug#21798 JOIN stores the pointer to the buffer for sorting
fields. It is used while sorting for grouping and for ordering. If ORDER BY
clause has more elements then the GROUP BY clause then a memory overrun occurs.
Now the length of the ORDER BY list is always passed to the
make_unireg_sortorder() function and it allocates buffer big enough to be
used for bigger list.
With MySQL 3.23 and 4.0, the syntax 'LIMIT N, -1' is accepted, and returns
all the rows located after row N. This behavior, however, is not the
intended result, and defeats the purpose of LIMIT, which is to constrain
the size of a result set.
With MySQL 4.1 and later, this construct is correctly detected as a syntax
error.
This fix does not change the production code, and only adds a new test case
to improve test coverage in this area, to enforce in the test suite the
intended behavior.
Post-commit issues fixed
* Test results for other tests fixed due to added error #s
* Memory allocation/free issues found with running with valgrind
* Fix to mysql-test-run shell script to run federated_server test (installs
mysql.servers table properly)
account predicates that become sargable after reading const tables.
In some cases this resulted in choosing non-optimal execution plans.
Now info of such potentially saragable predicates is saved in
an array and after reading const tables we check whether this
predicates has become saragable.
strings
MySQL is setting the flag HA_END_SPACE_KEYS for all the keys that reference
text or varchar columns with collation different than binary.
This was done to handle correctly the situation where a lookup on such a key
may return more than 1 row because of the presence of many rows that differ
only by the amount of trailing space in the table's string column.
Inserting such values however appears to violate the unique checks on
INSERT/UPDATE. Thus that flag must not be set as it will prevent the optimizer
from choosing a faster access method.
This fix removes the setting of the HA_END_SPACE_KEYS flag.
After the patch for big 21698 equality propagation stopped
working for BETWEEN and IN predicates with STRING arguments.
This changeset completes the solution of the above patch.
while space allocation
Under some circumstances DISTINCT clause can be converted to grouping.
In such cases grouping is performed by all items in the select list.
If an ORDER clause is present then items from it is prepended to group list.
But the case with ORDER wasn't taken into account when allocating the
array for sum functions. This leads to memory corruption and crash.
The JOIN::alloc_func_list() function now allocates additional space if there
is an ORDER by clause is specified and DISTINCT -> GROUP BY optimization is
possible.
const tables. This resulted in choosing extremely inefficient
execution plans in same cases when distribution of data in
joined were skewed (see the customer test case for the bug).
- Make the range-et-al optimizer produce E(#table records after table
condition is applied),
- Make the join optimizer use this value,
- Add "filtered" column to EXPLAIN EXTENDED to show
fraction of records left after table condition is applied
- Adjust test results, add comments
When optimizing conditions like 'a = <some_val> OR a IS NULL' so that they're
united into a single condition on the key and checked together the server must
check which value is the NULL value in a correct way : not only using ->is_null
but also check if the expression doesn't depend on any tables referenced in the
current statement.
This additional check must be performed because that optimization takes place
before the actual execution of the statement, so if the field was initialized
to NULL from a previous statement the optimization would be applied incorrectly.
The problem was in that opt_sum_query() replaced MIN/MAX functions
with the corresponding constant found in a key, but due to imprecise
representation of float numbers, when evaluating the where clause,
this comparison failed.
When MIN/MAX optimization detects that all tables can be removed,
also remove all conjuncts in a where clause that refer to these
tables. As a result of this fix, these conditions are not evaluated
twice, and in the case of float number comparisons we do not discard
result rows due to imprecise float representation.
As a side-effect this fix also corrects an unnoticed problem in
bug 12882.
The problem was that when converting a string to an exact number,
rounding didn't work, because conversion didn't understand
approximate numbers notation.
Fix: a new function for string-to-number conversion was implemented,
which is aware of approxinate number notation (with decimal point
and exponent, e.g. -19.55e-1)
When an alias is set to a column leading spaces are removed from the alias.
But when this is done on aliases set by user this can lead to confusion.
Now Item::set_name() method issues the warning if leading spaces were removed
from an alias set by user.
New warning message is added.
Adding decimal "digits" in multiplication resulted in signed overflow and
producing wrong results.
Fixed by using large enough buffers and intermediary result types :
dec2 (currently longlong) to hold result of adding decimal "digits"
(currently int32).
