[Tarantool-patches] [PATCH v1 1/1] sql: fix INSTEAD OF DELETE trigger for VIEW
Nikita Pettik
korablev at tarantool.org
Thu Jan 30 01:48:21 MSK 2020
On 22 Jan 16:54, imeevma at tarantool.org wrote:
> diff --git a/src/box/sql/delete.c b/src/box/sql/delete.c
> index 8d9ae4f..e9edbcc 100644
> --- a/src/box/sql/delete.c
> +++ b/src/box/sql/delete.c
> @@ -231,7 +231,7 @@ sql_table_delete_from(struct Parse *parse, struct SrcList *tab_list,
> int eph_cursor = parse->nTab++;
> int addr_eph_open = sqlVdbeCurrentAddr(v);
> if (is_view) {
> - pk_len = space->def->field_count;
> + pk_len = space->def->field_count + 1;
> parse->nMem += pk_len;
> sqlVdbeAddOp2(v, OP_OpenTEphemeral, reg_eph,
Please, add comments explaining what's going on here. Despite the seeming
simplicity, this change is way far from being obvious. This time I've added
comment myself:
+ /*
+ * At this stage SELECT is already materialized
+ * into ephemeral table, which has one additional
+ * tail field (except for ones specified in view
+ * format) which contains sequential ids. These ids
+ * are required since selected values may turn out to
+ * be non-unique. For instance:
+ * CREATE TABLE t (id INT PRIMARY KEY, a INT);
+ * INSERT INTO t VALUES (1, 1), (2, 1), (3, 1);
+ * CREATE VIEW v AS SELECT a FROM t;
+ * Meanwhile ephemeral tables feature PK covering ALL
+ * fields of format, so without id materialization
+ * processing is impossible. Then, results of
+ * materialization are transferred to the table being
+ * created below. So to fit tuples in it we must
+ * account that id field as well. That's why pk_len
+ * has one field more than view format.
+ */
Next time I'll reject such patches.
Pushed to master, 2.3 and 2.2.
More information about the Tarantool-patches
mailing list