[Tarantool-patches] [PATCH v2 1/1] sql: do not change order of inserted values

Vladislav Shpilevoy v.shpilevoy at tarantool.org
Sun Apr 12 19:00:45 MSK 2020


On 12/04/2020 17:21, Imeev Mergen wrote:
> 
> On 4/12/20 5:02 PM, Vladislav Shpilevoy wrote:
>>>>> --- a/src/box/sql/vdbe.c
>>>>> +++ b/src/box/sql/vdbe.c
>>>>> @@ -2694,23 +2694,10 @@ case OP_Column: {
>>>>>            pC->cacheStatus = p->cacheCtr;
>>>>>        }
>>>>>        enum field_type field_type = field_type_MAX;
>>>>> -    if (pC->eCurType == CURTYPE_TARANTOOL) {
>>>>> -        /*
>>>>> -         * Ephemeral spaces feature only one index
>>>>> -         * covering all fields, but ephemeral spaces
>>>>> -         * lack format. So, we can fetch type from
>>>>> -         * key parts.
>>>>> -         */
>>>>> -        if (pC->uc.pCursor->curFlags & BTCF_TEphemCursor) {
>>>>> -            field_type = pC->uc.pCursor->index->def->
>>>>> -                    key_def->parts[p2].type;
>>>>> -        } else if (pC->uc.pCursor->curFlags & BTCF_TaCursor) {
>>>>> -            field_type = pC->uc.pCursor->space->def->
>>>>> -                    fields[p2].type;
>>>>> -        }
>>>>> -    } else if (pC->eCurType == CURTYPE_SORTER) {
>>>>> +    if (pC->eCurType == CURTYPE_TARANTOOL)
>>>>> +        field_type = pC->uc.pCursor->space->def->fields[p2].type;
>>>>> +    else if (pC->eCurType == CURTYPE_SORTER)
>>>>>            field_type = vdbe_sorter_get_field_type(pC->uc.pSorter, p2);
>>>>> -    }
>>>>>        struct Mem *default_val_mem =
>>>>>            pOp->p4type == P4_MEM ? pOp->p4.pMem : NULL;
>>>>>        if (vdbe_field_ref_fetch(&pC->field_ref, p2, pDest) != 0)
>>>> Shouldn't this be a part of the previous commit, which
>>>> introduces the ephemeral space format?
>>> They should, along with 3962. But, they have different milistones,
>>> I wasn't sure that it is right to put all three tickets in one
>>> patch-set.
>> But they already depend on each other. So what is a problem?
> True. Should I resend the patch-set?

I guess Nikita anyway should take a look. So it would be good.
Better send them all as one patchset, I think. Since they can't
be applied separately.


More information about the Tarantool-patches mailing list