Hybrid functions (IF, COALESCE, etc) did not preserve the JSON property
from their arguments. The same problem was repeatable for single row subselects.
The problem happened because the method Item::is_json_type() was inconsistently
implemented across the Item hierarchy. For example, Item_hybrid_func
and Item_singlerow_subselect did not override is_json_type().
Solution:
- Removing Item::is_json_type()
- Implementing specific JSON type handlers:
Type_handler_string_json
Type_handler_varchar_json
Type_handler_tiny_blob_json
Type_handler_blob_json
Type_handler_medium_blob_json
Type_handler_long_blob_json
- Reusing the existing data type infrastructure to pass JSON
type handlers across all item types, including classes Item_hybrid_func
and Item_singlerow_subselect. Note, these two classes themselves do not
need any changes!
- Extending the data type infrastructure so data types can inherit
their properties (e.g. aggregation rules) from their base data types.
E.g. VARCHAR/JSON acts as VARCHAR, LONGTEXT/JSON acts as LONGTEXT
when mixed to a non-JSON data type. This is done by:
- adding virtual method Type_handler::type_handler_base()
- adding a helper class Type_handler_pair
- refactoring Type_handler_hybrid_field_type methods
aggregate_for_result(), aggregate_for_min_max(),
aggregate_for_num_op() to use Type_handler_pair.
This change also fixes:
MDEV-27361 Hybrid functions with JSON arguments do not send format metadata
Also, adding mtr tests for JSON replication. It was not covered yet.
And the current patch changes the replication code slightly.
Also fixes MDEV-24619 Wrong result or Assertion `0' in Item::val_native / Type_handler_inet6::Item_val_native_with_conversion
Type_handler_inet6::create_item_copy() created a generic Item_copy_string,
which does not implement val_native() - it has a dummy implementation
with DBUG_ASSERT(0), which made the server crash.
Fix:
- Adding a new class Type_handler_inet6
which implements val_native().
- Fixing Type_handler_inet6::create_item_copy()
to make Item_copy_inet6 instead of Item_copy_string.
IFNULL(inet6_not_null_expr, 'foo') erroneously set its nullability to NOT NULL.
Fix:
- Moving the line "maybe_null= args[1]->maybe_null" before the call
of fix_length_and_dec2(), so the call of Type_handler method
Item_hybrid_func_fix_attributes() can reset it when desired.
- Fixing Type_handler_inet6::Item_hybrid_func_fix_attributes()
to ignore args[0] when detecting nullability of IFNULL().
for the functionality that was part of the server for years
and is moved in 10.5 into a [statically linked] plugin
without changing the underlying code.
When some expression of an INET6 data type involves conversion to INET6 from
other data types, e.g. in:
- CAST:
SELECT CAST(non_inet6_expr AS INET6)
- CASE and hybrid functions:
SELECT CASE WHEN expr THEN inet6_expr ELSE non_inet6_expr END
- UNION:
SELECT inet6_expr UNION SELECT non_inet6_expr
the result column must be fixed as NULL-able even if the non-inet6 expression itself
is not NULL-able, because at the execution time the conversion can fail.
Details:
- Forcing NULL-ability if conversion from some data type to INET6 is involved
(for non-constant or for expensive expressions).
- Non-expensive constant expressions are tested for NULL-ability at fix_fields() time,
so things like `CAST('::' AS INET6)` are still detected as NOT NULL.
- Adding "bool warn" parameter into a few methods, to avoid redundant warnings
at fix_fields() time when calculating NULL-ability of constant values.
For field with type INET, during EITS collection the min and max values are store in text
representation in the statistical table.
While retrieving the value from the statistical table, the value is stored back in the original
field using binary form instead of text and this was resulting in the crash.
Introduced 2 functions in the Field structure:
1) store_to_statistical_minmax_field
2) store_from_statistical_minmax_field
Item_sum_sp did not override val_native(). So the reported script
crashed in the default implementation in Item::val_native() on DBUG_ASSERT().
Implementing a correct Item_sum_sp::val_native().
When my_vsnprintf() is patched, the code protected disabled with
'WAITING_FOR_BUGFIX_TO_VSPRINTF' should be enabled again. Also all %b
formats in this patch should be revert to %s again
This task deals with packing the sort key inside the sort buffer, which would
lead to efficient usage of the memory allocated for the sort buffer.
The changes brought by this feature are
1) Sort buffers would have sort keys of variable length
2) The format for sort keys inside the sort buffer would look like
|<sort_length><null_byte><key_part1><null_byte><key_part2>.......|
sort_length is the extra bytes that are required to store the variable
length of a sort key.
