Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!usc!wuarchive!emory!ogicse!orstcs!cutler.uucp!michael
From: michael@cutler.uucp
Newsgroups: comp.databases
Subject: Re: how to implement cursor based UPDATE/DELETE
Message-ID: <1991May15.191824.40@cutler.uucp>
Date: 15 May 91 19:22:55 GMT
References: <22626@shlump.lkg.dec.com> <12260@mentor.cc.purdue.edu> <24334@oolong.la.locus.com>
Organization: Cutler & Company, Inc., Medford OR (503) 770-9000
Lines: 24

In article <24334@oolong.la.locus.com>, jfr@locus.com (Jon Rosen) writes:

> ... Usually, the ROWID is a logical construction and has nothing (or
> little) to do with the physical location/inode/record number of the
> actual row.  So moving the row will generally not change its ROWID. ...
> 
> Unfortunately, the ROWID solution begs several questions.  First of all,
> a materialized ROWID is by no means universal.  Most RDBMS (including
> DB2, OS/2 EE, RDB, Ingres, etc.) do not have such an abstraction. ...

Just for the record, INGRES does have (and has always had) such an
abstraction.  Each row in an INGRES table or index has a TID, or
tuple identifier.  Use it as you would any other column.

If memory serves, the TID is indeed related to physical location.  The
TID is something to the effect of the high order m bits is the INGRES
page within the table the row is located, and the low order n bits
is the index of the row within the INGRES page.

Michael D. Shields        {UUCP: ...rutgers!ogicse!orstcs!cutler!michael}
AVP, MIS                  {Internet: michael%cutler.uucp@cs.orst.edu}
Cutler & Company, Inc.
Medford, OR  97504
(503) 770-9000