On 2011-04-06 10:27, Sergey Glukhov wrote:
> Better place for the fix is matching_cond() function,
> the reason is that the function can not return TRUE
> if item evaluation is failed. So suggested fix
> could look like:
>
> == modified file 'sql/opt_sum.cc'
> --- old/sql/opt_sum.cc 2010-06-11 07:38:29 +0000
> +++ new/sql/opt_sum.cc 2011-04-05 10:39:44 +0000
> @@ -732,6 +732,9 @@
>
> if (is_null || (is_null_safe_eq && args[1]->is_null()))
> {
> + if (current_thd->is_error())
> + DBUG_RETURN(FALSE);
> +
> part->field->set_null();
> *key_ptr= (uchar) 1;
> }
>
>
> Regards,
> Gluh
Hi Gluh
I like the idea of catching the error as early as possible, so I'll
implement that.
However: opt_sum_query() should also catch the error, and return
HA_ERR_XXX to JOIN::optimize:
- that's the documented behaviour for opt_sum_query
- if we don't return an error code, then the optimizer will continue
(or *it* will have to check for thd->is_error)
I'm not so happy about changing the interface of matching_cond() and
find_key_for_maxmin()
from
@retval false Index can't be used.
to
@retval false Index can't be used, or error has occured.
but I guess we cannot avoid it.
What do you think?
-- didrik
