Bug #94560 record comparison in spatial index non-leaf rtree node seems incorrect
Description:
According to the function cmp_rec_rec_with_match:
---------
for (; cur_field < rec1_n_fields && cur_field < rec2_n_fields;
cur_field++) {
ulint mtype;
ulint prtype;
/* If this is node-ptr records then avoid comparing node-ptr
field. Only key field needs to be compared. */
if (cur_field == dict_index_get_n_unique_in_tree(index)) {
break;
}
...
const dict_col_t* col;
col = dict_index_get_nth_col(index, cur_field);
mtype = col->mtype;
prtype = col->prtype;
/* If the index is spatial index, we mark the
prtype of the first field as MBR field. */
if (cur_field == 0 && dict_index_is_spatial(index)) {
ut_ad(DATA_GEOMETRY_MTYPE(mtype));
prtype |= DATA_GIS_MBR;
}
}
...
ret = cmp_data(mtype, prtype,
rec1_b_ptr, rec1_f_len,
rec2_b_ptr, rec2_f_len);
if (ret) {
goto order_resolved;
}
}
---------
If the record is located in a non-leaf node of spatial index, it consists of two fields, mbr and node-ptr.
"if (cur_field == 0 && dict_index_is_spatial(index)) prtype |= DATA_GIS_MBR;"
It only deals with the first col.
For the second field, mtype will be set to col->mtype.
It equals to index->fields[1]->col->mtype, which is the primary key's mtype.
So that we compare node-ptr field by pk's mtype.
It is ok in most cases since pk is always set to a int type in real world.
But if pk is varchar type or others, some unexpected problem may happen.
How to repeat:
Currently, I just read the code direcly and find the problem.
Suggested fix:
if (dict_index_is_spatial(index)
&& (!page_is_leaf(page_align(rec1)) || !page_is_leaf(page_align(rec2)))) {
cmp_spatial_index_nonleaf = true;
}
if (cur_field == 1 && cmp_spatial_index_nonleaf) {
/* mark as node-ptr */
mtype = DATA_SYS_CHILD;
prtype = DATA_NOT_NULL;
}
Portions of this website are copyright © 2001, 2002 The PHP Group
Page generated in 0.040 sec. using MySQL 8.4.7-u6-cloud
Timestamp references displayed by the system are UTC.
Content reproduced on this site is the property of the
respective copyright holders.
It is not reviewed in advance
by Oracle and does not necessarily represent the opinion of
Oracle or any other party.
Description: According to the function cmp_rec_rec_with_match: --------- for (; cur_field < rec1_n_fields && cur_field < rec2_n_fields; cur_field++) { ulint mtype; ulint prtype; /* If this is node-ptr records then avoid comparing node-ptr field. Only key field needs to be compared. */ if (cur_field == dict_index_get_n_unique_in_tree(index)) { break; } ... const dict_col_t* col; col = dict_index_get_nth_col(index, cur_field); mtype = col->mtype; prtype = col->prtype; /* If the index is spatial index, we mark the prtype of the first field as MBR field. */ if (cur_field == 0 && dict_index_is_spatial(index)) { ut_ad(DATA_GEOMETRY_MTYPE(mtype)); prtype |= DATA_GIS_MBR; } } ... ret = cmp_data(mtype, prtype, rec1_b_ptr, rec1_f_len, rec2_b_ptr, rec2_f_len); if (ret) { goto order_resolved; } } --------- If the record is located in a non-leaf node of spatial index, it consists of two fields, mbr and node-ptr. "if (cur_field == 0 && dict_index_is_spatial(index)) prtype |= DATA_GIS_MBR;" It only deals with the first col. For the second field, mtype will be set to col->mtype. It equals to index->fields[1]->col->mtype, which is the primary key's mtype. So that we compare node-ptr field by pk's mtype. It is ok in most cases since pk is always set to a int type in real world. But if pk is varchar type or others, some unexpected problem may happen. How to repeat: Currently, I just read the code direcly and find the problem. Suggested fix: if (dict_index_is_spatial(index) && (!page_is_leaf(page_align(rec1)) || !page_is_leaf(page_align(rec2)))) { cmp_spatial_index_nonleaf = true; } if (cur_field == 1 && cmp_spatial_index_nonleaf) { /* mark as node-ptr */ mtype = DATA_SYS_CHILD; prtype = DATA_NOT_NULL; }