maria-developers team mailing list archive

[Alexey Botchkov <holyfoot@xxxxxxxxxxxx>]

18.12.2012 15:30, Sergei Golubchik wrote:
> > And one more thing.
> > It's easy to imagine a user that likes to do the ALTER IF EXISTS
> > queries often.
> >
> > But even if the 'new_field' exists already, ADD FIELD IF EXISTS
> > new_field still does a lot.
> > The list of calls is:
> >        mysql_create_table_no_lock()  - creates the .frm file with the
> >                                        temporary name
> >        mysql_rename_table()
> >        mysql_rename_table()   once more
> >        quick_rm_table()  - removes the old .frm
> >        query_cache_invalidate()
> >        reopen_table()
> Hmm, I don't see it. If the field exists, mysql_alter_table will notice
> it in compare_tables() - that is, before mysql_create_table_no_lock().

It's easy to see, just run the ALTER TABLE ADD FIELD IF NOT EXISTS (int 
existing_field).
If the added field already exists, the ALTER TABLE will recreate the 
table anyway.

The can mysql_compare_tables notice it. But in the IF EXISTS case we 
don't break
the command. So that the alter_table process will be continued.
We can check in the mysql_compare_tables() if the tables are equal and
return the specific code for that. But it's not trivial to implement.
It wasn't a problem until this MDEV as the ALTER list could never be 
empty. But
now i think it can be useful to perform the NOOP ALTER possible faster.

Otherwise i'm working now to reform the mysql_alter_table() so we can do 
the IF EXISTS
checks in the same places where we return the errors about (not yet) 
existing objects.
It seems working, but the modification is going to be significant. Are 
we ok with big changes
in this part of code?

Regards.
HF