Added test case for bug#18759 Incorrect string to numeric conversion.
select.test:
Added test case for bug#18759 Incorrect string to numeric conversion.
item_cmpfunc.cc:
Cleanup after fix for bug#18360 removal
The Field::eq() considered instances of Field_bit that differ only in
bit_ptr/bit_ofs equal. This caused equality conditions optimization
(build_equal_items_for_cond()) to make bad field substitutions that result
in wrong predicates.
Field_bit requires an overloaded eq() function that checks the bit_ptr/bit_ofs
in addition to Field::eq().
3.23 regression test failure
The member SEL_ARG::min_flag was not initialized,
due to which the condition for no GEOM_FLAG in function
key_or did not choose "Range checked for each record" as
the correct access method.
a worse execution plan than in 4.1 for some queries.
It happened due the fact that at some conditions the
optimizer always preferred range or full index scan access
methods to lookup access methods even when the latter were much
cheaper.
The problem was not observed in 4.1 for the reported query
because the WHERE condition was not of a form that could
cause the problem.
Equality propagation introduced on 5.0 added an extra
predicate and changed the WHERE condition. The new condition
provoked the optimizer to make a bad choice.
The problem was fixed by the patch for bug 17379.
Re-work best_access_path() and find_best() to reuse E(#rows(range access)) as
E(#rows(ref[_or_null](const) access) only when it is appropriate.
[This is the final cumulative patch]
Multiple equalities were not adjusted after reading constant tables.
It resulted in neglecting good index based methods that could be
used to access of other tables.
Absence of table in left part of LEFT/RIGHT join wasn't checked before
name resolution which resulted in NULL dereferencing and server crash.
Modified rules:
"table_ref LEFT opt_outer JOIN_SYM table_ref" and "table_ref RIGHT opt_outer
JOIN_SYM table_ref"
NULL check is moved before push_new_name_resolution_context()
functions are involved.
When subselect is a join with set functions and no record have been found in
it, end_send_group() sets null_row for all tables in order aggregate functions
to calculate their values correctly. Normally this null_row flag is cleared for
each table in sub_select(), but flush_cached_records() doesn't do so.
Due to this all fields from the table processed by flush_cached_records() are
always evaluated as nulls and whole select produces wrong result.
flush_cached_records() now clears null_row flag at the very beginning.
select result
Item equal objects are employed only at the optimize phase. Usually they are not
supposed to be evaluated. Yet in some cases we call the method val_int() for
them. Here we have to take care of restricting the predicate such an object
represents f1=f2= ...=fn to the projection of known fields fi1=...=fik.
Added a check for field's table being const in Item_equal::val_int().
If the field's table is not const val_int() just skips that field when
evaluating Item_equal.
cmp_item_sort_string::cmp() wasn't checking values_res variable for null.
Later called function was dereferenced it and crashed server.
Added null check to cmp_item_sort_string::cmp().
crash
resolve_const_item() substitutes item which will evaluate to constant with
equvalent constant item, basing on the item's result type. In this case
subselect was resolved as constant, and resolve_const_item() was substituting
it's result's Item_caches to Item_null. Later Item_cache's function was called
for Item_null object, which caused server crash.
resolve_const_item() now substitutes constants for items with
result_type == ROW_RESULT only for Item_rows.
the same column as an aliased and as a non-aliased column.
The problem was that Item_direct_view_ref::eq() was first comparing view columns
by name, and in this case the name of one of them is different since it is aliased.
Invalid date like 2000-02-32 wasn't converted to int, which lead to not
using index and comparison with field as astring, which results in slow
query execution.
convert_constatn_item() and get_mm_leaf() now forces MODE_INVALID_DATES to
allow such conversion.
The cause for the bug is that the priorities of all rules/terminals
that process the FROM clause are not fully specified, and the
parser generator produces a parser that doesn't always parse
the FROM clause so that JOINs are left-associative. As a result
the final join tree produced by the parser is incorrect, which
is the cause for subsequent name resolution to fail.
ESCAPE has length of 1 if specified and sql_mode is NO_BACKSLASH_ESCAPES
or has length of 0 or 1 in every other situation.
(approved patch applied on a up-to-date tree re-commit)
DISTINCT wasn't optimized away and caused creation of tmp table in wrong
case. This result in integer overrun and running out of memory.
