[Tarantool-patches] [PATCH] lua/key_def: fix compare_with_key() part count check

[Alexander Turenko]

On Thu, Dec 17, 2020 at 04:27:09PM +0300, Sergey Nikiforov wrote:
> > > diff --git a/src/box/lua/key_def.c b/src/box/lua/key_def.c
> > > index a781aeff9..674891a85 100644
> > > --- a/src/box/lua/key_def.c
> > > +++ b/src/box/lua/key_def.c
> > > @@ -362,6 +362,15 @@ lbox_key_def_compare_with_key(struct lua_State *L)
> > >   	size_t key_len;
> > >   	const char *key_end, *key = lbox_encode_tuple_on_gc(L, 3, &key_len);
> > >   	uint32_t part_count = mp_decode_array(&key);
> > > +
> > > +	if (part_count > key_def->part_count) {
> > > +		region_truncate(region, region_svp);
> > > +		tuple_unref(tuple);
> > > +		diag_set(ClientError, ER_KEY_PART_COUNT,
> > > +			 key_def->part_count, part_count);
> > > +		return luaT_error(L);
> > > +	}
> > 
> > Why this check and the call below can't be all simply
> > replaces with box_key_def_validate_key() call?
> 
> Because we need part_count later. With box_key_def_validate_key() we would
> have to call mp_decode_array() twice or add yet another parameter to
> box_key_def_validate_key(). Is that good idea?

We can't change box_key_def_validate_key() parameters, it is in the
public C API.

The code would look more accurate if we'll reuse the public functions
(box_key_def_validate_key() and box_tuple_compare_with_key()) here.
However, right, it'll decode the msgpack array size twice.

I think it is negligible comparing to the validation of the key against
given key_def. If'll need maximum performance from the module, we'll add
an option to skip validation at all.

> 
> > > +
> > >   	if (key_validate_parts(key_def, key, part_count, true,
> > >   			       &key_end) != 0) {
> > >   		region_truncate(region, region_svp);