Re: Indexed PF vs. LF performance

[Booth Martin <booth@xxxxxxxxxxxx>]

My understanding of the reason so many of us are accustomed to not 
keying the PF is that early on keys could not be modified.  So, if the 
key was on last name and she got married, her record could not be 
corrected.  It was copy & delete, which was a major pain if the name 
file had many child files.  That was changed decades ago, but it is a 
tough habit to break.

On 7/31/2012 7:58 AM, rob@xxxxxxxxx wrote:
> Whenever I can I key my physical files and make them unique.
>
> I have some table files that I
> RGZPFM FILE(MYFILE) KEYFILE(*FILE)
> whenever I add records.  Doing it on most files make sense.  UNLESS you
> primarily access by a different order.  An example might be a name file.
> If a vast majority of accesses are by name and rarely by the 'key' it
> might make sense to reorg that by logical built over the name.  Even this
> could be improved upon.  Most PF's that are keyed omit the active record
> code from the key.  Those it might be better to reorg by a key that is
> built over active record id and then the key.
>
> Our ERP software never keys the PF.  They never specify unique in their
> LF's either.  They try to get uniqueness by imbedding the logic into
> programs.  When we do merges, acquisitions, migrations, etc., we often
> have to scrub the data to get uniqueness.  Imagine an item number in the
> file twice.  Once with an active record flag, again with an inactive
> record flag.
>
> I prefer to protect database integrity by using primary unique keys, by
> using foreign keys, by using check constraints, by journalling, etc. Stuff
> that most of DDS rarely even pays lip service to.
>
> Rob Berendt