The MariaDB handler API changed for write_row(). The 'buf' parameter is
now a 'const'. This meant that our implementation didn't match the
virtual call so ours was no longer called. This implemented the 'const'.
Used Item attribute getters introduced by 10.4
Make changes to support Item::CONST_ITEM introduced by 10.4
as a replacement for INT_,REAL_,STRING_ ITEM.
Replaced QT_INFINIDB_DERIVED and similar flags
with correponded flags for Item->print().
Replaced or commented out infinidb_ variable names with columnstore_
where applicable.
SELECT_LEX had been moved in THD so changed all references.
Avoid writing CS decimal scales into MDB decimal fields
d-only dec attribute. WIP
Replaced infinidb_vtable with a singleton MIGR.
Merged with MCOL-2121.
Added new wsrep include paths needed by UDaF code.
Removed .vcxproj from Connector code.
If the first byte of a char/varchar was > 0x80 then it will break the
min/max values for an extent during cpimport. This patch makes the
min/max compare unsigned and only switches to signed when storing.
In addition send all the LDI / INSERT...SELECT data to cpimport, not
truncated. Let cpimport figure out the truncation point.
This patch will allow MariaDB replication into UM1 when enabling the
following is added to the SystemConfig section of Columnstore.xml:
<ReplicationEnabled>Y</ReplicationEnabled>
The intended use case is to replication from an InnoDB MariaDB server
into ColumnStore. You would need to create the tables on the ColumnStore
slave as "ColumnStore" and the same tables in the master as InnoDB.
At the moment the use case is narrow and could be prone to problems so
this will use the hidden flag until we can improve it.
If a VARCHAR was defined as less than 255 characters and a data via LDI
or INSERT...SELECT was > 127 characters the length field would contain a
negative number.
In 1.0.11 and 1.1.0 this was not a problem because we cast the negative
back again. With MCOL-877 we stoped doing the double-cast. This patch
casts properly the first time.
BLOBs that are longer than half the maximum length of the column would
be turned into a negative length due to signed int casting. This would
cause them to turn into NULL inserts. Unsigned now used.
* 64KB TEXT column had off-by-one length pointer counting
* TEXT I_S/LDI was looping where it shouldn't causing pointer issues
* TEXT data type wasn't fully understood by cpimport
* TEXT and BLOB now have separate identifiers internally
* TEXT columns are identified as such in system catalog
* cpimport only requires hex input for BLOB, not TEXT
This does the following:
* Switch resource manager to a singleton which reduces the amount of
times the XML data is scanned and objects allocated.
* Make the I_S tables use the FE implementation of the system catalog
* Make the I_S.columnstore_columns table use the RID list cache
* Make the extentmap pre-allocate a vector instead of many small allocs
When a table has 8 columns that could be NULL and a NOT NULL column
after the NULLable columns the check to see if we have gone over the
NULL bitmap byte limit is run prematurely trigging an error.
This patch moves the check to only run when we are looking at NULLable
columns.
This changes the warning for truncation to the correct MariaDB error
code (1264).
In addition it passes the strict mode up into the DML class to roll back
correctly.
It also sets the abort_on_warning flag for updates as this isn't set on
the rnd_init phase but is needed for strict mode to work.
When infinidb_use_import_for_batchinsert is enabled the NULL handling
for the batch insert was incorrect. This is due to an off-by-one on the
NULL bitmap. This also affects INSERT...SELECT.
NULL is now pushed through the MariaDB storage engine plugin down to the
insert processing. A '0000-00-00' date is now a separate value to NULL.
This is more in-line with MariaDB's handling.