Fix backported from 4.1. Now if optimizer founds that in result be only 1
row it removes distinct.
field::sort_key() now adds length last for varbinary/blob
VARBINARY/BLOB is now sorted by filesort so that shorter strings comes before longer ones
Fixed issues in test cases from last merge
When fixing Item_func_plus in ORDER BY clause field c is searched in all
opened tables, but because c is an alias it wasn't found there.
This patch adds a flag to select_lex which allows Item_field::fix_fields()
to look up in select's item_list to find aliased fields.
After SHOW TABLE STATUS last_insert_id wasn't cleaned, and next select
erroneously rewrites WHERE condition and returs a row;
5.0 isn't affected because of different SHOW TABLE STATUS handling.
last_insert_id cleanup added to mysqld_extend_show_tables().
The problem was in that when finding the last table reference in a nested join tree,
the procedure doing the iteration over the right-most branches of a join tree
was testing for RIGHT JOINs the table reference that represents the join, and not
the second operand of the JOIN. Currently the information whether a join is LEFT/RIGHT
is stored not on the join object itself, but on one of its operands.
resolve_const_item() assumed to be not called for Item_row items. For
ensuring that DBUG_ASSERT(0) was set there.
This patch adds section for Item_row items. If it can it recursively calls
resolve_const_item() for each item the Item_row contains. If any of the
contained items is null then whole Item_row substitued by Item_null. Otherwise
it just returns.
The problem was in the way table references are pre-filtered when
resolving a qualified field. When resolving qualified table references
we search recursively in the operands of the join. If there is
natural/using join with a merge view, the first call to find_field_in_table_ref
makes a recursive call to itself with the view as the new table reference
to search for the column. However the view has both nested_join and
join_columns != NULL so it skipped the test whether the view name matches
the field qualifier. As a result the field was found in the view since the
view already has a field with the same name. Thus the field was incorrectly
resolved as the view field.
Optimizer did choose "Range checked for each record" for one of the tables.
For first few loops over that table it choose sequential access, on later
stage it choose to use index. Because table was previously initialized for
sequential access, it skips intitialization for index access, and when
server tries to retrieve data error occurs.
QUICK_RANGE_SELECT::init() changes so if file already initialized for
sequential access, it calls ha_rnd_end() and initializes file for index
access.
The problem was that in the first production in rule 'join_table', that
processes simple cross joins, the parser was processing the second join operand
before the first one due to unspecified priorities of JOINs. As a result in the
case of cross joins the parser constructed a tree with incorrect nesting:
the expression "t1 join t2 join t3 on some_cond" was interpreted as
"t1 join (t2 join t3 on some_cond)" instead of
"(t1 join t2) join t3 on some_cond".
Because of this incorrect nesting the method make_join_on_context picked an
incorrect table as the first table of the name resolution context.
The solution assignes correct priorities to the related production.
(Server crash on DO IFNULL(NULL,NULL)
(fixes also "SELECT CAST(IFNULL(NULL,NULL) as DECIMAL)" unreported
crash)
(new revampled fix with suggestions from Igor)
* Provide backwards compatibility extension to name resolution of
coalesced columns. The patch allows such columns to be qualified
with a table (and db) name, as it is in 4.1.
Based on a patch from Monty.
* Adjusted tests accordingly to test both backwards compatible name
resolution of qualified columns, and ANSI-style resolution of
non-qualified columns.
For this, each affected test has two versions - one with qualified
columns, and one without.
When copying varchar fields with field_conv() it's not taken into account
that length_bytes of source and destination fields may be different.
This results in saving wrong data in field and making wrong key later.
Added check so if fields are varchar and have different length_bytes they
are not copied by memcpy().
"Process NATURAL and USING joins according to SQL:2003".
* Some of the main problems fixed by the patch:
- in "select *" queries the * expanded correctly according to
ANSI for arbitrary natural/using joins
- natural/using joins are correctly transformed into JOIN ... ON
for any number/nesting of the joins.
- column references are correctly resolved against natural joins
of any nesting and combined with arbitrary other joins.
* This patch also contains a fix for name resolution of items
inside the ON condition of JOIN ... ON - in this case items must
be resolved only against the JOIN operands. To support such
'local' name resolution, the patch introduces a stack of
name resolution contexts used at parse time.
NOTICE:
- This patch is not complete in the sense that
- there are 2 test cases that still do not pass -
one in join.test, one in select.test. Both are marked
with a comment "TODO: WL#2486".
- it does not include a new test specific for the task
