Bug #8925 Error reported from handler leads to assertion

• Submitted: 3 Mar 2005 15:48
• Modified: 16 Dec 2005 19:13
• Reporter: Mikael Ronström
• Status: Closed
• Category: MySQL Server
• Severity: S3 (Non-critical)
• Version: 5.0.3
• OS: MacOS (Mac OS X)
• Assigned to: Ingo Strüwing
• CPU Architecture: Any

[3 Mar 2005 15:48] Mikael Ronström

Description:
If the handler reports an error on index_read_idx
when doing the query
select * from t1 where a = 1;
where a is the primary key then the server core's on an assert of that
table->key_read != 0 in close_thread_table

The problem is that the cleanup doesn't work properly in this error case

How to repeat:
Mix with a handler such that it reports error on any index_read_idx
create table t1 (a int not null, primary key(a));
insert into t1 values (1);
select * from t1 where a = 1;
Crash occurs on select statement

[3 Mar 2005 16:04] Mikael Ronström

The following code is the problem:
The original code didn't handle the call to extra(HA_EXTRA_NO_KEYREAD) in the
error situation so this call was done before the error was checked.

(The code is in join_read_const_table in sql_select.cc)

Patched code:
    error=join_read_const(tab);
    if (table->key_read)
    {
      table->key_read=0;
      table->file->extra(HA_EXTRA_NO_KEYREAD);
    }
    if (error)
    {
     tab->info="unique row not found";
      /* Mark for EXPLAIN that the row was not found */
      pos->records_read=0.0;
      if (!table->maybe_null || error > 0)
	DBUG_RETURN(error);
    }

[22 Jul 2005 19:18] Mikael Ronström

This bug has been fixed in the 5.1 tree

[19 Oct 2005 11:39] Mikael Ronström

Bug not fixed in 5.0 release but fixed in 5.1 release

[20 Oct 2005 20:34] Paul DuBois

Please indicate the three-part version number when the fix
was made.  Thanks.

[16 Dec 2005 19:13] Jim Winstead

If it was fixed in 5.1 tree in October, it was before any 5.1 releases and thus doesn't need documentation.