maria-discuss team mailing list archive

[Sergei Golubchik <serg@xxxxxxxxxxx>]

Hi, Tom!

On Mar 01, Tom Worster wrote:
> I discovered this showstopper on Friday while working with real data in
> development of a new app. I managed to narrow it down today and filed the
> bug report.
> 
> https://mariadb.atlassian.net/browse/MDEV-7650
> 
> Looks like Maria saves an illegally formatted dyncol string if a dynamic
> column is longer than 64kB and is not alphabetically last by name among the
> dynamic column in the blob.
> 
> Tbh, even is this were fixed tomorrow so I could proceed with my work, I'm
> not sure I would use dyncols. My confidence in the feature has been shaken.
> It's a pity because the Active Record ORM extension I wrote was working.

You could've paid attention to warnings from your test case:

===========
MariaDB [test]> insert into t (a, b, dcols) values (1, 'two', column_create('one', 1, 'css', @txt));
Query OK, 1 row affected, 1 warning (0.00 sec)

Warning (Code 1265): Data truncated for column 'dcols' at row 1
===========

Indeed, blob size is limited by 64K. Another test case for your bug is:

===========
MariaDB [test]> select length(column_create('one', 1, 'txt', @txt));
+----------------------------------------------+
| length(column_create('one', 1, 'css', @txt)) |
+----------------------------------------------+
|                                        65563 |
+----------------------------------------------+
1 row in set (0.00 sec)

MariaDB [test]> select column_check(left(column_create('one', 1, 'txt', @txt), 65535));
+-----------------------------------------------------------------+
| column_check(left(column_create('one', 1, 'txt', @txt), 65535)) |
+-----------------------------------------------------------------+
|                                                               1 |
+-----------------------------------------------------------------+
1 row in set (0.00 sec)

MariaDB [test]> select column_check(left(column_create('one', 1, 'css', @txt), 65535));
+-----------------------------------------------------------------+
| column_check(left(column_create('one', 1, 'css', @txt), 65535)) |
+-----------------------------------------------------------------+
|                                                               0 |
+-----------------------------------------------------------------+
1 row in set (0.00 sec)
===========

So the bug is that column_check() doesn't always detects that the dyncol
was truncated.

Either way, even if column_check() returns 1, you probably shouldn't
use truncated dynamic columns in your application :)

Use LONGBLOB instead.

Regards,
Sergei