IDS Forum

Re: Alignment of predefined UDT

Posted By: Wang Shiyi
Date: Monday, 5 March 2012, at 2:19 a.m.

In Response To: Re: Alignment of predefined UDT (Madison Pruet)

Great, I solved the problem of SBLOB's alignment.

I spend whole for testing the alignment, different database collate and
different ids version and different table rowsize.
Finally, I found it. It is a little strange that the alignment is depend on
the table definition not the real data.

Thanks Pruet.

BlackThought

On Mon, Mar 5, 2012 at 7:55 AM, Madison Pruet <mpruet@us.ibm.com> wrote:

> Alignment is required when the data within the opaque type is store 'as
> is'.
>
> As an example, suppose that an opaque type contains an integer data type
> and is stored as is. Then the opaque type would have to be stored on a
> word boundary. So the number of alignment bytes would depend on the byte
> that it was stored. If the definition of the table was...
>
> define table tab (
>
> col1 char(2),
>
> col2 my_UDT_containing_an_integer
>
> )...
>
> Then it would be required to have 2 alignment bytes. However, if col1 was
> char(4), then no alignment bytes would be required.
>
> (Actually the UDT contains an additional flag so the actual alignment bytes
> would be different. But to make my point, we'll ignore that for now.)
>
> So the key thing determining how many alignment bytes is not so much the
> UDT type as where the UDT falls within the row. The definition of the UDT,
> however, does contain an alignment value and that is based on any alignment
> requirements within the UDT itself.
>
> From: "Wang Shiyi" <osdbms@gmail.com>
>
> To: ids@iiug.org
>
> Date: 03/04/2012 07:48 AM
>
> Subject: Alignment of predefined UDT [26437]
>
> Sent by: ids-bounces@iiug.org
>
> According to the incl/public/sqltypes.h
>
> */* Alignment required by predefined UDT's.
> * Only a few are currently defined. At a minimum,
> * all with alignment not 1 should be defined here
> * and used throughout the code.
> */
> #define ALNINDEXKEYARRAY 4
> #define ALNSTAT 8*
>
> and through query table "sysxtdtypes", I can find the definition of xtd
> types.
>
> BUT, I wonder where has the definition of these alignments?
>
> for example: extended_id name
> type align
>
> 1
> lvarchar 40 1
>
> 10
> blob 41 4
>
> 11
> clob 41 4
>
> in my test, the alignment is not static, like the alignment of blob could
> be 1 or 4 in different table in same database.
>
> So my question is "is there a table defined each column of a table's
> alignment length?", or "how alignments used? "
>
> regards.
>
> BlackThought
>
> --e89a8f23458f816f3504ba6b0d27
>
>
>
> *******************************************************************************
>
> Forum Note: Use "Reply" to post a response in the discussion forum.
>
>
>
> *******************************************************************************
> Forum Note: Use "Reply" to post a response in the discussion forum.
>
>

--f46d0401fceddd1eab04ba79bd6c

Messages In This Thread

• Alignment of predefined UDT
  Wang Shiyi -- Sunday, 4 March 2012, at 8:47 a.m.
  • Re: Alignment of predefined UDT
    Madison Pruet -- Sunday, 4 March 2012, at 6:55 p.m.
    • Re: Alignment of predefined UDT
      Wang Shiyi -- Monday, 5 March 2012, at 2:19 a.m.
      • Re: Alignment of predefined UDT
        Art Kagel -- Monday, 5 March 2012, at 5:51 a.m.

IDS Forum is maintained by Administrator with WebBBS 5.12.