RE: SV: Embedded SQL

[Joel Cochran <jrc@xxxxxxxxxx>]

On Thu, 2005-05-12 at 16:51, Wilt, Charles wrote:
> Good point Joel, I for one wasn't aware of it.
> 
> However, it could still be a problem since depending on such behavior assumes 
> that columns will be added to the end of the file, unused columns will never 
> be removed and in general the order will never change.

> If you really need every column, maybe using select * could be excused.  But 
> even then, it's simply the lazy way out.  It doesn't really take that much 
> effort to get a list of the fields a file.  What to me is inexcusable, is the 
> use of select * when you don't actually need all the fields.

Charles,

These are completely valid points, and I agree that when feasible you
should specify selected columns. I was just pointing out that simply
adding a field to a file does not *require* a program to be recompiled:
someone might get the wrong idea that this created some kind of level
check problem.  

As for using *, it can be seen as lazy but it fits better with what
people are used to using Native I/O: they read a record and all the
fields are there.  Using select * in conjunction with an externally
defined DS based on the same file is a simple way to get native I/O
programmers comfortable with using embedded select.

Also, if the file is very large and you need *most* of the fields, then
I think using * is perfectly valid.  I'll give you an example: the
master file for one of our old Sys-36 apps has 237 fields in it.  We
have a print program that uses about 175 of those fields: if I wanted to
use SQL and NOT use *, I would have to list all 175 of them.  That just
doesn't seem very programmer or maintenance friendly to me.  Of course,
many would argue that for such an application I should use native I/O
anyway, but that is neither here nor there...

And just for the archives, or if anyone is curious, even when I select
individual field names, I still use an externally described DS based on
the same file.  I do this for Select, Insert, and Update statements:
first of all, it guarantees that the field definitions are correct which
helps prevent errors; second, since the field definitions are the same,
there are no hidden conversions that have to take place, so performance
is better; finally, since no conversions have to take place, you can
(typically) rely on '00000' and '02000' SQLSTT values, which means less
error checking.  Just some random thoughts...

Joel Cochran
http://www.rpgnext.com