maria-developers team mailing list archive

[Oleksandr Byelkin <sanja@xxxxxxxxxxxxxxxx>]

On 15.12.14 22:03, Sergei Golubchik wrote:
> Hi, Sanja!
>
> On Nov 17, sanja@xxxxxxxxxxxx wrote:
> > ------------------------------------------------------------
> > revno: 4350
> > revision-id: sanja@xxxxxxxxxxxx-20141117132635-gf281ipiomifw1oc
> > parent: psergey@xxxxxxxxxxxx-20141113111559-cuhsyfhor141qo6a
> > committer: sanja@xxxxxxxxxxxx
> > branch nick: work-maria-5.5-MDEV-6985
> > timestamp: Mon 2014-11-17 14:26:35 +0100
> > message:
> >    MDEV-6985: MariaDB crashes on stored procedure call
> >    
> >    Item_ident fixed to allow double cleanup().
> > === modified file 'sql/item.cc'
> > --- a/sql/item.cc	2014-10-06 17:53:55 +0000
> > +++ b/sql/item.cc	2014-11-17 13:26:35 +0000
> > @@ -862,12 +862,20 @@ void Item_ident::cleanup()
> >   		       field_name ? field_name : "(null)",
> >                          orig_field_name ? orig_field_name : "(null)"));
> >   #endif
> Before pushing, please remove the above #ifdef, in a separate commit.
> thanks!
>
> > +  bool was_fixed= fixed;
> >     Item::cleanup();
> >     db_name= orig_db_name;
> >     table_name= orig_table_name;
> >     field_name= orig_field_name;
> >     /* Store if this Item was depended */
> > -  can_be_depended= test(depended_from);
> > +  if (was_fixed)
> > +  {
> > +    /*
> > +      We can trust that depended_from set correctly only if this item
> > +      was fixed
> > +    */
> > +    can_be_depended= test(depended_from);
> > +  }
> Are double cleanups generally normal? Why this item is cleaned up twice?
Of course it is not so efficient, but we have it (SP works so that it 
can do cleanup for statement as it executed, than as part of procedure, 
if you need exact path I can find it again).

Also I can imagine how it can happens in "usual" execution for example 
we have PS and try to execute it with lack of a table (which cause an 
error) and so items will be cleanup second time without being used then 
return table in place and execute PS again and hit this bug.
> >     DBUG_VOID_RETURN;
> >   }
>
> Regards,
> Sergei