[tarantool-patches] Re: [PATCH v2 4/4] sql: remove OP_LoadPtr

  • From: Vladislav Shpilevoy <v.shpilevoy@xxxxxxxxxxxxx>
  • To: "n.pettik" <korablev@xxxxxxxxxxxxx>, tarantool-patches@xxxxxxxxxxxxx
  • Date: Tue, 10 Jul 2018 23:23:51 +0300



On 10/07/2018 21:34, n.pettik wrote:


@@ -2969,12 +2964,9 @@ sql_create_index(struct Parse *parse, struct Token 
*token,
                        goto exit_create_index;

                sql_set_multi_write(parse, true);
-
-
-               sqlite3VdbeAddOp2(v, OP_SIDtoPtr, BOX_INDEX_ID,
-                                 index_space_ptr_reg);
-               sqlite3VdbeAddOp4Int(v, OP_OpenWrite, iCursor, 0,
-                                    index_space_ptr_reg, 6);
+               sqlite3VdbeAddOp4(v, OP_OpenWrite, iCursor, 0, 0,
+                                 (void *)space_by_id(BOX_INDEX_ID),
+                                 P4_SPACEPTR);
                sqlite3VdbeChangeP5(v, OPFLAG_SEEKEQ);

Wait, AFAIK SIDtoPtr was used deliberately taking into consideration
the fact that DDL may change ptr to space. So conversion id -> space ptr
is delayed until vdbe execution (right before cursor opening). Am I wrong?

Yes, you are wrong here, because it was used to restore pointer to
struct space "_index", but "_index" is a system space, and it is never
altered. So pointer to "_index" is always valid, and it makes no sense
to restore it on runtime.



Other related posts:

  1. Home
  2. tarantool-patches
  3. Archive
  4. 07-2018