Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
5.5(EOL), 10.0(EOL), 10.1(EOL)
Description
Because MariaDB does not accept a constant virtual column (other DBMS's do) I was obliged to write the noconst UDF as:
/***********************************************************************/
|
/* Returns its argument saying it is not a constant. */
|
/***********************************************************************/
|
my_bool noconst_init(UDF_INIT *initid, UDF_ARGS *args, char *message)
|
{
|
if (args->arg_count != 1 || args->arg_type[0] != STRING_RESULT) {
|
strcpy(message, "noconst unique argument must be a string");
|
return true;
|
} // endif arg
|
|
|
initid->const_item = false; // The trick!
|
return false;
|
} // end of noconst_init
|
|
|
char *noconst(UDF_INIT *initid, UDF_ARGS *args, char *result,
|
unsigned long *res_length, char *, char *)
|
{
|
return args->args[0];
|
} // end of noconst
|
Now, creating the table t1, inserting one line to it and displaying it does work:
MariaDB [test]> create table t1 (
|
-> n int key not null auto_increment,
|
-> msg char(20) as (noconst('Hello World')) virtual);
|
Query OK, 0 rows affected (0.06 sec)
|
|
|
MariaDB [test]> insert into t1() values();
|
Query OK, 1 row affected (0.01 sec)
|
|
|
MariaDB [test]> select * from t1;
|
+---+-------------+
|
| n | msg |
|
+---+-------------+
|
| 1 | Hello World |
|
+---+-------------+
|
1 row in set (0.00 sec)
|
However, later on when stopping the server a crash occurs in the function:
void udf_handler::cleanup()
|
{
|
if (!not_original)
|
{
|
if (initialized)
|
{
|
if (u_d->func_deinit != NULL)
|
{
|
Udf_func_deinit deinit= u_d->func_deinit;
|
(*deinit)(&initid);
|
}
|
free_udf(u_d);
|
initialized= FALSE;
|
}
|
if (buffers) // Because of bug in ecc
|
delete [] buffers;
|
buffers= 0;
|
}
|
}
|
See the attach udf_crash.log.
The same kind of crash occurs with different other UDF's.
Attachments
Activity
|
Stack trace from 5.5 commit 29ac245dd04c5416d57a90b0ab3c4d08cbc4d723 |
#3 <signal handler called>
|
#4 0x00000000008500ae in udf_handler::cleanup (this=0x7f11c2875968) at 5.5/sql/item_func.cc:3454
|
#5 0x00000000008514e0 in Item_udf_func::cleanup (this=0x7f11c28758a0) at 5.5/sql/item_func.cc:3766
|
#6 0x00000000006073a1 in Item::delete_self (this=0x7f11c28758a0) at 5.5/sql/item.h:1434
|
#7 0x000000000060162f in Query_arena::free_items (this=0x7f11c2874bf8) at 5.5/sql/sql_class.cc:3164
|
#8 0x00000000006fcf32 in closefrm (table=0x7f11c2828e60, free_share=true) at 5.5/sql/table.cc:2773
|
#9 0x00000000005d4378 in intern_close_table (table=0x7f11c2828e60) at 5.5/sql/sql_base.cc:929
|
#10 0x00000000005d43e5 in free_cache_entry (table=0x7f11c2828e60) at 5.5/sql/sql_base.cc:952
|
#11 0x00000000005d4763 in close_cached_tables (thd=0x0, tables=0x0, wait_for_refresh=false, timeout=31536000) at 5.5/sql/sql_base.cc:1059
|
#12 0x00000000005d3111 in table_def_start_shutdown () at 5.5/sql/sql_base.cc:410
|
#13 0x0000000000586faf in clean_up (print_message=true) at 5.5/sql/mysqld.cc:1822
|
#14 0x0000000000586d9e in unireg_end () at 5.5/sql/mysqld.cc:1735
|
#15 0x0000000000586cc0 in kill_server (sig_ptr=0x0) at 5.5/sql/mysqld.cc:1663
|
#16 0x0000000000586cdd in kill_server_thread (arg=0x7f11ed3b9e98) at 5.5/sql/mysqld.cc:1686
|
#17 0x00007f11ecfaab50 in start_thread (arg=<optimized out>) at pthread_create.c:304
|
#18 0x00007f11eb26095d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:112
|
I cannot repeat this. elenst, could you please show a complete test case?
I do the following:
- build the attached noconst.c as g++ -shared -o noconst.so noconst.c -I/home/elenst/git/5.5/sql -I/home/elenst/git/5.5/include
- start the debug server with noconst.so in the plugin dir;
- run
CREATE FUNCTION noconst RETURNS STRING SONAME 'noconst.so';create table t1 (n int key not null auto_increment, msg char(11) as (noconst('Hello World')) virtual);insert into t1() values();select * from t1; - shutdown server;
=> observe the crash.
Reproducible on 5.5 commit 13ad179c96ee8c8c4043806b8575c851e3676f0d
-
Does this mean that UDFs are no more banned from virtual columns?
If so, it should be documented.
They were not banned, were they? If they were banned, one would not be able to create a virtual column with a UDF in the first place.
serg, it was the point of my initial comment: the documentation says they cannot be used, while in fact they can.
I don't know whether they had ever been (or supposed to be) forbidden, or it is just a documentation mistake.
That's a surprise it works at all, documentation says UDFs are not allowed for virtual columns.
bertrandop,
Could you please attach the whole c file with your UDF function? It will be faster than me creating missing parts.