Re: [Tarantool-patches] [PATCH] sql: support constraint drop

[Nikita Pettik <korablev@tarantool.org>]

On 21 Feb 00:09, Vladislav Shpilevoy wrote:
> Hi! Thanks for the patch!
> 
> > -	vdbe_emit_fk_constraint_drop(parse_context, constraint_name,
> > -				     child->def);
> > -	/*
> > -	 * We account changes to row count only if drop of
> > -	 * foreign keys take place in a separate
> > -	 * ALTER TABLE DROP CONSTRAINT statement, since whole
> > -	 * DROP TABLE always returns 1 (one) as a row count.
> > -	 */
> >  	struct Vdbe *v = sqlGetVdbe(parse_context);
> > +	assert(v != NULL);
> > +	assert(id->type < constraint_type_MAX);
> > +	switch (id->type) {
> > +	case CONSTRAINT_TYPE_PK:
> > +	case CONSTRAINT_TYPE_UNIQUE: {
> > +		uint32_t index_id = box_index_id_by_name(space->def->id, name,
> > +							 strlen(name));
> 
> 3. This is definitely not ok. It is not just looking at space *, it is a
> select during parsing. _index has a unique index by space id and name, so
> you can emit a deletion opcode without learning an index id. It shouldn't
> be hard.

Fair note. Then additional request from me: let's remove
box_index_id_by_name() from sql_drop_index() and pragma_index_info()
as well (in a separate patch ofc).