On Tue, 29 Oct 2002 16:37:16 -0500, Ian Whitlock <WHITLOI1@WESTAT.COM> wrote:

>Jack, > >It doesn't look accurate in my test. > >1645 data w ; >1646 do x = 1 to 5 ; output ; end ; >1647 run ; > >NOTE: The data set WORK.W has 5 observations and 1 variables. >NOTE: DATA statement used: > real time 0.11 seconds > > >1648 >1649 proc sql noprint ; >1650 select * from w ; >1651 %put &sqlobs ; >1

<snipped>

Has this odd "gotcha" been discussed in SAS-L before? If so, my apologies and please read no further. If not, then my question is, has anybody else ever found this behavior of sqlobs awfully counter-intuitive? In the on- line documentation, "Using Macro Variables Set by Proc SQL", the example in fact demonstrates the precise behavior Ian points out: if noprint is set, and you just do a select, sqlobs=1, regardless of how many rows would have been returned if noprint were off. Remove the noprint, and you get (the correct) 5. Leave the noprint in and add a "create table as. . ." and you get (also intuitively correct) 5. But just do a select with noprint, you get 1. At the very least, shouldn't it be 0? There are a couple other gotchas with sqlobs that escape me now.

In most of the other SQL implementations of which I'm aware, the corresponding system variable contains the # of rows returned, period. I think SI might want to consider changing the function of this variable, or at least documenting it with more than just a "for example." Any thoughts?

Karl