3) When packing of sort key is done we store the ORIGINAL VALUES inside
the sort buffer and not the STRXFRM form (mem-comparable sort keys).
4) Special comparison function packed_keys_comparison() is introduced
to compare 2 sort keys.
This patch also contains contributions from Sergei Petrunia.
This patch changes the way how INET6 is packed to the RBR binary log:
- from fixed length 16 bytes
- to BINARY(16) compatible variable length style
with trailing 0x00 byte compression.
This is to make INET6 fully compatible with BINARY(16) in RBR binary logs,
so RBR replication works in this scenarios:
- Old master BINARY(16) -> New slave INET6
- New master INET6 -> Old slave BINARY(16)
A new class StringPack was added to share the code between
Field_string and Field_inet6.
- Defining MariaDB_FUNCTION_PLUGIN
- Changing the code in /plugins/type_inet/ and /plugins/type_test/
to use MariaDB_FUNCTION_PLUGIN instead of MariaDB_FUNCTION_COLLECTION_PLUGIN.
- Changing maturity for the INET6 data type plugin from experimental to alpha.
This clause in CREATE TABLE:
PARTITION BY LIST COLUMNS (inet6column)
(PARTITION p1 VALUES IN ('::'))
was erroneously written to frm file as:
PARTITION BY LIST COLUMNS(inet6column)
(PARTITION p1 VALUES IN (_binary 0x3A3A))
I.e. the text value '::' was converted to HEX representation
and prefixed with _binary.
A simple fix could write `_latin1 0x3A3A` instead of `_binary 0x3A3A`,
but in case of INET6 we don't need neither character set introducers,
nor HEX encoding, because text representation of INET6 values consist
of pure ASCII characters.
So this patch changes the above clause to be printed as:
PARTITION BY LIST COLUMNS(inet6column)
(PARTITION p1 VALUES IN ('::'))
Details:
The old code in check_part_field() was not friendly to pluggable data types.
Replacing this function to two new virtual methods in Type_handler:
virtual bool partition_field_check(const LEX_CSTRING &field_name,
Item *item_expr) const;
virtual bool partition_field_append_value(String *str,
Item *item_expr,
CHARSET_INFO *field_cs,
partition_value_print_mode_t mode)
const;
so data type plugins can decide whether they need to use character set
introducer and/or hex encoding when printing partition values.
The code erroneously assumed that only Field_str descendants can
store character set information. After adding Field_inet6, it's
not true anymore.
Also, after adding Field_inet6, storing field->charset() become not correct either:
- Field_inet6::charset() return &my_charset_latin1,
because clients see INET6 as VARCHAR(39).
- Field_inet6::binlog_type_info().m_cs returns &my_charset_bin
because storage engines see INET6 as BINARY(16).
We need to store &my_charset_bin to the binlog metadata,
so the slave sees INET6 as BINARY(16), like storage engines do,
to make the slave treat the replicated data as binary IPv6 address
representation (rather than text representation).
The correct character set that needs to be stored to the metadata
is already populated to binlog_type_info_array[i].m_cs. So the
fixed code version uses this value rather than field->charset().
Type_handler_inet6::is_val_native_ready() was not overriden and
returned "false" by default, so Item_sum_min_max::update_field()
erroneously went through the min_max_update_str_field() rather than
min_max_update_native_field() execution path. As a result '8888::'
was compared to 'fff::' in string format (rather than INET6 binary format)
and gave "less" instead of "greater".
Adding the forgotten overriding method returning "true".
The patch for `MDEV-20795 CAST(inet6 AS BINARY) returns wrong result`
unintentionally changed what Item_char_typecast::type_handler()
returns. This broke UNIONs with the BINARY() function, as the Aria
engine started to get columns of unexpected data types.
Restoring previous behaviour, to return
Type_handler::string_type_handler(max_length).
The prototype for Item_handed_func::return_type_handler() has changed
from:
const Type_handler *return_type_handler() const
to:
const Type_handler *return_type_handler(const Item_handled_func *) const
Suppress warnings when Field_inet6::store() is called from
read_statistics_for_table() and other optimizer related routines.
This patch does for Field_inet6 the same thing with what Monty's
patch previously did for other Field_xxx:
> commit 1bbe8c5e0f
> Author: Michael Widenius <monty@mariadb.org>
> Date: Sun Sep 22 04:08:48 2019 +0300
>
> Proper fix for disabling warnings in read_statistics_for_table().
> MDEV-20589: Server still crashes in Field::set_warning_truncated_wrong_value
Alas, some meaningful warnings disappeared.
