Details
-
Bug
-
Status: Closed (View Workflow)
-
Critical
-
Resolution: Fixed
-
10.4.13, 10.4(EOL), 10.5
-
Debian 10
Description
Changing the name of a primary key column with a foreign key constraint fails with the following error since 10.4.13:
[HY000][1553] (conn=17) Cannot drop index 'PRIMARY': needed in a foreign key constraint
create schema test collate utf8_unicode_ci;
|
create table test.input_type
|
(
|
input_type_id int(11) unsigned auto_increment
|
primary key
|
);
|
|
|
create table test.input
|
(
|
input_id int unsigned auto_increment
|
primary key,
|
input_type_id int unsigned not null,
|
constraint ibfk
|
foreign key (input_type_id) references test.input_type (input_type_id)
|
);
|
|
|
create index type_id
|
on test.input (input_type_id);
|
|
|
alter table test.input_type change input_type_id id int(11) unsigned auto_increment;
|
Attachments
Issue Links
- is duplicated by
-
-
- Closed
-
Activity
By running git-bisect, I founded that the bug is introduced by df07e00a810890f6f6eb1334c76ee22133750777.
The following diff is probably the point. old_part.length and new_part.length are equal but old_cs->mbmaxlen and new_cs->mbmaxlen are not in the Alice's test case.
- if (old_part.length != new_part.length) |
+ if (old_part.length / old_cs->mbmaxlen != new_part.length / new_cs->mbmaxlen) |
(gdb) p old_part.length
|
$1 = 4
|
(gdb) p new_part.length
|
$2 = 4
|
(gdb) p old_cs->mbmaxlen
|
$3 = 1
|
(gdb) p new_cs->mbmaxlen
|
$4 = 3
|
nayuta-yanagisawa, thank you very much. It looks like the CHARACTER SET is wrongly affecting the check of the INT column.
bar, can you suggest how we could refine that code to determine the proper length of the column? The intention of the change was to allow an instant conversion from utf8mb3 to utf8mb4. Character sets are not applicable to columns that do not store character data, yet the information appears to be stored for them. How can we detect that old_part and new_part do not store character data, and thus we should ignore any change of character set?
sanja, can you please review?
The patch for 10.5
- Branch:
https://github.com/MariaDB/server/tree/bb-10.5-bar-MDEV-22775 - A direct link to the patch:
https://github.com/MariaDB/server/commit/ff1ee3fdd92ed5d91fa1641d35c2774fdd3fca13
The patch for 10.4:
- Branch:
https://github.com/MariaDB/server/tree/bb-10.4-bar-MDEV-22775 - A direct link to the patch:
https://github.com/MariaDB/server/commit/bbd216ec520c0c6ee160602062c6589fddd1d7c6
Are there, by the way, any known bug reports over at MySQL for the same problems? I tested and had the same issue in MySQL 8, but not in 5.7, so I suspect they have the same issue as well