Re: Newbie question on table design.

From: Brian Selzer <brian_at_selzer-software.com>
Date: Sat, 28 Apr 2007 11:55:23 -0400
Message-ID: <M7KYh.1671$tp5.1031_at_newssvr23.news.prodigy.net>


"-CELKO-" <jcelko212_at_earthlink.net> wrote in message news:1177767078.449343.78520_at_n59g2000hsh.googlegroups.com...
>>>
> I can easily think of 2 ways to support record [sic] un-/deletion.
>
> 1. Add an 'is_deleted' column that can take a binary 1/0 value (1 =
> deleted, 0 = not-deleted). .. <<
>
> Rows are not records....

You're wrong, Joe. Record, tuple, row can all mean a fact recorded in a database. Record emphasizes the action required to make a fact part of a database. Tuple emphasizes the formal mathematical structure of data recorded in a relational database. Row emphasizes the informal visual layout of data recorded in an SQL database.

> ...Among the MANY differences is that rows are
> shared and have transaction control problems. Records in a single
> application file system do not. So you probably will want to have
> flags that show (delete_status, commit_status). That would allow the
> engine to support uncommitted reads and the other transaction levels.
> Then there is the need for a housekeeping routine to physically remove
> the (delete, commit) rows, fix the indexes and re-do the stats.
>
> Oh, then that delete_status flag needs to work with updates and
> inserts who also have those transaction levels to consider, so it is
> probably going to be several bits. Database internals are harder than
> file systems.
>
Received on Sat Apr 28 2007 - 17:55:23 CEST

Original text of this message