Date:         Sun, 2 Aug 1998 14:55:18 +0200
Reply-To:     martin.gregory@bigfoot.com
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Martin Gregory <martin@DJOUCE.MANNHEIM-NETZ.DE>
Subject:      Re: Bug or Feature With No Obvious Use?
Comments: To: kjb@GA.UNC.EDU
In-Reply-To:  message from Keith Brown on Fri, 31 Jul 1998 14:06:15 -0500
Content-Type: text/plain; charset=US-ASCII

Hi Keith,
 
    Keith>    I recently ran across (got bitten by, actually) the
    Keith> situation illustrated below.  I had the same variable (N1)
    Keith> in two different datasets.  In one, it was character; in
    Keith> the other, it was numeric.  I renamed the character version
    Keith> to be consistent with a character variable in the other
    Keith> dataset, and recomputed the numeric variable when it was
    Keith> missing (or so I thought).  Somehow, the numeric variable
    Keith> was retained from the first observation in dataset B.  Does
    Keith> anybody know where this "feature" is documented, if at all?
    Keith> I'd like to think that, if it's part of SAS, there's a
    Keith> reason for it, but danged if I can imagine a situation
    Keith> where this is what I want to happen.  Thanks for any
    Keith> pointers,
 
The answer lies in understanding when the dataset data vector is
initialised.The initialisation of the data vector is done when an
observation is read from the dataset, but my understanding is that
only the variables from the dataset being read are initialised. It
also seems that in a simple concatenating SET, the variables from the
first dataset are initialised during the execution of the dataset
subsequent to the last observation being read from that dataset.
 
Your variable N1 is only on dataset A, the first one to be read. You
explicitly set N1 after the first obs has been read from B, and this
never gets initialised again because no more obs are read from A. Thus
your condition "n1=." is only true on the first obs from B.
 
You can see what happens by putting the values before and after the
set statement (I've dropped some of the lines for clarity):
 
   49   data c;
   50   put '++> ' _all_ ;
   51   set  a
   52        b(rename=(n1=c1));
   53   put '--> ' _all_ ;
   54   if n1=. then n2=input(c1,4.2);
   55   if n1=. then n1=input(c1,4.2);
   56   run;
 
   ++> ID=  C1=  N1=. N2=. _ERROR_=0 _N_=1
   --> ID=A1 C1=0100 N1=1 N2=. _ERROR_=0 _N_=1
   ++> ID=A1 C1=0100 N1=1 N2=. _ERROR_=0 _N_=2
   --> ID=A2 C1=0200 N1=2 N2=. _ERROR_=0 _N_=2
   [...]
   --> ID=A9 C1=0900 N1=9 N2=. _ERROR_=0 _N_=9
   ++> ID=A9 C1=0900 N1=9 N2=. _ERROR_=0 _N_=10
   --> ID=B1 C1=0100 N1=. N2=. _ERROR_=0 _N_=10
   ++> ID=B1 C1=0100 N1=1 N2=. _ERROR_=0 _N_=11
   --> ID=B2 C1=0200 N1=1 N2=. _ERROR_=0 _N_=11
   ++> ID=B2 C1=0200 N1=1 N2=. _ERROR_=0 _N_=12
   [...]
   NOTE: The data set WORK.C has 18 observations and 4 variables.
   NOTE: DATA statement used:
         real time           0.17 seconds
         cpu time            0.02 seconds
 
cheers, Martin
--
Martin Gregory                     Steubenstr. 52-54
martin@djouce.mannheim-netz.de     D-69121 Heidelberg