Re: ctlib compute columns (was: "Fix affected rows after insert with prepared query" side effect)

It's always a good idea to open a new thread for a new topic!

2016-05-11 21:47 GMT+01:00 John Kendall <[email protected]>:
>
> On May 11, 2016, at 1:23 PM, Frediano Ziglio <[email protected]> wrote:
>
>> 2016-05-11 6:07 GMT+01:00 John Kendall <[email protected]>:
>>>
>>>> On May 10, 2016, at 3:34 PM, Frediano Ziglio <[email protected]> wrote:
>>>>
>>>> 2016-05-10 22:31 GMT+01:00 John Kendall <[email protected]>:
>>>>> It appears the patch for this ("fix ct-library rows count") has broken compute in ctlib:
>>>>>
>>>>>
>>>>> select col=1 into #tbl insert #tbl values (2) select * from #tbl compute sum(col)
>>>>> (1 row affected)
>>>>> (1 row affected)
>>>>>  col
>>>>>  -----------
>>>>>            1
>>>>>            2
>>>>>            2
>>>>>
>>>>> (3 rows affected)
>>>>>
>>>>> dblib (and ctlib RC1) does the expected:
>>>>>
>>>>> select col=1 into #tbl insert #tbl values (2) select * from #tbl compute sum(col)
>>>>> (1 row affected)
>>>>> (1 row affected)
>>>>>  col
>>>>>  -----------
>>>>>            1
>>>>>            2
>>>>>  sum
>>>>>  ===========
>>>>>            3
>>>>>
>>>>> (3 rows affected)
>>>>>
>>>>>
>>>>> Sorry to keep finding things!
>>>>>
>>>>> John
>>>>>
>>>>>
>>>>
>>>> Can you put some more detailed examples? Which program did you use?
>>>>
>>>> Frediano
>>>
>>> I'm using sqsh.
>>>
>>> The compute clause should return an extra result line.  In the example I have given above, the table is a single column table with two rows with values 1 and 2.  Adding the compute sum clause to the select should return a compute result of 3, instead it is returns an extra copy of the last row.  My example is not a special case, no compute (sum, max, min, avg, etc) results seem to be returning from any queries with a compute clause.  I never saw this problem before RC4. Backing out the 'fix ct-library rows count' patch on ct.c makes it return the expected result.
>>>
>>> John
>>>
>>
>>
>> Just to be sure, can you test 1.0rc5 ?
>>
>
> I already have.   And it looks good.  Thanks.
> While I have your attention, and you are in that part of the code, I wonder if you might look into this:
>
> http://lists.ibiblio.org/pipermail/freetds/2016q1/029725.html
>
> I want to clarify what I'm asking you to look into.  That was a multi-topic thread.
> The problem I am still interested in fixing is the ct_compute_info return values.
> (The rest of that thread turned out to just be client differences.)
>

Patch looks good. Take a look at "compute operator" in include/freetds/tds.h.

Looks like for dblib constants are defined as protocol so no
transformation is required however ctlib (and our headers!) contain
different constants. Your patch is good however:
- I would add a small function to do the conversion;
- I would add support for additional mssql operators (I would collapse
CNT_BIG into count and *U ones in no -U constants).

Frediano